3C4
Yaap
PREDICTION METHODS AND PAROLE
A Study of Factors Involved in the Violation or Non-violation of Parole in a Group of Minnesota Adult Males
By
GEORGE B. VOLD, Ph.D.
Assistant Professor of Sociology University of Minnesota
THE SOCIOLOGICAL PRESS
Hanover	Minnesota	Liverpool
New Hampshire	Minneapolis	England
19 3 1PREDICTION METHODS AND PAROLE
A Study of Factors Involved in the Violation or Non-violation of Parole in a Group of Minnesota Adult Males
By
GEORGE B. VOLD, Ph.D.
Assistant Professor of Sociology University of Minnesota
THE SOCIOLOGICAL PRESS
Hanover	Minneapolis	Liverpool
New Hampshire	Minnesota	England
19 3 1Copyright, 1931
By
GEORGE B. VOLD
(Printed in the United States of America, by The Patriot Press, InC., Putnam, Connecticut).3 fct
PREFACE
Present parole practice in Minnesota ’grants parole privileges to about one-third of the men incarcerated in the State Prison or in the Reformatory. The question is therefore continually in the foreground: What kind of men should be paroled and what kind should be left in confinement within the institution? The present study represents an attempt to apply prediction methods to this problem of selection.
No attempt has been made to write a history of parole in Minnesota nor has it been the concern of the present writer to evaluate or pass judgment on past or present methods. Parole practice in this state probably measures up quite well with that in other states. In common with the practice in other states, it rests on common sense judgments and conclusions in regard t0 men rather than on any scientific analysis of the situations involved. All that the present study attempts is to show how the information already in the parole records may be utilized for a more systematic and intelligent selection of the men to be placed on parole. All the essential source materials and tables have been included, not because they have any particular value in and of themselves, but in order that other research workers interested in the same problem may have opportunity further to check and test and supplement the present work.
Students in. the field will recognize the writer’s heavy indebtedness to the pioneer studies of Professors Burgess and Glueck for fundamental methodology and approach. This is particularly true of Professor Burgess’ earlier study of parole prediction in Illinois. Appropriate foot notes and citations seek to render this indebtedness more concrete and specific. For a fundamental point of view in Criminology the writer is primarily indebted to Professor E. H. Sutherland, teacher and friend. It was at his suggestion and under his sympathetic and stimulating criticism that the present study was begun. Grateful acknowledgment is also made of the invaluable assistance rendered by Professor M. M. Wiley who has read the entire manuscript with care and offered many worthwhile suggestions. Stimulating criticisms and helpful suggestions were also received from Professors D. G. Paterson, P. A. Sorokin, G. A. Lundberg, and F. Stuart Chapin.
Indebtedness is further acknowledged to the Minnesota State Board of Parole for permission to make use of their records and confidential files; to the Honorable C. J. Swendsin, Chairman of the Board of Parole for personal interest and good will throughout; to J. H. Dewitt, State Parole Agent for valuable suggestions and friendly cooperation; to Miss Ella S. Hudgins, Secretary to the Chairman of the Parole Board for patient helpfulness in locating records in history data; to Miss Pearl Putnam, Secretary Registration Division for assistance and counsel in securing information about the men; to the entire personnel of the Parole Department and the Registration Division for cheerful acceptance of the added inconvenience of another investigator under «extremely crowded office conditions.	,
Finally, the writer desires to acknowledge his profound indebtedness to his wife, Ila R. Void for assistance, not only with routine tabulations, but also in countless other ways.
GEORGE B. VOLD
University of Minnesota January, 1931.TABLE OF CONTENTS
Chapter	Page
I.	Introduction ........................................  1-5
1.	General ..................................... 1
2.	Organization of Parole in Minnesota ........1-3
3.	The Problem v.............................. 3-5
II.	Earlier Studies and	Material	Dealing with
Parole ..........................................  6-19
1.	Discussions of the Desirability or Undesirability of Parole..... .............................6-8
2.	Discussions of the Aims, Ideals, and Technique
of Parole Administration ..................8-10
3.	Traits or Conditions of the Individual and the
Outcome of Penal Treatment................10-12
4.	Specific Quantitative Studies of Parole
Factors ..................................12-19
III.	The Method and	Technique	of	the
Present Study	............................    20-29
1.	The Setting: Population Statistics for the
Institutions .............................20-24
2.	Collecting the Data ......................25-28
3.	The Plan of Analysis .................... 28-29
IV.	Some Results : Distributions and Violation Rates for all of the Major Categories of Information ...............................................  30-57
1. General Results: Distribution of Violators and Non-violators for each of the Institutional Groups ..............................  30-32
2.	Factors Involving the Circumstances and Conditions, of the Trial and Commitment .........32-38
3.	Factors Involving the Circumstances and Conditions of the Inmate’s Social Background 39-43
4. Factors Involving the Traits, Habits, and Characteristics of the Individual ............43-46
5.	Factors Associated, with the Period of Stay
in the Institution .......................46-47
6.	Factors Associated with the Period on
Parole .................................  48-50
7.	Miscellaneous Factors of Interest in Connection with Parole ........................ ....50-54
8.	The Relative Importance of the Various Fac-
tors Involved in Violation or Non-violation of Parole ...................................54-56
9.	Summary .....................................57
viTABLE OF CONTENTS
Chapter	Page
V.	The Question of Reliability or Consistency ... 58-82
1.	The Problem of Reliability in' Reclassification 58
2.	Procedure for Checking' Reliability in This
Study ...................................   58-60
3.	The Question of Statistical Methods and
Procedures .................................60-67
4.	Scatter Diagrams for “Reliability Series A” 67-73
5.	Tabular Presentation of Reliability or Consistency Results ...............................73-81
6.	Summary ...................................... 82
VI.	The Possibility of Predicting Outcome on
Parole.........................................83-103
1.	The Problem of Utilizing Parole Record
Information ................................83-86
2.	Prediction Methods Attempted in the
Present Study ........................... ..86-88
3.	Some Suggestive Results from the Attempts
at Predicting Outcome on Parole in the Present Study .............................88-100
4.	Some Considerations in the Application of Prediction Tablesi.............................. 100-103
VII.	Summary and Conclusions ............104-106
APPENDIX A ...................  107-118	.
APPENDIX B .................... 119-121
APPENDIX C .....................123-125
APPENDIX D ...................  127-131
APPENDIX E .....................133-135
LIST OF TITLES CITED ....... ....137-138
VIILIST OF TABLES
Page
Table
I. Inmate’s Work Record Prior to Conviction in Relation to Parole Violation for 1000 Joliet Cases as Found in Burgess’ Study of Parole in Illinois, 1922-24 ...............14
II. Experience and Expectancy Rate of Parole Violation for 1000 Joliet Cases Found by Burgess in His Study of Parole in Illinois, 1922-24 ................................. 15
III. Probable Post-parole Criminality Rates Found by the Gluecks’ Study of the Massachusetts Reformatory, 1921-22. Based on the total failure scores on the six highest pre-reformatory factors and the highest reformatory
factor ..................................................    18
IV.	Population Statistics for the Minnesota State Reformatory for the Five Year Period, 1922-27, as Indicated in the Published Biennial Reports................................. 21
V.	Population Statistics for the Minnesota State Prison for the Five Year Period, 1922-27, as Indicated in the Published Biennial Reports.......................................... 22
VI.	Statistics of Inmates Paroled from the Minnesota State Reformatory during the Five Year Period, 1922-27, as Indicated in the Published Biennial Reports ..................... 23
VII.	Statistics of Inmates Paroled from the Minnesota State Prison during the Five Year Period, 1922-27, as Indicated in the Published Biennial, Reports ..........................24
VIII.	Comparison of the Yearly Totals of Inmates Released on Parole Obtained in This Study and Those Published in the Biennial Reports of the Institutions. (Minnesota State Prison and the State Reformatory, 1922-27) .................. 24
IX.	Distribhtions as to Outcome on Parole for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27 ........................ 32
X.	Distributions of Major and Minor Violations in the Total Group of Violators for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27 ....................................... 32
XI.	Total Number in Each and Percentage Violating Parole Shown for Each Subclass of 9 Different Categories Involving Factors of the Trial and Commitment for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27 .......... 33
viiiLIST OF TABLES
Table
XII.
XIII.
XIV.
XV.
XVI.
XVII.
XVIII.
XIX.
Page
Sigma Values and the Range of Plus and Minus Sigma for ‘‘Percentage who Violate” in the Subclasses of the Category “County from which Received” for the Minnesota State Prison, 1922-27 ....... 37
Standard Errors of the Differences in Violation Rates for all Subclasses Under the Category “County from Which Received” with Probability Stated in Terms of “Chances in 100 that the True Difference is Greater than Zero.” (Minnesota State Prison, 1922-27) ............................................... 38
Total Number in Each and Percentage Violating Parole Shown for Each Subclass of 12 Different Categories Dealing with the Inmate’s Social Background for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27 ..................................... 39
Total Number in Each and Percentage Violating Parole Shown for Each Subclass of 10 different Categories Dealing with the Traits, Habits, and Characteristics of the Individual for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory,	1922-27 ............. 44
Total Number in Each and Percentage Violating Parole Shown for Each Subclass of 5 Different Categories Dealing with Factors Associated with the Stay in the Institution, for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27 ....................... 47
Total Number for Each and Percentage Violating Parole Shown for Each Subclass of 9 Different Categories Dealing with Factors Involved in the Period on Parole for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27 .................................. 48
Time Interval Between Discharge from Parole and Expiration of Maximum Sentence for the Group of Non-violators in the Minnesota State Prison, the
State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27 ...................... 51
Number of Months Served on Parole by the Group of Non-violators in the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27 ...................... 52
IXLIST OF TABLES
Table
XX.
XXI.
XXII.
XXIII.
XXIV.
XXV.
XXVI.
XXVII.
Page
Number of Months Served on Parole before Viola tion in the Group of Parole Violators for the Min-nesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory,
1922-27 .............................................. 52
Number of Months Served on Parole before Violation Indicated by Monthly Periods and with Cumulative Percentage Figures for the Combined Group of Minnesota State Prison and the State Reformatory, 1922-27 ....................................... 53
Number and Percentage of Men Placed on Parole Between July 1st, 1922 and June 30th, 1924 who had Subsequent Criminal Records by June 30th, 1929 (four to six years after discharge from parole). Information from the official parole files for the Minnesota State. Prison and the State Reformatory, 1922-29 ..............................................  54
Contingency Coefficients Obtained from the Comparison of the Relation Between “Outcome on Parole” and Each of 44 Other Factors, Arranged in Rank Order for the Combined Minnesota State Prison and State Reformatory Group, 1922-27 ............................ 56
Comparison of r and C Values Obtained in Reliability Series A (Own 1st entry with own 2nd entry) on the Category “Occupation at or before Conviction,” for Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 ....................................... 63
Summary Table of the Reliability Results for Series A (Own 1st entry with own 2nd entry) on All Categories Except those Dealing with “Habits and Character Traits of the Inmate” for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 ................. 79
Percentage in Full Agreement on the Scatter Shown for All Categories for Each of the Four Reliability Series. Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 ......................................  80
Percentage in Full Agreement on the Scatter Shown for All Five Categories Dealing with “Habits and Character Traits of the Inmate” for Each of the Four Reliability Series. Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 .......................... 81
xLIST OF TABLES
Table
XXVIII.
XXIX.
XXXIII.
Pa?,e
Contingency Coefficients, in Rank Order, Obtained in Comparing “Outcome on Parole” with 34 Preparole Factors in the Combined Minnesota State Prison and State Reformatory Group, 1922-27 ............... 84
Experience and Expectancy Rate of Parole Violation in Minnesota State Prison and Reformatory, 1922-27;
Whole Group of 1192 Cases. Distributions based on the average violation rate for Prison and Reformatory (21.8%). (Burgess method on 17 highest preparole factors) ........................................... 90
XXX.
Experience and Expectancy Rate of Parole Violation in Minnesota State Prison and Reformatory, 1922-27: Operating Group of 597 Cases. Distributions based on the average violation rate for the Prison and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors) ........................
XXXI.
XXXII.
Experience and Expectancy Rate of Parole Violation in Minnesota State Prison, 1922-27: Whole Group of 542 Cases. Distributions based on the average violation rate for the Prison (24.7). (Burgess method on the 17 highest pre-parole factors) ....................
92
93
Experience and Expectancy Rate of Parole Violation in Minnesota State Reformatory, 1922-27:	Whole
Group of 650 Cases. Distributions based on the average violation rate for the Reformatory (19.4%). (Burgess method on the 17 highest pre-parole factors) ............................................ 94
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison and Reformatory, 1922-27: Operating Group of 597 Cases. Distributions based on violation rate percentages for the Prison and Reformatory group. (Glueck method on the 17 highest pre-parole factors) ................. 96
XXXIV. Experience and Expectancy Rate of Parole Violation in Minnesota State Prison, 1922-27:	Whole
Group of 542 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on the 17 lowest pre-parole factors) ...........................................
99
XXXV. Experience and Expectancy Rate of Parole Violation in Minnesota State Prison, 1922-27:	Whole
Group of 542 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on 25 selected pre-parole factors)....101
xiLIST OF TABLES
Table
XXXVI.
XXXVII.
XXXVIII.
XXXIX.
XL.
XLI.
XLII.
Page
Experience and Expectancy Rate of Parole Violation in Minnesota State Reformatory, 1922-27:
Whole Group of 650 Cases. Distributions based on the average violation rate for the Reformatory (19.4%). (Burgess method on 25 selected preparole factors) ......................................102
Experience and Expectancy Rate of Parole Violation in Minnesota State Prison and Reformatory, 1922-27:	Control Group of 595. Distributions
based on the average violation rate for Prison and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors) ........ .................108
Experience and Expectancy Rate of Parole Violation in Minnesota State Prison, 1922-27:	Whole
Group of 542 Cases. Distributions based on the average violation rate for the Prison and Reformatory (21.8%). (Burgess Method on the 17 highest pre-parole factors) ..................................109
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Operating Group of 272 Cases. Distributions based on the average violation rate for the Prison and Reformatory (21.8%) (Burgess method on the 17 highest pre-parole factors) .......................110
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Control Group of 270 Cases. Distributions based on the average violation rate for the Prison and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors) ..........................Ill
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27:
Whole Group of 650 Cases. Distributions based on the average violation rate for the Prison and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors) ..........................112
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27: Operating Group of 325 Cases. Distributions based on the average violation rate for the Prison and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors) ...........................113
xiiLIST OF TABLES
Table
XLIII.
XL IV,
XLV.
XLVI.
XLVII.
XLVIII.
XLIX.
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27:
Control Group of 325 Cases. Distributions based on the average violation fate for the Prison and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors) ......................
Experience and Expectancy Rate of . Parole Violation in the Minnesota State Prison, 1922-27: Operating Group of 272 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on the 17 highest pre-parole factors ................................................115
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Control Group of 270 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on the 17 highest pre-parole factors) ...............................................116
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27: Operating Group of 325 Cases. Distributions based on the average violation rate for the Reformatory (19.4%). (Burgess method on the 17 highest preparole factors) ........................................117
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27:
Control Group of 325 Cases. Distributions based on the average violation rate for the Reformatory (19.4%). (Burgess method on the 17 highest preparole factors) ........................................118
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Operating Group of 272 Cases. Distributions based on violation rate percentages for the Prison and Reformatory Group. (Glueck method on the 17 highest pre-parole factors) ......,.........................120
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27: Operating Group of 325 Cases. Distributions based on violation rate percentages for the Prison and Reformatory group. (Glueck method on the 17 highest pre-parole factors) ............................121
xiii
Page
114LIST OF TABLES
Table	Page
L. Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Operating Group of 272 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on the 17 lowest preparole factors) ...........................,..............124
LI. Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Control Group of 270 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on the 17 lowest pre-parole factors) ...................................................125
LII. Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Operating Group of 272 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on 25 selected pre-parole factors)..........128
LIII. Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27:	Con-
trol Group of 270 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on 25 selected pre-parole factors)....129
LIV. Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27: Operating Group of 325 Cases. Distributions based on the average violation rate for the Reformatory (19.4%). (Burgess method on 25 selected preparole factors) ............................................  130
LV. Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27:
Control Group of 325 Cases. Distributions based on the average violation rate for the Reformatory . (19.4%) ...........................................    131
xivLIST OF FIGURES
Table	Page
1.	Scatter Diagrams for the Category “Occupation at or be-
fore Conviction” (Scale I) on Reliability Series A (Own 1st entry with own 2nd entry) for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 .....................................64
2.	Scatter Diagrams for the Category “Occupation at or be-
fore Conviction” (Scale II) on Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 ............... 65
3.	Scatter Diagrams for the Minnesota Institute of Child Wel-
fare Classification of “Occupation at or before Conviction” on Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 ........................................  68
4.	Scatter Diagrams for the Category, “Social Type of Inmate”
on Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 ......................................... 69
5.	Scatter Diagrams for the Category, “Type and Size of Com-
munity in Which Inmate was Brought Up,” Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.....71
6.	Scatter Diagrams for the Category “Type and Size of Com-
munity in Which Offense was Committed,” Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.....72
7.	Scatter Diagrams for the Category “Type and Size of Com-
munity Into Which Inmate was Paroled,” Reliability Series A for the Minnesota State Prison,” the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.....73
8.	Scatter Diagrams for the Category “Inmate’s Work Habits
Prior to Conviction,” Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 .....................74
9.	Scatter Diagrams for the Category “Outcome on Parole,”
Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 ...........................................75
10.	Scatter. Diagrams for the Category “Mobility before Conviction, Native-born and Foreign-born,” Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 .............76
xvLIST OF FIGURES
Table	Page
11.	Scatter Diagrams for the Category “Mobility before Con-
viction, Native-born only,” Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 ................77
12.	Scatter Diagrams for the Five Categories Dealing with “Hab-
its and Character Traits of the Inmate,” Reliability Series A (Own 1st entry with own 2nd entry) for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory 1925 ........ .................... 78
13. Scatter Diagram Obtained in Applying the Two Scoring . Methods (Burgess and Glueck) to the Data of the Minnesota Prison and Reformatory “Operating” Group of 597 Cases, 1922-27 ......................................................97
xviCHAPTER I.
Introduction
1.—General
PAROLE has become a recognized and widely used part of the American penal system.1 By 1925 forty-six states had made definite legal provisions for the release of prisoners on parole. Mississippi and Virginia were alone in having no such laws. 2
As the term is ordinarily used, parole has reference to the release from a correctional institution of an offender who, however, remains under the control of the correctional authorities and subject to whatever supervision they may prescribe until such a time as a final discharge is granted. As such, it presupposes confinement in a prison or reformatory. This imprisonment is presumed to have some sort of cathartic or reformatory effect upon the individual so that upon release he will “go his way and sin no more.” Parole is the last step in this correctional scheme. It seeks, on the one hand, to discover if the offender is prepared (i. e., sufficiently reformed) to take his place in society as a free man, and on the other hand, to facilitate the process of reentry into society by exercising some degree of supervision.
The procedure of parole may be incorporated into a system of indeterminate sentences, or, it may be utilized in connection with specific sentences. The term “indeterminate sentence” refers to the fact that the exact period of imprisonment is not fixed before custody begins but is determined by an administrative board sometime after incarceration. The term “parole” refers to the fact that an administrative board may authorize a portion of the period of custody to be spent outside of the institution subject to such conditions and supervision as the board may prescribe.
2.«—Organization of Parole in Minnesota
When the Minnesota State Reformatory (St. Cloud) was established in 1889 a law was enacted providing for a “reformatory plan” of sentences which included a system of parole. In 1911 a general parole
1	This should not be construed to mean that parole practices are peculiar to American penology. Analagous English and Continental practices have not been considered since the chief purpose of this study has been the application of certain specific methodological techniques to the problem of “predicting” the outcome on parole in Minnesota on the basis of information in the present parole records of this state.
2	This statement of the extent of the parole system in the United States is taken from Parole and the Indeterminate Sentence by A. A. Bruce, A. J. Har-uo, W. E. Burgess, and J. Landesco, p. 48. Illinois Board of Parole, Chicago, 1928.PREDICTION METHODS AND PAROLE
and indeterminate sentence law was enacted by the legislature and made applicable to the State Prison (Stillwater) as well. This act specifically provided that
the Court in imposing sentence shall not fix a definite term of imprisonment but shall sentence every such person to the state reformatory or to the state prison, as the nature of the case may require, and every such sentence shall be without limit as to time, and the person sentenced shall be subject to release on parole and to final discharge by the Board of Parole as hereinafter provided, but imprisonment under such sentence shall not exceed the maximum term provided by law for the offense for which such person shall be convicted; .... 1
The law also set up a separate Board of Parole to administer the parole function and to determine the exact period of servitude that should be required in each case.
Relatively few changes have been made in the parole system since 1911. One of the most important changes, perhaps was that involved in the amendment of 1917 which authorized courts to fix a lesser maximum sentence in individual cases than that provided in the statute for that class of offenses. With the establishment of the State Reformatory for Women (Shakopee) (Act of 1915) the indeterminate sentence and parole features were made to apply to that as well.
For the past several years the Parole Board has consisted of five members, namely: (l)i the member of the State Board of Control oldest in continuous service as a member of that board is ex-officio a member of and the chairman of the Board of Parole; (2) the warden of the State Prison; (3) the superintendent of the State Reformatory for Men; (4) the superintendent of the State Reformatory for Women; (5) a citizen member appointed by the governor with the consent of the senate. Of these five, only three have an active vote in deciding a case up for parole, namely, the chairman, the citizen member, and the warden or superintendent of the institution in which the inmate is confined. The other two members may discuss the case and offer advice but may not vote. Thus, for1 example, at the State Prison paroles are granted or denied by a majority vote of the chairman, the citizen member and the warden. At the State Reformatory those voting are the chairman, the citizen member, and the suprintendent of the reformatory.
The law places on the Parole Board the responsibility for determining the exact length of sentence to be exacted of a man and also charges it with determining whether any part of such sentence shall be spent outside of the institution on parole. In any given case there are, therefore, several courses of action which the Parole Board may follow:
1. It may keep the prisoner confined in the institution until the expiration of the maximum sentence fixed by statute for his offense (or a lesser maximum fixed by the court).
1 General Statutes of Minnesota, 1911, Chapter 298, Senate File No. 9.CHAPTER ONE
3
, 2. It may keep the prisoner confined in the institution for some period of time within the limits of the minimum and maximum sentence fixed by statute or court and then give him his liberty on a final discharge.
3.	It may release a prisoner from the institution and place him on parole. When such release shall take place and the length of time under supervision outside the institution before discharge are questions determined by the Parole Board alone,—^within the limits of the minimum and maximum sentence. That is, no one may be discharged from parole before the expiration! of his minimum sentence (less good time) and no one may be.kept on parole beyond the expiration of his maximum sentence (less good time).
4.	It may release a man from the institution or from parole by means of a “conditional discharge.” A condition frequently imposed in such cases is that the offender leave the state and stay out of it at least until the expiration of his maximum sentence. Evidence of the violation of the conditions imposed is sufficient to return a man so discharged to the institution for the balance of his sentence.
The various alternatives open to the parole board may perhaps be made more clear by means of a concrete illustration showing the meaning of “minimum” and “maximum” sentence in actual practice. Suppose a man is convicted on a charge of 2nd degree robbery and the court imposes the statutory sentence. For this , offense the law provides a penalty of “imprisonment in the state prison for not less than two years and not more than fifteen years.” This, however, is subject to “good time” reduction,—that is, the time actually to be served is shortened by two months for each period of six months of “good behavior” in prison.1 The limits then become: minimum one year, seven months, and six days; the maximum ten years, three months, and eighteen days. In this case, then, whatever alternative the parole board may choose to follow, its authority is confined by law to action within the limits indicated.
3.—-The Problem
The foregoing review of the alternatives open, to the parole board in any given case serves to make clear the complex nature of its function. The most important practical question confronting the board is that of determining what part of an offender’s sentence is to be spent in confinement within the institution and what part (if any) is to be served outside under supervision. In other words, it must determine for every case, first, whether the man is to be released on. parole, and second, at what time he is to be released. These are the two major questions involved in the parole function; the length of time on parole, the nature of the supervision, or special conditions or restrictions to be imposed, are of secondary importance.
1 This is an abbreviation of the detailed schedule for calculating “good time” and is not strictly accurate but gives a fairly close approximation for purposes of general reference.4
PREDICTION METHODS AND PAROLE
How is the board to determine these questions? On the basis of what information is it to reach its decisions? The law makes general provisions but leaves the determination of specific procedure entirely to the discretion of the board. The law provides that
In considering applications for parole, or final release, said Board shall not be required to hear oral argument from any attorney or other person not connected with the prison or reformatory in favor of or against the parole or release of any prisoners, but it may institute inquiries by correspondence, taking testimony or otherwise, as to the previous history, physical or mental condition and character of such prisoner, and each member of said Board is hereby authorized to administer oaths to witnesses for every such purpose. . . . Each prisoner shall be credited for good prison demeanor, diligence in labor and study and results accomplished, and be charged for derelictions, negligences and offenses under such uniform system of marks or other methods as shall be prescribed by the Board. . . . Said Board of Parole shall have power from time to time to make, alter, amend and publish rules governing the granting of paroles and final discharges and the procedure relating thereto, ... as said Board may deem proper.1
In an attempt to answer the questions raised above, the first Parole Board adopted (in 1911) a rule which has since been in continuous effect:
Rule 5. In considering the question of the parole of an inmate, the Board will take into account:
(a)	His conduct while an inmate of the institution.
(b)	His history previous to commitment.
(c)	His character, capacity, habits, tendencies and attitude.
(d)	The nature and circumstances of his crime.
(e)	The probabilities from all the facts that the prisoner will lead
a correct life at liberty, and
(f)	The effect of his parole upon the administration of justice. 2
Rules six and seven, also in continuous effect since 1911, (with slight changes in wording) make the further statement:
Rule 6. Each prisoner, shortly after entering the institution, will be required to give a statement of his past history, his past employment, references as to character, and detailed statement of his offense. This statement shall be a part of the permanent record of each inmate and will be used for the purpose of investigation and verification as to his past record, and will be presented with all other records and information to the State Board of Parole when the inmate’s application for parole or discharge is under consideration. 3
1	Minnesota General Statutes, sections 9273, 9274, 9279.
2	First Biennial Report, Minnesota State Board of Parole, 1911-1912, p. 5.
3	This wording is from the Ninth Biennial Report, 1926-1928, p. 53; the same rule appears in almost the same words in the First Biennial Report, p. 5.CHAPTER ONE
5
Rule 7. An inmate declining or failing to give a truthful and satisfactory history of his past life will not be paroled and will be discharged only by expiration of sentence. The purpose of requiring information is to enable the officials to properly discharge their several duties and not for the purpose of humiliating the inmate or his friends. 1
It would appear from the above that the only information given definite weight in determining parole is the matter of a “truthful and satisfactory history of his past life.” Failure to give such a history will, according to the rule, prevent parole and lead to discharge only by expiration of sentence. How to determine whether the history given is “truthful” or “satisfactory” is not indicated. What weight should be given to the various factors mentioned in rule five is nowhere indicated. Apparently that remains unverbalized, a changing and indefinite element playing an inexplicable and uncertain role in the actual parole decisions of the board.
What information, in the parole records thus accumulated by the Board, is important as an indicator of probable conduct on parole? How may the Board know, in any given case, whether it is taking a serious chance or acting on a relative certainty in the matter of an inmate’s probable conduct on parole? It is to these questions that the present study is addressed. It seeks to formulate a technique whereby the information in the records of the Board of Parole may be utilized in a systematic manner in arriving at an answer to the question of whether any given man is to be paroled. If, from the study of a man’s past life the question of his behavior on parole can be consistently answered in terms of probability, thén it would seem that a device of great practical value is at hand for the better discharge of the parole function.
1 First Biennial Report, p. 5. (Also in all subsequent Biennial Reports).CHAPTER II.
Earlier Studies and Material Dealing With Parole
The widespread incorporation of parole into American penal treatment has called forth a considerable amount of discussion, pro and con. The many official reports of parole boards or other correctional officers nearly all offer some discussion of parole with special reference to a justification of their own particular procedure. In addition there have been numerous contributions by interested individuals in the form of books, pamphlets, articles, and interviews. Without making any attempt at a complete survey, it may still be profitable to review briefly some phases of this material.
This mass of literature may, for the present purposes, be roughly classified as of four kinds: (1) material dealing with the general question of the desirability or undesirability of parole; (2) material dealing with the specific problems involved in the selection, supervision and care of inmates on parole; (3) general quantitative studies of the results or effects of penal treatment; (4) specific quantitative studies of factors involved in conduct on parole. A brief glimpse of each of these fields follows in order.
1—Discussions of the Desirability or Undesirability of Parole
The great mass of material in this field can be passed over without comment.. Much of it is incorporated into discussions of the indeterminate system and of the ideal of reformation as the proper objective of penal treatment. It is largely of a theoretical and non-quantitative nature with frequent appeals to sentiment. The numerous articles, speeches, and special reports of such men as Wines, Brockway, Hubbell, and others in connection with the establishing and developing of “the Elmira System” fall into this $lass. 1 Here, too, belong many of the reports of the increasingly numerous “crime commissions.”
The desirability of parole is sometimes deduced, however, from
1 Especially noteworthy from the historical standpoint is Brock-way's famous paper “The Ideal of a True Prison System for a State” read before the first meeting of the American Prison Congress in 1870; also the report by Wines and Dwight .to the New York Legislature in 1867 on “Prisons and Reformatories of the United States and Canada” in which they urge reformation as an ideal and advocate the “Irish system” including “ticket^of-leave” or parole.
6CHAPTER TWO
7
studies of a more or less quantitative nature. A study by Gault 1 in 1915 may be taken as an illustration of this type. After listing the (then) thirty-five states that utilized some form of parole, he attempted to get some measure of the extent to which the practice may be said to be successful. For this purpose he took the official parole reports of eighteen states involving a total of 38,593 inmates placed on parole. Of this number only 17.4% were reported as “parole violators”. Considering the question of the kind of conduct involved in parole violation, Gault found, from incomplete reports, that only about 5% of those reported as “violators” had violated by committing a felony. His conclusion is consequently highly optimistic. He says, “It seems safe to say, therefore, that 80% of men paroled fulfill the conditions and become law-abiding citizens.” 2
The obvious criticism of Gault’s conclusion, and of the many similar ones repeatedly published by parole and correctional officers, is that it is utterly unwarranted. The fact that only 15-20% of men violate their parole agreements in such a manner that they are reported as “violators” does not warrant the conclusion that the other 80-85% have become law-abiding citizens. One of the major difficulties involved in attempts to measure results of the penal system lies in the fact that the correctional authorities have no way of keeping in touch with a man once he is discharged. Men who have been discharged from parole may, therefore, be guilty of considerable petty crime and vagrancy without coming again to the attention of the penal authorities; they may change their names and thus serve repeatedly as first offenders, especially if they are vagrants drifting about from place to place; they may leave the state and become involved in further crime without having the authorities in the first state aware of that fact; they may, so long as there is no central clearing bureau of information and identification, remain quite anonymous and succeed in concealing their past records. That being true, it is preposterous to take the figures for non-violation of parole in the various states as synonymous with law observance and non-criminality in the group discharged from parole.
Other studies using different methods have arrived at results somewhat similar to those reported by Gault. Thus Clark, 3 reporting on the condition of 252 Whittier School boys two years .after release from the institution, gives the following distribution:
' 42.7 % doing well 4 30.2% doing fairly well 27.1% doing poorly
1	Gault, Robert H., “The Parole System, A Means of Protection,” Journal of Criminal Law and Criminology, 5:799 et seq., March 1915.
2	Gault, R. H., op. cit., p. 801.
3	Clark, W. W., “Success Records of Prisoners and Delinquents,” Journal of Delinquency, 6:444 et seq., July 1921.
4	For a more detailed statement of the meaning of these, terms see page 11 of this chapter.8
PREDICTION METHODS AND PAROLE
Just how information on conduct after discharge was obtained is not made clear. As the results stand, however, 72.9% are indicated as “doing well” or “fairly well” with only 27.1% reported as having a definitely negative and unsatisfactory record.
Recently Knox 1 reported on a “ten years after” follow up study of 103 Whittier School boys admitted to the institution in 1917. Most of these boys had been discharged for eight or more years. Information was obtained on only eighty boys for the follow-up study in 1927. Success was rated by Knox as:
This indicates only 38% as known failures leaving 62% to be classed as “success.” With incomplete information, as in this case, unknown selective factors make the figures untrustworthy as they stand.
The relatively high percentages of “successes” indicated in these three studies are interesting primarily by way of contrast when compared with the results of the Gluecks, recent careful and painstaking study. They arrive at almost diametrically opposite conclusions: “. . . out of 510 men who left Massachusetts Reformatory during the years 1911-1922, eighty percent were not reformed five to fifteen years later, but went right on committing crimes after discharge.” 2 The Gluecks* study, it should be added, rests on far more complete and reliable information than the others. The validity of the general conclusion need not be discussed here. Other aspects of the study will be taken up later in this chapter.
2.—Discussions of the Aims, Ideals and Technique of Parole Administration
Much attention has been given problems of selection, supervision, and care of inmates on parole. Most of this material is of the “common sense” type of discussion. Only a few examples can be touched upon here. Many verbal formulations have been made in attempts to set up general principles governing release on parole. The following, by Spalding, from a symposium on parole, is a good example:
The only questions for the parole board are, “Can the prisoner be trusted with limited liberty ?” “Will he probably use it in a way which is consistent with his own best interest and
1	Knox, E. E., “Follow-up of 103 Whittier State School Boys Ten Years after Admission,” Journal of Delinquency, 12:261-269, Sept.-Dee. 1928.
2	Glueckj Sheldon and Eleanor T., 500 Criminal Careers, Foreword, p. XII. (Forword by Richard C. Cabot). New York, Alfred A. Knopf, 1930.
A—Notable success B—Very good success C—Average success D—Poor success (doubtful) E—Known Failures
2.5%
7.6%
48.1%
3.8%
38.0%CHAPTER TWO
9
the interests of the community?” Even these questions need not be answered positively. It is no more required of a board of parole that it be certain that he will be a good citizen after his discharge, than it is required of the court in fixing in advance the date of discharge, under the definite sentence. . . . The advantage of the parole system is that if the parole board makes a mistake it can correct it by returning the offender to prison, while if the judge makes a mistake . . . it can not be corrected.1
It is probably true that, this principle of “fitness for liberty” is, in general, accepted by most parole boards in determining who shall be paroled. But an abstract principle such as this must somehow be made concrete. Hence the boards have been in the habit of considering various information regarding the nature of the crime, the prior criminal record, conduct in prison, length of time already spent in prison, mental and physical condition, alien-st’s report, doctor's statement of medical diagnosis, etc. The report of the Prison Survey Committee of New York offers the following suggestions:
The committee recommends as a matter of supreme importance that the board of pardon and parole base its discussions for granting or refusing parole upon definite and supporting data. When the prisoner applies for parole, the board should have laid before it the following records: (a) a record of main facts of the trial resulting in sentence; (b) a history of the prisoner, supplied by the receiving station; (c) a certificate from the receiving station stating the prisoner’s mental and physical ability and limitations; (d) a record of the prisoner’s conduct during his prison sentence; (e) a record showing the prisoner’s work record, including the work to which he has been assigned and his application and progress therein together with the amount of money he has earned; (f) a record showing the extent of the prisoner’s education,—if he was an illiterate on his entry the record will show his progress in the prison school in reading, writing and speaking of English, if already proficient in elementary studies, school attendance will show his interest in self-improvement; (g) a statement of the person or agency assuming responsibility for the prisoner and the employment which he is to undertake. 2
The above was formulated in 1920, but it will be observed that it is no more definite as to how much weight is to be assigned to any one factor or item of information than the rule adopted by the Minnesota Board of parole in 1911. A number of studies have been made, however, in attempts to find out what connection there may be between certain traits or conditions in the prisoner or in his circumstances and his success or failure on parole. Several of these
1	Spalding, W. F., “Parole and Probation, A Symposium,” Proceedings of American Prison Association 1916, pp. 464-465.
2	Prison Survey Committee Report, State of New York, Albany, 1920, p. 254. Of this same “common sense” advice type are the following: Doll, “Classification of Prisoners for Purposes of Training Work and Parole,” Journal of Criminal Law and Criminology, 14:110 ff., May 1923; Butler, “What Prisoners Should Be Eligible to Parole and What' Considerations Should Govern the Granting of It,” Journal of Criminal Law and Criminology, 12:549-553, February 1922.10
PREDICTION METHODS AND PAROLE
are not restricted to parole alone, but seek to measure the role of traits or conditions of the individual and the outcome of penal treatment as a whole. This group is considered in the next section.
3.—Traits or Conditions of the Individual and the Outcome of Penal Treatment
Intelligence is a trait of the individual frequently held to be an important factor involved in crime causation and in the success or failure of penal treatment. Many studies have been made of the distribution of intelligence in prison and reformatory groups. These hardly come within the scope of the present effort. A few studies, however, have tried to measure the relation of intelligence to the “success” or- “failure” of an individual after release from penal treatment. It is to this type that attention is directed in this section.
An interesting study of this sort was made by Pintner and Reamer and reported in 1918. 1 Working with a group of twenty-six delinquent girls (wards of the Big Sisters Association), ranging in intelligence from 70 to 109 C. M. A. (coefficient of mental ability, Yerkes-Bridges scale), they tried to get a measure of “success” that could be compared with intelligence data. The two authors together with the matron of the Association home prepared, separate ratings of the “future probable success” of the girls. The combined judgments of the three observers was then taken as the rating for aftersuccess. This was then compared with the intelligence scores, giving an r of + .16 for the whole group of twenty-six girls. The correlation for the individual ratings of the judges, however, vary widely. Thus:
Mental Ability	and	A’s	judgment	+.32
Mental Ability	and	B’s	judgment	—.02
Mental Ability	and	C’s	judgment	+.03
Their conclusion is that “mental tests are not prognostic of their (the girls’) success after leaving the Big Sisters Home.” The weakness of this study (aside from the small number of cases), as admitted by the authors, lies in that “it lacks a sound criterion by which to measure success in the world.” Obviously the “rating” of the three observers as to future success is itself an uncertain element, the degree of accuracy of which must be established before it can have much meaning. The significance of this study does not lie in its materials, nor in its results, nor in its conclusions, but in its recognition of a problem for quantitative study.
A more detailed study of 301 Whittier School boys “paroled, furloughed, or discharged . . . during the biennium of July 1, 1916 to June 30, 1918” has been reported by Clark. 2 He had( information on
1	Pinter, R. and Reamer, J. C., “Mental Ability and Future Success of Delinquent Girls,” Journal of Delinquency, 3:143-148.
2	Clark, W. W., “Success Records of Delinquent Boys in Relation to Intelligence,” Journal of Delinquency 5:174-182, 1920.CHAPTER TWO
11
the subsequent conduct of 223 boys, using the following criteria of success:
1.	Doing well—those who “have developed into men of good character fulfilling the requirements of good citizenship/’ or, if minor wards, are “meeting the requirements of good homes.”
2.	Doing fairly well—Less successful than above but “give promise of becoming self-supporting, respectable citizens.”
3.	Doing poorly—those “regarded by the community in which they live as undesirable citizens” or who “do not give promise of becoming useful.”
In comparing this three place scale of success or failure with intelligence, Clark found an r of +*19. His cases were divided into six industrial groups and a wide range of differences appeared in the correlations between, intelligence and after-success in the various groups. He concludes, “There is a distinct tendency for the boys of higher intelligence to have a better record of success than those of lower mentality.” The greatest weakness of this study lies in the selective factors that may be involved in that part of the group on which information about subsequent records is lacking. There is considerable possibility that the delinquent ones would take pains to conceal their identity, or would be more likely to leave the territory than the non-delinquents.
Even if taken at their face value, Clark’s results show only a relatively slight positive correlation. This is in keeping with the results of another study1 of his involving the relation of several factors to “conduct under supervision” of a group of 143 Whittier School cases. Supervising officers were required to grade the conduct responses of the boys in their charge on a five place scale as, bad, poor, fair, good, excellent. Values from 1 to 5 were assigned to these and comparisons made with factors such as the following: length of stay in institution, race, age, mental age, influence of previous offenses, mental retardation (I. Q.) and home conditions. In every case he found little or no relation between the specific factors and conduct while under supervision.
At best these studies show a low positive correlation between specific factors in the individual’s make up, or in his surroundings, and his conduct while under supervision or after discharge from supervision. Another study along this line is that by Heacox 2 of the parole violators at Auburn for the year 1915. It dealt with only twenty-nine cases and scarcely warrants extensive discussion. Taking factors such as home condition, mental peculiarity, unfavorable married life—ten in all, he sought to determine for each case whether the factor was major or minor in its influence. The distrib-
1	Clark, W. W., “Supervised Conduct Response of Delinquent Boys,” Journal of Delinquency, 6:387-401, May 1921.
2	Heacox, F. L., “A Study of One Year’s Parole Violators Returned to Auburn Prison,” Journal of Criminal Law and Criminology, 7:233-258, July 1917.12
PREDICTION METHODS AND PAROLE
ution of major and minor factors in the group of violators was then compared with their distribution in an unselected group of prisoners. In no case does any very pronounced difference in the distribution of major and minor factors appear in the group of violators when compared with the unselected prison group.
None of the studies of the relation between the traits and conditions of the individual and his conduct during a period of time, discussed in the last few pages, is very satisfactory either from the standpoint of method or of results. Yet, the general trend of the results may have some significance. Low positive correlations are common, but in no case is there evidence of any trait or condition of outstanding importance.
4.—Specific Quantitative Studies of Parole Factors
Mention has already been made of the common practice of parole boards of collecting various sorts of information about the inmate as a guide in determining whether or not to grant a parole. An elaborate attempt to determine the importance of these different items of information in the parole records of the Massachusetts Reformatory was made by Warner for the “Committee on Criminal Records” of the Institute of Criminal Law and Criminology. 1 His study covered the ten year period 1912-1921, during which 5,434 prisoners had been received at the Reformatory and a total of 4,789 granted release on parole. The actual study deals with 680 cases. “From the alphabetical file in the office of the Commissioner of Correction, starting with the J’Sy every folder was examined until 680 had been covered, 300 parole successes, a like number of parole violators, and eighty not paroled.” He secured quantitative data for analysis on some sixty odd items of information in the records. The plan of analysis adopted may be seen in the distribution of his cases under the cate-
gory “Character of Father.”			Not
Character of Father	Successes	Violators	Paroled
Good	62%	57%	64%
Fair	3%	4%	4%
Questionable	2%	2%	0
Bad	3%	7%	6%
Drunk	25%	22%	15%
Unknown	5%	8%	11%
	100	100	100
The percentage differences found above are too small to indicate very definitely any significant differences in the make-up of the three groups. Similar tables were worked out for each of the categories of information with very similar results. Information as to prior crim-
1 Warner, S. B., “Factors Determining Parole From the Massachusetts Reformatory,” (Preface by Sanford Bates, Commissioner of Correction for Massachusetts), Journal of Criminal Law and Criminology, 14:172-207, August 1923.CHAPTER TWO
13
inal record did reveal some differences. It was found, for instance, that 19% of parole successes and only 7% of parole violators had never been previously convicted. Warner, however, attaches little importance to his results and concludes that the parole board has little or no information on which to base a decision regarding fitness for release. He elaborates:
But poor as the criteria now used by the Board are, the Board would not improve matters by considering any of the sixty-odd pieces of information placed at its disposal, which it now ignores, except the Alienist’s report. ... No considerable improvement is possible without a complete change both in the methods of obtaining information for the Board and in the nature of the information obtained. The blame for the Board’s failure to obtain better criteria for parole should be placed upon the present undeveloped state of the science of criminology rather than upon either the Board of Parole or the Department of Correction. 1
To this conclusion Hornell Hart takes exception. His own words are:
That the percentage of violations of parole among men paroled from the Massachusetts Reformatory could be reduced one half through scientific utilization of data already being collected by the authorities of that institution, is the conclusion which should have been reached by analysis of statistical data presented by Professor Sam B. Warner in a recent study.
This conclusion, however, is quite at variance with those found by Professor Warner. . . . The reason for the discrepancy between . . . (his) conclusions and those of the present writer is that, while the former’s study is a most admirable, careful and valuable piece of work up to the point where he draws his conclusions from his tables, his failure to apply accurate statistical tests to determine which of the factors involved showed significant contrast, and which did not, resulted in his overlooking certain highly important differences between the men who violated their paroles and the men who succeeded. 2
Hart points out, as a specific instance, that Warner has dismissed as of no importance the fact that of 300 non-violators none reported his mother as having been arrested or sent to jail, while 4% of the violators admitted such a report of their mother. He shows that this difference (4%), small though it is, could not possibly be due to chance. He adds:	“Obviously, therefore, the sensible conclusion is that the
contrast between the two groups (violators and non-violators) is significant, and that reporting one’s mother as having been arrested tends to be prognostic of parole violation.”3
Hart elaborates his argument by pointing out that the information thus found to be of some significance under the several categories needs to be combined into prognostic scores so that the parole board may act
1	Warner, S. B., op. cit., p. 196.
2	Hart Hornell, “Predicting Parole Success,” Journal of Criminal Law and Criminology, 14:405, November 1923.
3	Ibid., p. 406.14
PREDICTION METHODS AND PAROLE
on a knowledge of the probabilities' in every case. He argues; that the board should be able to report, for example, that “Jim Jones has a score of ninety-three points—-in the past 19% of the men with this score have been violators; Smith has a score of twenty-one points— in the past 80% of men withj this score have violated.” 1
Warner’s study is thus seen to be similar in results to those discussed earlier in this chapter. Specific factors in a man’s nature or in his surroundings such as intelligence, home condition, prior criminal record, etc., are revealed as of no great significance, taken separately, in determining probable future conduct on parole. Hart argues that when many such factors are found, each showing a slight positive relation to future conduct, it should be possible to construct some kind of combined score that will take account of the total cumulative effect. Actual application of this method of cumulative prognostic scores to the problem of parole violation appears for the first time in Burgess’ section of the recently published study of parole in Illinois. 2 3
Burgess examined the official parole files of 1,000 inmates paroled from each of three Illinois institutions (3,000 cases in all) for information regarding offense, previous criminal record, family conditions, work record, age, mentality, social type, punishment record m the institution,—a total of twenty-one factors. Each of these classes was divided into appropriate subclasses and the proportion of parole violators determined for each. Table I makes this clear:
TABLE 13
Inmate’s Work Record Prior to Conviction in Relation to Parole Violation for 1,000 Joliet Cases as Found in Burgess’ Study of Parole in Illinois, 1922-24.
Previous Work Record	Violation Rate
at Joliet Prison
All persons (average for institution)	28.4%
No previous work record	44.4%
Record of casual work	30.3%
Record of irregular work	24.3%
Record of regular work	12.2%
It seems evident from the above that persons classed as “no previous work record” are more likely to violate parole than those with “record of regular work.” Yet not all those with no work records are violators, nor are all those with records of regular work free from violations. The matter of work record appears important, but not decisive. It can not be taken as the sole index of probable conduct on parole.
1	Hart, Hornell, op. cit., p. 412.
2	Bruce, A. A., Harno, A. J., Burgess, E. W., and Landesco, J., Parole and the Indeterminate Sentence, xiv + 277. Illinois State Board of Parole, 1928.
3	Burgess, E. W., op. cit., p. 229, Table XVIII. As given above the results for the other two institutions studied have been omitted with only the rates for Joliet indicated.CHAPTER TWO
15
Burgess worked out similar tables for each of the other factors on which he had information and found very similar results. That is, in every case, there were1 more or less significant differences in the violation rates of those entered in the various subclasses under each category. In this respect Burgess’ study shows accord with those already reviewed. In order to get some measure of the cumulative effect of these factors, no one of which appeared outstanding, he combined them into a single prognostic table. Taking the average rate of violation for the institution as a base, he credited each man with one point for each category under which he was classified in a subclass having a violation rate lower than the average. If he was classified in a subclass having a rate higher than the averge for his institution, no credit was given. Thus, for example, in the Joliet cases under the category “Previous work record” (Table I, p. 14 of this chapter), those men classified as having “record of irregular work” and “record of regular work” would be given credit since the violation rates of their respective subclasses are 24.3% and 12.2%, both of which are lower than the institutional average for Joliet of 28.4%. Those entered in subclasses with violation rates greater than 28.4% would then be given no credit. Since Burgess used information on twenty-one factors, the highest credit score possible for any man would be twenty-one points; the lowest would be 0. The distribution of the Joliet cases on this scale is given in the following table:
TABLE III
Experience and Expectancy Rate of Parole Violation for 1,000 Joliet Cases Found by Burgess in His Study of Parole in Illinois, 1922-24.
Points for number of factors above the average	Number of men in each group	Percent who violate parole
16-21	68	1.5
14-15	140	2.2
13	91	8.8
12	106	15.1
11	110	22.7
10	88	34.1
7-9	287	43.9
5-6	85	67.1
2-4	25	76.0
The last column in this table shows the proportion of parole violators in each group of the cumulative score, and, it may be presumed, “other factors remaining the same” that in the future men in the same score classes will show the same rate of parole violation. Of the sixty-eight men in the score class “16-21 points better than the average,” 1.5 percent were violators in the past; in the future the same proportion of offenders (scored according to the same scheme) with scores in this group may be expected to violate parole, and in terms of probability, the chances that a particular inmate with a score in, this group will violate parole are 1.5 in 100.
iBurgess, E. W., op. cit., p. 248, Table XXVII. The table is given here in somewhat simplified form.1(3
PREDICTION METHODS AND PAROLE
After this presentation of the method followed by Burgess, some of the more important criticisms of his study should be noted:
1.	It is confined to official records only and therefore incomplete and unreliable. It is likely that a number of men are actual violators of parole who escape official attention and are therefore listed as “nonviolators.”
2.	It is confined only to conduct during the official parole period. Many men discharged as “successes” on parole continue a career of crime, some serving several parole terms, each time as a “success.” The answér would seem to be that though this may to some extent be true, it is still desirable that the parole board be able to predict outcome on parole. After all, the fact that a man may commit crime after discharge from parole is no more chargeable to parole than to any other phase of the penal system. It is still important to be able to make intelligent selection of those placed on parole. If prediction tables can be developed showing the probabilities of crime after discharge, that is good and desirable, but this represents an essentially different problem.
3.	It uses categories of information that are in many cases highly subjective and overlapping. Thus, for example, the category of “social type” includes such subclasses as “hobo,” ne’er-do-well,” mean citizen, drunkard, gangster* recent immigrant, farm boy, and drug addict. The same individual could presumably easily be a hobo, a ne’er-do-well, a drunkard, a drug addict and a gangster at one and the same time. In that case the particular classification under which a man is entered would seem to be very much a matter of chance. At best it would probably be impossible for another individual to classify 1,000 cases in the same way that Burgess did.
4.	It lacks any measure of reliability or consistency. Since the classifications inevitably tend to be subjective, it is important to have a measure of the consistency with which the schedules were used. A simple test of this would have been the re-classification of a group of the same cases.
5.	It uses a system of scoring that assigns equal weight to every one of the twenty-one factors of information. It is apparent from the study itself, and from other studies, that some factors are more important than! others. This may lead to fallacious results in the cumulative index groupings.
The only effective and conclusive answer to these points of criticism is the pragmatic test of workability. If the method used by Burgess can be applied in practice to the problem of selecting men for parole and can be shown to work out, the questions of scoring method, subjectivity, and the like, dwindle in importance. Such a practical test has not, as yet, been applied.
The most complete and thoroughgoing American study of parole and the reformatory system yet published is the recent book, 500 CriminalCHAPTER TWO
17
Careers by the Gluecks. 1 They undertake the important task of actually checking up on the results of the whole reformatory system of penal treatment and, in passing, give a great deal of attention to parole. The quantitative portions are based on a rather complete study of the lives of 510 men whose parole from the Massachusetts Reformatory expired in 1921-22. In over 90% of the cases the history of each man included essentially complete information covering a five year post parole period.
Though this is a far more intensive and painstaking study than that of Burgess, the fundamental method is basically the same. The information about the men was reduced to quantitative form under some fifty categories with appropriate subclasses for each category. The proportion of men violating in each subclass was then determined much as in Burgess’ study. But where Burgess made use of all twenty-one of his factors in arriving at his “prediction tables,” Glueck sought to eliminate those that seemed of little importance as reflected by low coefficients of contingency when compared with conduct during the post parole period. That left the following list of factors on which to construct prediction tables: 2
I.	Pre-reformatory factors
1.	Seriousness and frequency of crime.
2.	Arrest for crime preceding the offense for which sentence to the reformatory was imposed.
3.	Penal experience preceding reformatory.
4.	Industrial habits.
5.	Economic responsibility.
6.	Mental abnormality.
II.	Reformatory factors
7.	Frequency of offenses in the reformatory.
III.	Parole factors 3
8.	Criminal conduct during parole.
IV.	Post-parole factors
9.	Industrial habits.
10.	Economic responsibility.
11.	Attitude toward family.
12.	Types of home.
13.	Use of leisure.
The individual inmate’s score was arrived at by adding the percentages of failure in the subclasses to which he belonged under selected factors in this list. The lowest possible score was 274,—the highest 499. The distribution of their cases with this method of scoring , is indicated in Table III on the next page :
1	Glueck, Sheldon and Glueck, Eleanor T., 500 Criminal Careers. New York, Alfred A. Knopf, 1930. The substance of Chapter 18, “Predictability in the Administration of Criminal Justice” was published separately as an article in The Harvard Law Review, 42:300-329, January 1929.
2	Glueck, Sheldon and Glueck, Eleanor T., op. cit., pp. 281-284.
3	The parole and post-parole factors were not used in the construction of parole prediction tables. They were used only in the tables in which the Gluecks sought to predict the outcome of the whole penal system.18
PREDICTION METHODS AND PAROLE
TABLE III i
Probable Post-parole Criminality Rates Found by the Gluecks’ Study of the Massachusetts Reformatory, 1921-22. Based on total failure scores on the six highest pre-reformatory factors and the highest reformatory factor.
Total Failure Probable status as to post-parole Score	criminality, (in percentages)
Total Failure
274-325	7.1
326-425	40.6
426-475	73.2
476 and over	82.8
Total (average)	63.3
The Gluecks say that this, . . prognostic table is presented as a possible model for parole boards in determining which men to release on parole, and in obtaining some true conception of the probable length of parole supervision needed in different cases.” 1 2 The same inferences with referenc to the probabilities as to future conduct may be made from this table as from those devised by Burgess. That is, 7.1% of those inmates with scores in the group 274-325 were “total failures” in the five post-parole years studied; in the future inmates with a similar score may be expected to be “total failures” at the same rate; the “chances” are 7 out of 100. This method differs from that of Burgess in that the Gluecks sought to eliminate the factors found to be of slight significance;, and, they weighted each factor used in terms of its respective percental importance.
The method used by the Gluecks meets some of the objections raised against that followed by Burgess. They studied conduct, not only during the parole period, but for a five year post-parole period; and their prediction tables are constructed on the basis of selected factors weighted in proportion to their importance. The procedure, however, raises other questions. Examination of the seven factors used in their parole prediction table makes it clear that there is heavy weighting of the factors of past criminality. Four out of the seven factors have reference to the violation of law or of prison regulations. May not this be giving too much weight to the assumption “once a criminal, always a criminal?” In the matter of selecting only the important factors, the question boils down to this, will the summation of factors, each of slight statistical significance, give results more reliable than the complete exclusion of these factors? In effect the Gluecks’ method assumes that the answer is negative; Burgess’ method assumes an affirmative answer. 3 Verbal argument can not decide which assumption is the more nearly correct. The only conclusive answer is the pragmatic one: try out both methods under conditions of experimental control.
1	Glueck, Sheldon and Glueck, Eleanor T., op. cit., p. 286, Table 113.
2	Ibid., p. 286.
3	The writer wishes to acknowledge indebtedness to Professor E. H, Sutherland for first calling attention to these points of criticism in a paper as yet unpublished.CHAPTER TWO
19
Further comparisons and criticisms of these studies will be presented in connection with some of the results of the present study. Attention should be called again, however, to the developing methodology for the study of parole procedure. The change from the early “common sense” observations of parole officers and other interested persons to the elaborate quantitative analyses of Burgess and of the Gluecks would seem to represent a step of some significance in the development of a beginning-to-be scientific criminology. The process has been marked by much fumbling and many deviations. Appeals to humanitarian sentiment have played their part; striking “case histories” have had their day; the traits of the individual have come in for their share of study; intelligence tests, psychiatric prognosis, and emotional questionnaires have been used; all have been important, all have had some significance, but none has given any very satisfactory answer to the question of how to judge in advance a man’s probable conduct on parole. The important point has been demonstrated in study after study, that no one thing, nor any combination of a few things, is very important in determining a man’s conduct on parole. (With the recent studies by the Gluecks and by Burgess, the principle has been given practical recognition that the cumulative effects of many factors, individually of little significance, may become very important when operating together. !
The present study has been made in full recognition of this principle and seeks to apply it to the parole situation^ in Minnesota. It is hoped that the present effort may become one more step in the chain of studies indicating the development of a more scientific method in the study of crime and penal administration.CHAPTER III
The Method and Technique of the Present Study
Materials for the present study have been taken from the official parole files of the Minnesota State Board of Parole.1 A careful examination was made of the records of all men placed on parole during the five year period from July 1, 1922 to June 30, 1927 from the two adult male penal institutions, the State Reformatory at St. Cloud and the State Prison at Stillwater. The Reformatory receives, in general, the young adult male offender (espeically the first offender) up to thirty years of age, while the Prison gets the older group of offenders and those with a previous criminal record. There are, however, no very hard and fast dividing lines. A man in the middle or late twenties is likely to be sent to the Reformatory if a first offender, but to the Prison if he has a previous reformatory record.
1.—The Setting: Population Statistics for the Institutions
The details of the distribution of the numbers involved in the present study as indicated in the published official Biennial Reports, are presented in Tables IV, V, VI and VII on the succeeding pages. Attention should particularly be directed to the “five year summaries” in each table. It will be observed that of the total admissions to the institutions for the period, only 35.8% were placed on parole from the Reformatory and 32.4% from the Prison. In other words, only about one third of the inmates received were paroled.
The relatively large number whose penal experience is terminated by “expiration of sentence” (either while in the institution or while on parole) should be noted. For the Reformatory “expiration of sentence” accounted for the release of 18.5% directly from the institution and of 8% of those placed on .parole. Or, putting it in other words, 26.5% of the total Reformatory admissions during the five year period had their penal experience terminated by the legal expiration of sentence. Fori the Prison 30.8% of the total admissions were finally released by the legal termination, of sentence.
The figures from the Biennial Reports, presented in the tables in the preceding pages, indicate the proportion and general numerical significance of the parole function in relation to the entire group of inmates admitted to the two institutions. They serve to give the setting for the more specific problem of selecting the rather limited group of those actually placed ion parole. The yearly totals for the number paroled are not quite the same in the Biennial Reports as found in
1 The writer desires to express grateful appreciation for the many courtesies rendered by the personnel of the Minnesota Parole Department.
20CHAPTER THREE
21
TABLE IV i
Population Statistics for the Minnesota State Reformatory for the Five Year Period,11922-27, as Indicated in the Published Biennial Reports.
	1922- 1923	1923- 1924	1924- 1925	1925- 1926	1926- 1927
In institution at begin, of year	526	629	650	755	826
Admitted :					
First Admission	309	319	413	403	446
Readmission	1	3	0	5	7
Received by Transfer:					
From Reformatory	0	0	0	0	0
From Prison	0	0	2	0	1
Returned :					
From Parole	18	18	27	30	26
From Escape	3	1	2	2	3
From State Hospitals	0	1	0	0	0
Total under care	857	971	1094	1195	1309
Discharged :					
By expiration of sentence	46	31	62	103	112
By commutation of sentence	1	2	2	4	1
By Board of Parole	80	122	109	89	86
By Board of Pardons	4	4	9	0	0
By order of court	1	0	0	0	2
Transferred :					
To Prison	5	10	19	35	18
To State Hospitals	2	3	1	1	7
Died	1	2	2	1	2
Paroled	87	145	131	132	189
Escaped	1	2	3	3	2
In institution at end of year	629	650	756	827	890
Total under care	857	971	1094	1195	1309
Five Year Summary
Received :
First Admissions 1890 Readmission	16
Total	1906
Discharged by expiration of sentence Discharged by Parole Board Paroled
Total released
Retained in institutions or otherwise accounted for
Grand Total
	Percent of
N	admissions
354	18.5
486	25.5
684	35.8
1524	79.8
382	20.1
1906	99.9
1 Compiled from the Biennial Reports of the Minnesota State Reformatory for the years 1921-22; 1923-24; 1925-26; 1927-28. (Printed at the Reformatory).22
PREDICTION METHODS AND PAROLE
TABLE V 1
Population Statistics for the Minnesota State Prison for the Five Year Period, 1922-27, as Indicated in the Published Biennial Reports.
	1922- 1923	1923- 1924	1924- 1925	1925- 1926	1926- 1927
In institution at begin, of year	912	1033	1053	1082	1230
Admitted :					
First admission	254	219	231	257	299
Readmission	97	89	99	118	112
Received by transfer:					
From Reformatory for Men	5	11	19	34	20
Returned during year:					
From parole (in good standing) 17		12	18	18	11
From parole violation	12	22	8	6	8
For violation of conditional disch. 0		4	4	6	4
From, state hospitals	7	3	2	6	1
From escape	0	1	0	0	0
From reprieve	2	1	3	2	1
Total	1306	1395	1437	1529	1686
Discharged :					
By expiration of sentence	70	80	91	88	131
By commutation of sentence	6	14	20	7	2
By Board of Parole	61	83	101	85	82
By order of court	2	2	0	0	0
Transferred during year:					
To State Hospitals	15	18	6	23	19
To Reformatory for Men	1	0	2	0	1
Released on reprieve	2	0	5	5	3
Died	6	3	10	5	7
Pardoned	0	0	1	0	0
Paroled	110	140	119	86	120
Escaped	0	2	0	0	0
In institution at end of year	1033	1053	1082	1230	1321
Total	1306	1395	1437	1529	1686
Five Year Summary
Received:
First admission	1260
Readmission	515
Total	1775
Percent of
Released:	N	admissioi
Discharged by expiration of sentence	460	25.9
Discharged by Board of Parole	412	23.2
Paroled	575	32.4
Total released Retained in institution of otherwise	1447	81.5
accounted for	328	18.4
Grand Total	1775	99.9
1 Compiled from Biennial Reports of the Minnesota State Prison for the years 1921-22; 1923-24; 1925-26; 1927-28. (Printed at the Prison).CHAPTER THREE
23
TABLE VIi
Statistics of Inmates Paroled from the Minnesota State Reformatory during the Five Year Period 1922-27, as Indicated in the Published Biennial Reports.
	1922-	1923-	1924-	1925-	1926-
	1923	1924	1925	1926	1927
On parole at beginning of year	95	83	126	123	112
Paroled during year	87	145	131	132	189
Dismissed on parole	2	7	7-	3	2
Reinstated on parole	1	0	0	1	0
Total	185	235	264	259	303
Discharged from parole:					
By expiration of sentence	2	12	10	15	16
By commutation of sentence	0	0	0	0	0
By Board of Parole Returned to institution:	75	60	106	95	81
Temporarily (in good standing)	4	6	9	7	2
For violation of parole	10	13	11	20	14
Committed to other institutions 2		6	2	4	5
Fugitives from parole	7	11	3	6	6
Died	1	1	0	0	4
On parole at end of year	84	126	123	112	175
Total	185	235	264	259	303
Five Year Summary:					
				Percent of total	
			N	number paroled	
Total number paroled in five year period			684		100.0
Discharged from parole:					
By expiration of sentence			55		8.0
By Board of Parole			417		60.0
Total discharged			472		68.0
Parole violators, fugitives, dead, or					
otherwise accounted for			212		32.0
Total			684		100.0
the present study. The discrepancy is due mostly to the fact that the latter included only the cases of men actually sent out from the institutions on parole. The official figures for “paroled” include the total number voted paroles by the Parole Board. It not infrequently happens, however, that a man who has been voted a parole can not be satisfactorily placed in employment and he may then remain in the institution. Such cases would be counted as “paroled” in the Biennial Reports but would not be included in the present study. Then, too, those released on “medical paroles” are counted in the official reports but were not included in this study. The year by year comparison of totals is indicated in Table VIII on the next page.
The preceding pages should serve to make clear the numerical setting of this study. Further description of the procedure utilized follows in the next section.
1 Compiled from the Biennial Reports of the Minnesota State Board of Parole for the years 1920-22; 1922-24; 1924-26; 1926-28.24
PREDICTION METHODS AND PAROLE
TABLE VIIl
Statistics of Inmates Paroled from the Minnesota State Prison during Five Year Period, 1922-27, as Indicated in the Published Biennial Reports.
1922- 1923- 1924- 1925- 1926-1923 1924 1925 1926 1927
On parole at beginning of year	86	90	117	95	66
Paroled during year	110	140	118	86	120
Reinstated from escape	2	1	1	1	0
Total	198	231	236	182	186
Discharged from parole:			21	19	18
By expiration of sentence	11	18			
By Board of Parole Temporarily returned	67	55	90	66	36
(good standing)	3	11	3	6	2
V iolations—returned	15	13	15	10	9
Fugitives	12	15	11	13	8
Died on parole	0	2	1	2	2
On parole at end of year	90	117	95	66	111
Total	198	231	236	182	186
Five Year Summary:					
				Percent of total	
			N	number paroled	
Total number paroled during period Discharged from parole:			574		100.0
By expiration of sentence			87		15.1
By Board of Parole			314		54.8
Total discharged			401		69.9
Parole violators, fugitives, dead, or otherwise accounted for			173		30.0
Total			574		99.9
TABLE VIII
Comparison of the Yearly Totals of Inmates Released on Parole Obtained in This Study and Those Published in the Biennial Reports of the Institutions. (Minnesota State Prison and State Reformatory, 1922-27).
Year	Prison B. R.2	P. S.3	Reformatory B. R.2 P. S.3		Pris. B. R.2	& Ref. P. S.»
July 1st-June 30th						
1922—1923	110	92	87	78	197	170
1923—1924	140	127	145	143	285	270
1924—1925	118	117	131	122	249	239
1925—1926	86	78	132	125	218	203
1926—1927	120	128	189	182	309	310
Totals	574	542	684	650	1258	1192
Amount of discrepancy 32			34		66	
	542		650		1192	
1	Compiled from the Biennial Reports of the Minnesota State Board of Parole for the years 1920-22; 1922-24; 1924-26; 1926-28.
2	B. R.—Biennial Reports.
3	P. S.—Present Study.CHAPTER THREE
25
2.—Collecting the Data
The State Parole Department keeps a book of “Monthly releases on parole” into which is pasted a typewritten copy of the list of men voted a parole from each institution at each successive meeting of the Parole Board. This record contains a separate list of names of those paroled from each institution for each month, arranged in chronological order with the most recent releases at the first of the book. The list from each institution for each month is in alphabetical order, but nowhere is there a grouping of the names of all of the men released from any one institution, or for any one year, or other convenient unit of time. Every name on these monthly lists was taken for the period from July 1st, 1922 to June 30th, 1927 and the man’s record examined. Those voted paroles but not released from the institution, or voted medical paroles, or given a conditional discharge shortly after release, were omitted from further study. This gave a net total of 542 Prison and 650 Reformatory cases that were actually released and served time on parole. The extent of the culling out process can be seen from Table VIII on the preceding page. The period 1922-27 was selected as the maximum one in which reasonably complete parole files could be found) The records become, in general, less complete and satisfactory the farther back one goes. It was impossible to make use of more recent records than June 1927 because some of the men released after that date were still on parole at the time the present investigation was made.
The contents of the ordinary case-record folder in the files of the Parole Department consist of a general mixture of many kinds of information clipped together into one record. As soon as any correspondence on the case develops involving the Parole Board a folder is started bearing the man’s name and prison or reformatory number. To keep the file in order, the material is usually fastened to a stiff-back holder by means of a two-pronged clip. Into this folder goes all material bearing on the case, clipped to the card-back in approximately the order in which it reaches the parole office. When the case first comes before the Board for a hearing a “face-sheet” is attached giving in summary form the name, number, age, county, crime, sentence, previous criminal record, marital status, occupation, physician’s report, results of mental tests (if any are available), report on prison conduct and sometimes a few other items that can conveniently be summarized. With this is a mass of correspondence (either in the original or carbons) from the prisoner himself, his friends and relatives, officials and others interested; also the letters of those persons whose names the prisoner has given as “references.” Reports of the parole agents, statements of doctors and alienists, and the like are included; also the formal “ monthly reports” of the inmate while on parole. This conglomerate mass of material is the “parole file.”
The face sheet entries can not be taken as complete or’final summaries of the record. Information dealing with the inmate’s background found on the face sheet may represent only the prison-26
PREDICTION METHODS AND PAROLE
er’s own unverified statements from the interviews at the time of admission to the institution. As further investigation and verification brings in new information this is entered in the record in the form of the original letters or reports by which it has been transmitted and the face sheet is not always changed to correspond. At subsequent hearings new face sheets are sometimes prepared, more or less completely filled out, but this is not always the case. There has been no great uniformity in this and in every case it was necessary to read the entire) record with some carle in order to arrive at a judgment as to what constituted the best and most accurate information available. The records range in size from a few pages to a hundred or more. An occasional record has several hundred pages included in the folder. The average is, perhaps, about 20 or 25 pages for each case, not counting the “monthly report” blanks.1
How reduce this chaotic mass of materials to organized and systematic information? That was a question to which considerable attention was given. The writer spent several half-days reading records more or less at random and getting familiar with the filing system and other details of present parole procedure. This “browsing” did not serve to give any clear notion of what kind of information the records actually contained, but it did help to give insight into the way in which the records were made up and the sources of the information. Six or seven records were then carefully read and rather complete notes taken on all the information that they contained. Further examination of these notes led to a listing of the major categories on which information had been found. Mimeographed copies of this list of categories (distributed in double-space or triple-space arrangement on legal size paper) were struck off and this tentative schedule was then used as a basis for the careful reading of another group of about fifty haphazardly selected cases. In this tentative list no subclasses were provided for the categories and descriptive phrases or exact information was entered directly on the mimeographed sheet. Where the categories already on the sheet did not serve, new ones were written in. The information from this group of cases was then studied and experimented with in an effort to get a satisfactory list of categories with appropriate subclasses that would accurately record in codified form the facts that had been written in on the mimeographed sheets. The list of items or factors on which information was at hand is given below:
I.	Factors Involving the Circumstances and Conditions of the Trial and Commitment
1.	County from which received
2.	Previous criminal record
3.	Number of associates in offense for which convicted
4.	Length of maximum sentence imposed
5.	Kind of plea entered
6.	Nature of the offense
l It should be noted, however, that the sheets are frequently only partially filled.CHAPTER THREE
27
7.	Statement of trial judge in regard to clemency
8.	Statement of county attorney in regard to clemency
9.	Size and type of community in which offense was committed
II.	Factors Involving Circumstances and Conditions of the Social Background
10.	Marital status of inmate
11.	Number of children, if married
12.	Home condition
13.	Number of siblings (other children in family)
14.	Education of inmate
15.	Mobility of inmate before conviction
16.	State or country where inmate was born
17.	Occupation at conviction
18,. Occupation at conviction according to the Minnesota Institute of Child Welfare classification of occupations
19.	Inmate’s work habits prior to conviction
20.	Social type.1
21.	Size and type of community in which inmate was brought up
*
III.	Factors Involving the Traits, Habits, and Characteristics of the Individual
22.	Age at conviction
23.	Mental age
24.	Intelligence Quotient
25.	Height (in inches)
26.	Weight (in pounds)
27. Whether inmate should be considered lazy or ambitious 1
28. Whether inmate should be considered honest or dishonest
29.	Whether or not inmate has record of the use of liquor
30.	Whether or not inmate uses drugs
31.	Whether or not inmate has a record of sexual immorality
IV.	Factors Associated with the Period of Stay in the Institution
32.	Prison official’s estimate of inmate’s mentality
33.	Prison official’s judgment of conduct: deportment
34.	Punishment record in the institution
35.	Months in the institution before parole
36.	Whether aid was given dependents while man was in the institution
V.	Factors Associated with Period on Parole 2
37.	Size and type of community into which paroled
38	Where living on parole in relation to where brought up
39	Where living on parole in relation to where offense was committed
40. Inter-relation between place where brought up, where offense was committed, and where placed on parole in terms of communities or neighborhoods
41	Inmate’s work record while on parole
42	Wage received on parole
43.	Number of visits to inmate by state parole agent
44.	Number of hearings before Parole Board before parole was granted
1	Categories 27 and 28 represent the present writer’s judgment of the inmate’s character as reflected in the record as a whole.
2	None of the factors associated with the period on parole were used in the construction of parole prediction tables.28
PREDICTION METHODS AND PAROLE
VI. Miscellaneous Items Checked for Statistical. Purposes
45.	Outcome on parole: non-violators, minor and major violators
46.	Institution from which paroled and year of parole
47.	Number of months on parole before discharge (for non-violators)
48.	Time interval between discharge from parole and the expiration of maximum sentence
49.	Number of months on parole when violation took place (for the violators)
The subclasses used in connection with each of these items of information need not be given here. They appear, for most of the categories, in connection with the tables and discussion of results in Chater IV.
Typewritten copies of the above schedule, including a full listing of all subclasses used with each category, were pasted on two large pieces of trunk-board which were tied together at one end much like the covers of a large loose-leaf note book. This “dummy” book could then be set up on a table or desk in front of one when reading the case record and the information entered according to its proper schedule classification. To facilitate the recording, 4x6 cards were printed (with appropriate blanks at the top for identification information) listing in order the numerals from 1 to 45. The number of the subclass in the schedules under which an item should be classified was then entered after the numeral on the card corresponding to the number of the major categories. Thus for example “age at conviction” is number 22 in the list of categories that was used, and subclass number 1 under that is “18 years or less” ; if, in a specific record, the man’s' age was given as eighteen years that fact would be recorded by writing the number one after the printed numeral 22 on the 4x6 printed card. Similar entries were made for each of the other categorias. The information thus codified and entered on the 4x6 cards was then “punched” on regular tabulating machine cards. Practically all of the actual counting was done by machine. But for the use of these mechanical sorting and counting devices, the physical labor of counting and getting the numerical distribution in each class or group would have made a study such as this practically impossible.
3.—The Plan of Analysis
With the mass of materials reduced to code and punched on tabulating cards, only relatively simple forms of anaylses were attempted. The more important efforts may be briefly indicated as follows:
1. The cards were first sorted into the three “outcome on parole” groups, namely, “non-violators”, “minor violators”, and “major violators.” These groups were then counted separately to get the distribution of cases in each subclass of each category for the Prison and Reformatory separately and for the combined group of Prison and Reformatory.CHAPTER THREE
29
2.	The percentage distribution for each subclass under each category was then computed for the institutions separately and for the combined group.
3.	The percentage of the cases violating parole in each subclass of each category was also computed for each institution separately and for the combined group.
4.	Cumulative tables for “predicting outcome on parole” were then constructed for the institutions separately and for the combined group. Several different schemes were experimented with for these tables.
5.	Two separate tests or checks on the reliability or consistency of the classifications used were attempted, namely: (a) a reclassification by the writer, at a later date, of two hundred randomly selected cases, and (b) the employment of an outside person for a further check on sixty-three of these same cases using the same schedules and classifications.
The distribution for each category and the violation rate for each subclass are presented in condensed form in the next chapter. Following this comes the discussion of reliability and the presentation of the prediction tables for outcome on parole.CHAPTER IV
Some Results: Distribution and Violation Rates For All of the Major Categories of Information
Adequate presentation of the results of a study such as this involves considerable difficulty. Space limitations make it impractical to list fully all the material that has been worked into tabular form. Yet it is desirable to give sufficient source data to enable the reader to follow the discussion intelligently and to check up on the conclusions attained. This chapter will attempt a brief survey of the distributions found for each category of information. By way of introduction to the specific materials there will be a brief discussion of the general results. Otherwise the same major divisions of subject matter used in the schedule 1 will be followed here.
1.—General Results :	Distribution of Violators and Non-
violators For Each of the Institutional Groups
The underlying purpose of the entire study has been to discover what relation there might be between the various factors involved in an inmate’s past life, upon which there is some information in the parole records, and his conduct while on parole. Three categories of classification were adopted for the statistical consideration of “outcome on parole”, namely, non-violators, major violators, and minor violators. These terms are largely self-explanatory, but a brief elaboration may help make clear the more exact meaning utilized for this study.
1. Non-violator:	Under this classification were entered all cases
of those men discharged from parole in which the record gave no indication of misconduct or of violation of the rules laid down. This definition means, of course, that some error may be introduced through considering as non-violators all those having no record of violation. There is the possibility that some men may have violated their parole agreements without the fact coming to the attention of the authorities and thus succeeded in obtaining their discharge as non-violators ■when in reality they were violators. This difficulty, however, is inherent in any use of official records and information. The only way to avoid it would be to supplement the official records with information obtained through private channels and outside sources from a separate and independent investigation. Even then, of course, the possibility of undiscovered violation would still exist, though presumably in a smaller number of cases. No funds were available for such supplementary information in this study.
1 See pp. 26-28, Chapter III.
30CHAPTER FOUR
31
2.	Major violator:	Under this head were entered all those with
a definite record of serious misconduct. By this is meant primarily the commission of a new offense (either misdemeanor or felony) or becoming a fugitive. Committing a new offense need not involve an overt, aggressive criminal act. Thus, for example, if a man has been committed under the statute provision for “failure to provide for minor child” and then fails to make his agreed payments while on parole (when earning money enough to do so), he would most likely be returned to the institution as a parole violator and he would be classified as a major violator even though his offense was merely the negative one of failing to pay. Such a case would be considered a major violation because the offense on parole is just as serious as the original crime for which convicted. Major violators, it was found, were almost without exception returned to the institution to complete their term, or to await their chances for a new parole.
3.	Minor violator:	This classification includes those who have
rather definite records. of violations, but whose misconduct has not been of any very serious or vicious nature. Violations of the technical rules of parole, such as getting married without the permission of the Parole Board, or changing residence, locality, or employment without giving immediate notification, or buying an automobile without permission, would come under this classification. So, too, would the more serious misconduct indicated by persistent drunkenness, gambling, idleness, and the like which leads the parole agent to reprimand the man but still giving-him “another chance.”
The two classes of violators thus distinguished are reasonably definite. Major violations were easy to recognize in the records from the statements and reports of the parole agents describing in some detail the misconduct and the official action taken. Minor violation was inferred from the same source. Where the agent reported to the Parole Board the details of a case of misconduct but suggested that the man be given another chance under new conditions that had been arranged, the entry would be made as “minor violator”. Minor violators wore usually not returned to the institution, though that occasionally happened as in ihe case of the boy who got married without the Board’s permission and then was “laid off” from work.
The distribution of cases under these categories of “outcome on parole” for the two institutions and for the combined group is indicated in tables IX	and X on the next page.	It	will be	noticed that
of the 542 cases paroled from the Prison 134	or	24.7%	were classed
as violators; of these 134 violators, 33 or 24.6% were classed as minor violators. Of	the 650 Reformatory cases,	126 or	19.4% were
classed as violators,	while of these latter, 23	or	18.3%	were classed
as minor violators. Why there should be almost identically the same proportions of minor violators in the group of violators as there are of violators in the total group of paroled cases is not apparent. As far as the data of this study go there is nothing to indicate a meaningful relation. In the Illinois study Burgess found nearly 40% of the total number of violations to be on minor or technical grounds. 1
1 Op. cit.y p. 219.32
PREDICTION METHODS AND PAROLE
TABLE IX
Distributions as to Outcome on Parole for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27.
Outcome on Parole	Prison N %	Reformatory N %	Pris. & Ref. N %	
Non violators Violators	408 75.3 134 24.7	524 80.6 126 19.4	932 260	78.2 21.8
Totals	542 100.0	650 100.0	1192	100.0
Of the total number paroled he found the proportion violating to be somewhat higher, namely, 22.1% for Pontiac, 28.4.% for Joliet, and 26.5%; for Menard.1
The relative proportions of major and minor violators in the total group of violators, is indicated in Table X below.
TABLE X
Distribution of Major and Minor Violations in the Total Group of Violators for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27.
Nature of	Prison		Reformatory		Pris. & Ref.	
Violation	N	%	N	%	N	%
Major Violation	101	75.4	103	81.7	204	78.5
Minor Violation	33	24.6	23	18.3	56	21.5
Totals	134	100.0	126	100.0	260	100.0
The distributions indicated above give the basis for further comparisons. The average rate of violation of parole for each institution, or for the combined group, may be taken as a convenient base or standard to which the violation rate found for any special subclass may be referred for interpretation and meaning. That is the major purpose of the following sections of this chapter.
2.—Factors Involving the Circumstances and Conditions of the Trial and Commitment
The specific factors in this group appear in Table XI on the following pages. Columns, one, three, and five give the number of men in each subclass for the three institutional groups. Columns two, four, and six give the violation rate for each subclass, that is the percentage of the total number in that class who are parole violators. Under category Number one (county from which received), for example, in the Prison there were 185 men from Hennepin county; 34.1% (63 cases) of these were parole violators. With this general scheme of tabular presentation in mind, brief comments and explanations bearing on the specific categories may be in order.
1. County from Which Received:	The classification used is self-
explanatory. Attention should be called to the low violation rates
i Ibid., p. 265.CHAPTER FOUR
33
TABLE XI
Total Number in each and Percentage Violating Parole Shown for Each Subclass of Nine Different Categories Involving Factors of the Trial and Commitment for the Minnesota State Prison, the State Reformatory and the Combined Group of Prison and Reformatory, 1922-27.
		Prison		Reformatory		Pris & Ref.	
			% who		% who		% who
Category		N	violate	N	violate	N	violate
		(i)	(2)	(3)	(4)	(5)	(6)
Average rate of violation			24.7		19.4		21.8
1.	County Received From Hennepin	185	34.1	150	30.7	355	32.5
	Ramsey	82	39.0	82	30.5	164	34.8
	St. Louis	44	15.9	55	23.6	99	20.2
	Rest of State	231	13.9	353	11.6	584	12.5
	Federal Prisoners	0	0.0	10	10.0	10	10.0
	Totals	542		650		1192	
2.	Prior Criminal Record None discovered	194	10.8	286	11.5	480	11.2
	Arrest, probation or fine 166		24.9	179	16.8	345	20.6
	One minor term served	52	30.8	87	33.3	139	32.4
	Two or more minor terms 41		58.5	77	36.4	118	44.1
	One major term One major and one	43	16.3	6	33.3	49	18.4
	minor term	37	51.4	15	26.7	5-2	44.2
	Two or more major terms 9		66.7	0	0.0	9	66.7
	Totals	542		650		1192	
3.	Number of Associates in	Offense for Which Convicted					
	None recorded	357	25.2	297	21.2	654	23.4
	One	99	24.2	192	16.1	291	18.9
	Two	59	22.0	74	21.6	133	21.8
	Three or more	27	25.3	' 87	18.4	114	20.2
	Totals	542		650		1192	
found in the distinctly rural counties in the state when compared with the two definitely urban counties. The cities of Minneapolis and Saint Paul constitute about ninety to ninety-five percent of the total population of Hennepin and Ramsey counties respectively. The population of Duluth is only about fifty percent of the total for Saint Louis county.
2. Prior Criminal Record:	By “minor term served” is meant ac-
tual incarceration in a jail, workhouse, juvenile training* school, or reform school. By “major term served” is meant actual incarceration in the state reformatory or the state prison, either in this state or elsewhere. The last subclass “two or more major terms” really means any combination of major and minor terms not included under the other subclasses. Of special interest is the apparent important relation between the occurrence of a “minor prior record” and the violation of parole. This is in general agreement with Burgess’ results though the methods of classification differ too much for direct comparison. 1
1 Burgess, E. W., op. cit.y p. 228.34	PREDICTION METHODS AND PAROLE
TABLE XI (Continued)
	Prison	Reformatory	Pris. & Ref.
	% who	% who	% who
Category	N violate	N violate	N violate
	(1) (2)	(3) (4)	(5) (6)
erage Rate of Violation		24.7		19.4		21.8
Length of Maximum Sentence		Imposed				
3 years or less	44	13.6	91	15.4	135	14.8
4-5 years	269	30.1	301	23.3	570	26.5
6-7 years	74	13.5	74	8.1	148	10.8
8-10 years	117	25.6	143	20.3	260	22.7
11-15 years	20	15.0	22	27.3	42	21.4
16-20 years	4	75.0	3	0.0	7	42.9
21-30 years	7	14.3	10	10.0	17	11.8
31-40 years	4	0.0	5	0.0	9	0.0
41 years and over	3	0.0	1	0.0	4	0.0
Totals	542		650		1192	
Plea Entered						
Guilty	405	27.2	558	20.3	963	23.2
' Not guilty	128	17.2	90	14.4	218	16.1
No information	9	28.6	2	0.0	11	18.2
Totals	542		650		1192	
Nature of the Offense						
Larceny	171	26.9	239	20.5	410	23.2
Robbery	25	32.0	32	21.9	57	26.3
Burglary	40	35.0	86	15.1	126	21.4
Fraud and Forgery	5f3	35.8	67	29.9	120	32.5
Using car w. o. permission 5		40.0	42	31.0	47	31.9
Arson	12	0.0	6	0.0	18	0.0
Desertion : wife and						
children	68	41.2	27	33.3	95	38.9
Sex offenses	113	9.7	124	9.7	237	9.7
Murder & manslaughter	10	30.0	4	25.0	14	28.6
Liquor and prohibition	8	0.0	5	0.0	13	0.0
All others	37	8.1	18	11.1	55	9.1
Totals	542		650		1192	
Statement of Trial Judge in Regard to Clemency						
In favor of clemency	70	18.6	110	12.7	180	15.0
Opposed to clemency	60	21.7	176	17.0	236	18.2
No recommendation	59	25.4	179	21.8	238	22.7
No information recorded 353		26.3	185	23.2	538	25.3
Totals	542		650		1192	
Statement of County Attorney in Regard to				Clemency		
In favor of clemency	58	10.3	78	14.1	136	12.5
Opposed to clemency	27	14.8	178	20.8	205	20.0
No recommendation	53	17.0	224	18.3	277	18.1
No information recorded 404		28.5	170	21.8	574	26.5
Totals
542
650
1192CHAPTER FOUR
35
9.
Size and Type of Community in Which Offense Was Committed
8.7	214	9.8
Farm (open country) Country town (up to 5,000) Small city (5-25000) Larger city (over 25000)
Totals
76	11.8	138
114	16.7	173
58	19.0	61
294	32.3	278
542		650
16.8	287	16.7
11.5	119	15.1
28.1	572	30.2
	1192	
3. Number of Associates in Offense for Which Convicted:	It should be noted that the number of associates in the offense for which convicted seems to bear but slight relation to the occurrence of parole violation. This is at variance with the results obtained by Burgess. He found the “lone wolf” much more likely to violate parole than the man who had one or more associates. 1
4.	Length of Maximum Sentence:	The distribution for each sub-
class is given here in order to make clear the small ¡number receiving sentences of any considerable length. The last five subclasses need to be grouped into one in order to compute a more accurate violation rate. 2
5.	Kind of Plea Entered:	No comment is necessary on the figures
for this category.
6.	Nature of the Offense:	The classification of offenses used here
is, in the main, the one appearing on the “face sheet” of the parole records. The only ambiguous class is that of “using car without permission.” Minnesota has a special statute specifying such an act a crime and a few offenders are convicted under that charge. The great majority of automobile thieves, however, are convicted of larceny. The special class therefore loses its apparent meaning.
7.	Statement of Trial Judge in Regard to Clemency:	Information
on this category was found in too few cases to attach much importance to the differences in violation rates that appear. The fact that the rate of violation of the large group of “no information recorded” is considerably higher than the average for the institutions would seem to indicate selection of some sort in the cases on which information is available.
8.	Statement of County Attorney in Regard to Clemency:	The
comment made above applies in all essentials to this category as well.
9.	Size and Type of Community in Which Offense Was Committed: The same tendency appears here which was noted under the category “county from which received”. The men whose offenses were committed in large cities have a decidedly higher rate of parole violation than those of the open country or small town.
The percentage of those who violate parole has been indicated for all subclasses of all the above categories, even where the total number in the class is too small to compute a valid proportion. This
1	Ibid., p. 223.
2	Such a grouping of the subclasses was used when this category was included among those on which prognostic tables for parole violation were constructed.36
PREDICTION METHODS AND PAROLE
has been done in order to enable the reader to infer the actual distributions of violators and non-violators obtained for each subclass. Thus, for example, under the category “prior criminal record,” the subclass “one major term”, for the Reformatory group, has a total of six cases of which 33.3% are violators. This percentage can not be taken as a valid measure of the probable future rate of violation of men in this group. All it means is that of the six cases found in this group, two were violators and four were non-violators. Not much weight can be attached to the violation rate, as such where the total number in the subclass is less than twenty-five. 1
The question of the statistical validity of the difference in the percentages of violators found may well be considered at this point. For purposes of illustration the differences in the violation rates indicated for the various subclasses of the category “county from which received” may conveniently be taken for the prison group. It appears (Table XI, category 1). that 39^0% of the Ramsey county cases violated while only 34.1% of the Hennepin county cases were violators. Is this difference statistically significant? What is the probability that other samples of randomly selected cases for these two counties would show an equally great difference; or at least a difference greater than zero?
The problem here is essentially the same as that involved in comparing the observed difference between any two means. The formula (in simplest form) given by Yule 2 for the standard error of a proportion is crp = VYV where p equals the percentage or proportion, q
Yf
equals the difference between p and 1.00, and N equals the number of cases. By using the Edgerton-Paterson3 tables, the sigma values may be read off directly. The sigma and sigma squared values thus obtained, with the range of plus and minus sigma, are indicated in Table XII on the next page.
Interpreting Table XII for Hennepin county, it may be said that the chances are 68 in 100 that the difference between the violation rate obtained in any random sample and the true violation rate for the county will not be greater than ±3.5% (.035). Or, in other words, the. chances are 68 in 100 that the true violation rate lies, between the limits 34.1 plus 3.5, and 34.1 minus 3.5, that is, between 37.6% and
1	By adopting this scheme for tabular presentation, it is possible to indicate all essential information about the distributions obtained and still preserve the advantages of parallel column comparisons for the three institutional groups. The amount of space required for tables is thus cut down by about two-thirds.
2	Yule, G. Udney, An Introduction to the Theory of Statistics, p. 257,.formula (2). London, Charles Griffin and Co., 1927 (Eighth edition). The particular form of the formula given above is that adapted by Edgerton and Paterson in their “Table of Standard Errors and Probable Errors of Percentages for Varying Numbers of Cases.”
3	Edgerton, Harold A., and Paterson, Donald G., “Table of Stand-
ard Errors and Probable Errors of Percentages for Varying Numbers of Cases.” The Journal of Applied Psychology, 10:	378-391, No. 3,
September 1926.CHAPTER FOUR
37
TABLE XII
Sigma Values and the Range of Plus and Minus Sigma for “Percent who Violate” in the Subclasses of the Category “County from Which Received” for the Minnesota State Prison, 1922-27.
		% who	S. E. of		Range of Sigma
		! violate	violation	(S. E.)1 2	plus and minus
		in each	rate. (Ex.		Expresed in
County	N	group	in deci.)	up 2	percentages
	(1)	(2)	up (3)	(4)	(5)
Hennepin	185	34.1	.035	.001225	30.6—37.6
Ramsey	82	39.0	.054	.002916	33.6—44.4
St. Louis	44	15.9	.055	.003025	10.4—21.4
Rest of State	231	13.9	.022	.000484	11.7—16.1
Total 542
30.6%.! Similar interpretations need to be applied to the other subclasses as well.
To determine whether the difference indicated between the violation rates of Hennepin and Ramsey counties is significant we need only apply the usual technique for getting the reliability of the difference between two means. The formula 2 for the standard error of the difference is <*diff.= V^Pj2 + <Tp22. Substituting the sigma squared values found from the Edgerton-Paterson table (column four Table XII) the following results are obtained:	adiff = V .001225 + .002916= V .004141 = .064.
The actual difference in violation rates for the two counties (expressed in decimals) is .049 whereas the standard error of the difference is .064. Since the actual difference obtained is smaller than the standard error of that difference, it would usually be dismissed as statistically insignificant. To constitute reasonable certainty a difference must be at least three times the standard error of the difference. We may say of the above that the chances are 68 in 100 that the obtained difference of .049 does not diverge from the true difference by more than plus or minus .064; also that the chances are 77 in 100 that the obtained difference represents a true difference greater than zero. 3
The statistical validity of all the differences for the various subclasses in Table XII are given in, Table XIII on the next page.
These results may be interpreted to mean that, if an infinite number of samples were taken from Hennepin and Ramsey counties, the violation rates for the latter would really be higher than for the former (difference greater than zero) in 77% of the samples; in 23% of the samples there might be no difference, or a difference in reverse direction. In the comparison of St. Louis county with the rest of the state, of an infinite number of samples a real difference (greater
1	This interpretation involves the usual assumptions with reference to the conditions of random sampling. See Yule, op. cit,, p. 259-262.
2	Edgerton, H. A., and Paterson, D. G., op. cit., p. 390. This form of the formula is an adaptation of that given by Yule, G. U., op. cit., p. 269 (formula 6).
3	This is following Garrett, H. E., Statistics in Psychology and Education, pp. 128-134. New York, Longman’s Green & Co., 1926.38
PREDICTION METHODS AND PAROLE
TABLE XIII
Standard Errors of the Differences in Violation Rates for All Subclasses Under the Category “County from Which Received'' With Probability Stated in Terms of “Chances in 100 that the true difference is greater than zero.” (Minnesota State Prison, 1922-1927).
Obtained difference (in decimals) D Subclasses compared (1)		°diff (2)	D ^diff. (3)	Chances in 1 2 100 (4)
Hennepin-Ramsey	.049	.064	0.767	77
Hennepin-St. Louis	.182	.065	2.800	100 (99.74)
Hennepin-Rest of State	.202	.041	4.926	100
Ramsey-St. Louis	.231	.076	3.039	100
Ramsey-Rest of State	.251	.058	4.328	100
St. Louis-Rest of State	.020	.058	0.344	64
than zero) would be found	in 64%	of the samples;		in 36% of the
samples there would be no	difference, or a		difference	» in reverse di-
rection. In the case of the other subclasses compared, an infinite num-
ber of samples would give real differences in 100 %i of the cases. We may say, therefore, that the differences indicated between violation rates for all the subclasses of the category “county from which received” are (for the Prison group) statistically significant except the differences between the rates for Hennepin and Ramsey counties and for St. Louis county and the rest of the state. Even in these the probabilities of a real difference being'expressed are considerably better than chance, which, in this case, is expressed as “50 chances in 100.” 2 '
1	The values for “chance in 100” in column four have been taken from Garrett, H. E., Statistics in Psychology and Education, Table XIV, p. 134.
2	No computation of the standard errors of the differences for all the categories has been attempted, due to the .great amount of time and labor involved. The method, however, for such computation has been indicated and the results are sufficient to give some idea of its probable effects if generally applied. If the reader will reexamine Table XI and turn ahead to Tables XIV, XV, XVI and XVII he will observe, on the one hand that relatively few of the differences obtained are as great, for example, as that between Ramsey and St. Louis counties, while, on the other hand, differences comparable in magnitude to those between St. Louis county and the rest of the state, or between Ramsey and Hennepin counties, are common. In other words, it appears that relatively few of the differences found are of definite and unquestioned statistical validity, but it is also apparent that most of them are greater than could be expected by chance.
As a practical aid or guide in viewing the probable reliability of the differences in rates, the following suggestion is offered for whatever it may be worth. Where the subclasses have a reasonably large number of cases, say fifty or more, a difference of 2 % in violation rates is a little better than a chance difference; a difference of 5% is probably about 50% better than chance difference (i. e., the “chances in 100” are about seventy-five); a difference of 15% indicates a reasonable certainty that there is a true difference greater than zero. Of course, no hard and fast rule of thumb can be applied here; the suggestion must be taken only as a very crude and general guide.CHAPTER FOUR
39
3.—Factors Involving Circumstances and Conditions of the Inmate’s Social Background
This group includes categories 10-21 inclusive. To conserve space, comment will be withheld except in cases where the essential meaning may be obscured unless a further word of explanation is added. The results appear in Table XIV, occupying this and three succeeding pages. The same plan of numbering the categories consecutively, used on pages 26-28 of this study, will be followed here. The general plan and meaning of the table is the same as for Table XI, pages 33-35.
TABLE XIV
Total Number in Each and Percentage Violating Parole Shown For Each Subclass of Twelve Different Categories Dealing with the Inmate’s Social Background, for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27.
Prison Reformatory Pris. & Ref. % who	% who	% who
Category	N violate (1) (2)		N (3)	violate (4)	N (5)	violate . (6) ‘
Average rate of violation		24.7		19.4		21.8
10. Inmate’s Marital Status						
Single	139	26.6	496	18.1	635	18.1
Married; lives with						
wife	221	20.4	83	14.5	304	18.7
Separated	63	27.0	21	23.8	84	26.2
Divorced	87	32.2	48	37.5	135	34.1
Widowed	32	21.9	2	50.0	34	23.5
Totals	542		650		1192	
11. No. of Children, if married						
None	108	27.8	40	19.5	148	25.5
1-2	143	25.9	101	23.8	244	25.0
3-4	100	21.0	13	30.8	113	22.1
5-6	34	17.6	* 0	0.0	34	17.6
7 or more	18	11.2	0	0.0	18	11.2
Total No. married	403		154		557	
No. single	139		496		635	
Totals	542		650		1192	
12. No. of Siblings						
None	39	25.6	26	11.5	65	20.0
One	67	30.0	38	31.6	105	30.5
Two	82	30.5	60	16.7	142	24.6
Three	54	22.2	39	23.1	93	22.6
Four	78	21.8	30	16.7	108	20.4
Five	43	25.6	21	23.8	64	25.0
Six	42	23.8	15	26.7	57	24.6
Seven	26	15.4	23	39.1	49	26.5
Eight	26	23.1	9	22.2	35	22.9
Nine or more	28	7.1	20	20.0	48	12.5
No information	57	29.8	369	17.1	426	18.8
Totals	542		650		1192	40
PREDICTION METHODS AND PAROLE
TABLE XIV (continued)
		Prison		Reformatory		Pris. & Ref.	
			% who		% who		% who
Category		N violate		N	violate	N	violate
		(i)	(2)	(3)	(4)	(5)	(6)
Average rate of violation			24.7		19.4		21.8
13.	Home Condition						
	Parents living, together Parents living,	133	23.3	356	19.7	489	20.7
	separated	21	45.8	28	14.3	49	24.5
	Parents living, divorced 14		28.6	38	26.3	52	26.9
	Father dead	115	26.9	99	17.2	214	22.4
	Mother dead	71	27.4	45	24.4	116	25.0
	Both parents dead	150	23.3	23	21.7	173	23.1
	No information	38	18.4	61	14.8	99	16.2
	Totals	542		650		1192	
14.	Education of, Inmate Unable to read or write	25	12.0	4	0.0	29	10.3
	0-3 grade	19	26.3	59	13.6	, 78	16.7
	4-6 grade	99	26.3	180	18.9	279	21.5
	7-8 grade	215	28.8	326	20.2	541	23.7
	Some H. S., not finished 43		30.2	53	18.9	96	23.9
	High School graduate	16	18.7	11	27.3	27	22.2
	Some col., not finished	12	25.0	8	12.5	20	20.0
	College graduate	4	0.0	0	0.0	4	0.0
	Read and write, No other information	105	16.2	9	44.5	114	18.4
	No information	4	50.0	0	0.0	4	50.0
	Totals	542		650		1192	
15.	Mobility before Conviction						
	N. B.1 Always in same						
	community	158	25.3	307	17.9	465	20.4
	N. B. Lived several places in or near state	182	21.4	101	15.8	283	19.4
	N. B. Some travel, “hum-						
	med” one or two years	91	34.1	199	24.6	290	27.6
	N. B. Transient, no fixed residence	39	38.5	26	23.1	65	32.3
	F. B.2 Always in same community	46	10.9	7	0.0 V	53	9.4
	F. B. Lived several places in or near state	13	23.1	5	0.0	18	16.7
	F. B. Some travel	7	0.0	2	0.0	9	0.0
	F. B. Transient	4	25.0	2	0.0	6	16.7
	No information	2	0.0	1	0.0	3	0.0
	Totals	542		650		1192	
1	N. B.—Native Born.
2	F. B.—Foreign Born Adult: a man entering the U. S. as an infant or dependent minor of less than 12 years is included with the native born in this check of mobility.CHAPTER FOUR
41
TABLE XIV (continued)
Prison % who Category N violate (1) (2)			Reformatory % who N violate (3) (4)		Pris. N (5)	& Ref. c/o who violate (6)
Average rate of violation		24.7		19.4		21.8
16.	State or County Where Born					
	In Minnesota 227	29.5	466	18.2	693	21.9
	In adjoining states 112	17.0	114	21.1	226	19.0
	Elsewhere in U. S. 71	33.8	23	21.7	94	30.9
	In Canada 1 10	30.0	5	20.0	15	26.7
	Other foreign-born 1 117	17.1	39	25.6	156	19.2
	No information 5	25.0	3	33.3	8	25.0
	Totals 542		650		1192	
17	Occupation at Conviction 2					
	Professional 5	0.0	1	0.0	6	0.0
	Student 0	0.0	10	10.0	10	10.0
	Business 41	7.3	9	11.2	50	8.0
	Clerks and bookkeepers 45	20.0	52	19.2	97	19.6
	Farmers 35	2.9	8	12.5	43	4.7
	Skilled and semi-skilled 165	38.8	140	20.7	305	30.5
	Farm hands 100	15.0	188	9.6	288	11.5
	General unskilled labor 151	27.8	242	27.4	393	27.6
	Totals 542		650		1192	
18.	Minnesota Institute of Child Welfare Classification				of	
	Occupation at Conviction 2					
	Group I (Professional) 5	0.0	0	0.0	5	0.0
	Group II (Proprietors) 37	8.1	9	22.3	46	10.9
	Group III (Upper trades					
	and specialities) 41	7.7	12	0.0	53	7.5
	Group IV (Trades and					
	and skilled workmen) 71	21.1	65	24.6	136	22.8
	Group V (Semi-skilled) 138	39.9	126	19.8	264	30.3
	Group VI (Laborers) 250	22.8	425	19.3	675	20.6
	No information 0	0.0	13	7.7	13	7.7
	Totals 542		650		1192	
19.	Previous Work Habits					
	Loafer (no work record) 98	40.0	203	26.6	301	32.6
	Odd jobs only 94	29.8	147	23.8	241	26.1
	Irregularly employed 189	25.9	179	13.4	368	19.8
	Regularly employed 158	8.2	117	9.4	275	8.7
	No information 3	0.0	4	50.0	7	28.6
	Totals 542		650		1192	
1 The foreign-born enumeration here is of place of birth in so far as possible. It differs from the F. B. designation used for category “Mobility” and the total numbers listed as “foreign born” are not comparable for the two categories.
■2 For more detailed account of the sub-classes used, turn to the discussion of reliability in Chapter V.42	PREDICTION METHODS AND PAROLE
TABLE XIV (continued)
Category	Prison % who N violate (1) (2)		Reformatory % who N violate (3) (4)		Pris. N (5)	& Ref. % who violate (6)
Average rate of violation		24.7		19.4		21.8
20. Social Type 1						
Responsible and						
substantial citizen	46	4.3	2	50.0	48	6.2
Respected workman	175	13.7	80	8.7	255	12.2
Irresponsible youngblood 12		8.3	81	13.6	93	12.9
Recent immigrant	10	20.0	4	0.0	14	14.3
Weak character	42	38.1	326	19.6	368	21.7
Ne’er-do-well	171	33.3	60	21.7	231	30.3
Transient (hobo)	43	37.2	26	23.1	69	31.9
Small town or country						
“bully”	22	27.3	34	23.5	56	25.0
“City tough”	18	55.6	34	47.1	52	50.0
No information	3	0.0	3	0.0	6	0.0
Totals	542		650		1192	
21 Size and Type of Community		in which Brought up				
Farm (open country)	81	14.8	153	9.8	234	11.5
Country town						
(up to 5000)	99	20.2	169	17.2	268	18.3
Small city (5-25000)	21	23.8	46	19.6	67	20.9
Larger city						
(over 25000)	130	36.9	233	28.3	363	31.4
Outside State,						
no other information	204	24.0	44	13.6	248	22.2
No information	7	0.0	5	20.0	12	8.3
Totals	542		650		1192	
Category Number 12 “Number of Siblings” really means the number of other children in the family and includes step-brothers and stepsisters if they have lived in the home with the inmate.
Category Number 15 “Mobility before Conviction” attempts to show something of the extent of moving about with consequent breaks in personal relations that the inmate may have undergone. The subclass “Always in same community” has reference to the man who has always been identified with one neighborhood or community. “Lived in several places in or near state” has reference to the class of individuals who have moved about some and have consequently been identified with several communities. For example, a man may have been born and raised on a farm in southwestern Minnesota, worked for three or four years as a young man in the packing houses in Sioux City, married and settled on a farm near Mason City, and then after some years moved to a farm in northern Minnesota where he comes into conflict with the law. Such a case would be classed as having
1 For a more detailed account of the subclasses used, turn to the discussion of reliability in Chapter V.CHAPTER FOUR
43
“Lived several places in or near state”. Under the subclass “Some travel, ‘bummed’ one or more years” were entered those individuals who have “knocked about” somewhat but still have some particular town or community which they return to (perhaps where they were born, or where their parents or other relatives live), and which they think of as “home”. This differs from the group “Lived several places. . .” in that it represents much more of an unstable “floater” type, being, however, less unsettled than the “Transient, no fixed residence” group. This latter group includes the drifting “homeless” man who is continually migrating from place to place as a hobo, tramp, or casual worker. In the case of the foreign born, it would seem that the move from the old country to America could hardly be considered as “mobility” for the individual unless he already had established definite contacts at home before coming here. It was arbitrarily decided to take the age 12 years as the time when such a move would take on real significance for the individual.
Categories 17, 18, 19, and 20, dealing with occupational classifications, previous work habits, and social type, will be discussed at some length in Chapter V and need not therefore be further elaborated here.
The implications of the distributions and differences in violation rates indicated for the categories in Table XIV can not be entered into at length. For the most part, they are in accord with what one might expect. Yet, there are some exceptions. Thus, illiteracy, or near illiteracy does not seem to be particularly indicative of parole violation, nor does a high school education appear to decrease violation. In the prison it would even appear that the illiterates were among the best parole “risks.” The matter of previous yvork habits shows, perhaps, the most definite relation to parole violation. Those who are classed as “Loafers” or “Odd jobs only” show a much higher rate of violation than those in the “Regularly employed” group.
4.—Factors Involving the Traits, Habits and Characteristics of the Individual
The categories and subclasses in this group are largely self-explanatory. The results appear in Table XV on the next three pages.
In the matter of age it should be recalled that only men below 30 are confined at St. Cloud. It appears that the younger men are somewhat more prone to violate parole than the older ones. This is true for both the prison and the reformatory.
Mental age and Intelligence Quotient figures were found in only about 16% of the cases at the prison. Little importance can be attached to the resulting violation rates indicated. In the Reformatory group, however, the information is fairly complete and the results must be taken into account. It should be noticed that, regardless of whether intelligence is stated directly as “mental age” or as a ratio with chronological age (Intelligence Quotient), the 'trend in violation rates is consistent. The two lowest classes, (in the Reformatory) under both devices of representation, show lower violation44
PREDICTION METHODS AND PAROLE
TABLE XV
Total Number in Each and Percentage Violating Parole Shown for Each Subclass of Ten Different Categories Dealing with the Traits, Habits, and Characteristics of the Individual for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27.
Category	Prison % who N violate (1) (2)		Reformatory % who N violate (3) (4)		Pris. & Ref. % who N violate (5) (6)	
Average rate of violation		24.7		19.4		21.8
22. Age at Conviction						
18 years and less	2	50.0	57	28.1	59	28.8
19-22 years	33	36.4	297	20.5	330	22.1
23-26 years	50	32.0	190	16.8	240	20.0
27-30 years	74	33.8	99	14.1	173	22.5
31-34 years	101	22.8	7	42.9	108	24.1
35-38 years	86	23.3			86	23.3
39-42 years	86	23.3			86	23.3
43-46 years	40	17.5			40	17.5
47-50 years	29	6.9			29	6.9
51-54 years	21	28.5			21	28.5
55-58 years	6	0.0			6	0.0
59 years and over	14	14.3			14	14.3
Totals	542		650		1192	
23. Mental Age (on Tvsts)						
8 years and below	7	57.1	22	13.6	29	24.1
9 years	10	30.0	37	8.1	47	12.8
10 years	22	22.7	68	19.1	90	20.0
11 years	18	33.3	54	20.4	72	23.6
12 years	13	53.8	116	22.4	129	25.6
13 years	5	40.0	102	17.6	107	18.7
14 years	4	25.0	90	20.0	94	20.2
15 years	4	25.0	41	29.3	45	28.9
16 years	1	0.0	24	25.0	25	24.0
17 years	3	0.0	12	0.0	15	0.0
18 years and over	4	0.0	65	18.5	69	17.4
No information	451	23.3	19	21.1	470	23.2
Totals	542		650		1192	
24. Intelligence Quotient						
0-64	14	42.9	47	10.6	61	18.0
65-74	30	23.3	105	18.1	135	19.3
75-84	23	39.1	103	21.4	126	24.6
85-94	12	33.3	132	22.0	144	22.9
95-104	5	40.0	78	23.1	83	24.1
105-114	3	0.0	31	12.9	34	11.8
115-124	1	0.0	25	16.0	26	15.4
125-134	2	0.0	25	12.0	27	11.1
135 and over	1	0.0	28	21.4	29	20.7
No information	451	23.3	76	21.1	527	23.2
Totals	542		650		1192	
rates on parole than do the intermediate groups which are less distinctly “sub-normal” in intelligence. Equally noetworthy is the fact that those of high intelligence, that is, distinctly above the average, are not especially inclined to be non-violators. In other words, theCHAPTER FOUR
45
TABLE XV (continued)
Prison Reformatory Pris. & Ref. % who	%who	% who
Category		N violate (1) (2)		N violate (3) (4)		N (5)	violate (6)
Average rate of violation			24.7		19.4		21.8
25.	Height (in inches)						
	63 and below	30	26.7	39	15.4	69	20.3
	64-65	61	29.5	115	18.3	176	22.2
	66-67	101	27.7	187	21.4	288	23.6
	68-69	77	22.1	163	20.2	240	20.8
	70-71	66	24.2	116	19.0	182	20.9
	72 and over	15	20.0	13	23.1	28	21.4
	No information	192	22.9	17	5.9	209	21.5
	Totals	542		650		1192	
26.	Weight (in pounds)						
	119 pounds or less	5	40.0	11	18.2	16	25.0
	120-139 pounds	43	23.3	139	20.1	182	20.9
	140-159 pounds	138	26.8	280	17.5	418	20.6
	160-179 pounds	104	23.1	163	24.5	267	24.0
	180-199 pounds	34	32.4	34	17.6	68	25.0
	200-219 pounds	9	11.2	6	0.0	15	6.7
	220 pounds and over	6	0.0	1	0.0	7	0.0
	No information	203	24.1	16	6.3	219	22.8
	Totals	542		650		1192	
27.	Character Traits						
	Lazy, shiftless	231	34.2	410	25.4	641	28.5
	Ambitious, worker	296	16.9	238	9.2	534	13.5
	No information	15	33.3	2	0.0	17	29.4
	Totals	542		650		1192	
28.	Character Traits						
	Dishonest, reckless	263	33.5	439	24.1	702	27.6
	Honest, reliable	264	15.5	209	9.5	473	12.9
	No information	15	33.3	2	0.0	17	29.4
	Totals	542		650		1192	
29.	Habits: Use of Liquor						
	Uses liquor	427	28.1	517	21.9	944	24.7
	No record of liquor	100	9.0	131	9.9	231	9.5
	No information	15	33.3	2	0.0	17	29.4
	Totals	542		650		1192	
30.	Habits: Use of Drugs						
	Drug addict	23	56.5	7	57.1	30	56.7
	No record of drugs	504	23.0	641	19.0	1145	20.8
	No information	15	33.3	2	0.0	17	29.4
	Totals	542		650		1192	46
PREDICTION METHODS AND PAROLE
TABLE XV (continued)
	Prison		Reformatory		Pris. & Ref.	
		% who		% who		% who
Category	N	violate	N	violate	N	violate
	(1)	(2)	(S)	(4)	(5)	(6)
Average rate of violation 31. Sex Habits		24.7		19.4		21.8
Apparently immoral No record of	308	27.6	469	21.7	777	24.1
immorality	219	20.1	179	13.4	398	17.1
No information	15	33.3	2	0.0	17	29.4
Totals	542		650		1192	
rate of parole violation is about the same whether a man' is of low or of high intelligence. No such distinct differences in violation rates appear for the extremes of the distribution on the mental scale as were found for the extremes in the groupings under “Previous work habits”; This would seem to indicate that intelligence, as measured by tests, bears bnly a relatively unimportant relation to conduct on parole.
Comment is unnecessary on the differences in violation rates for the categories dealing with character traits and habits. They are quite in keeping with what one would expect.
5.—Factors Associated with the Period of Stay in the Institution
The factors in this group are indicated in Table XVI on the next page. Important differences in violation rates appear in the table, but for the most part categories are sufficiently clear to call for, little or no comment.
The fallacy of estimates of mentality appear rather strikingly if one takes the trouble to compare the distributions here with those obtained from the use of standardized intelligence tests. The suspicion remains that one of the chief items determining the “estimate” made of intelligence was conduct in the institution. If the inmate violated prison rules and got himself into trouble thereby, he was probably therefore considered of below average intelligence. That would account for the relation that appears between the estimate of mentality and conduct on parole. Since the two categories dealing with conduct in the institution show a considerable relation with outcome on parole, a measure of intelligence based on this same fact would also show such a relation.
The length of time spent in the institution bears no very significant relation to violation of parole. If any, it would appear that the longer a man was confined the more likely would he be to violate. The proposition is not, however, so simple. It may be true that the more desperate and undesirable type of man tends to be confined longer when in the institution and when released he might also be more likely to violate. The information in these tables does not permit a satisfactory answer to that question.CHAPTER FOUR
47
TABLE XVI
Total Number in Each and Percentage Violating Parole Shown for Each Subclass of Five Different Categories Dealing with Factors Associated with the Stay in the Institution, for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27.
Prison Reformatory Pris. & Ref. % who	% who	% who
Category		N violate (1) (2)		N violate (3) (4)		N (5)	violate (6)
Average rate of violation			24.7		19.4		21.8
32.	Prison Official's Estimate of Inmate's Mentality					•	
	Above average	9	22.2	10	20.0	19	21.1
	Average	222	27.0	328	19.8	550	22.7
	Below average	38	15.8	310	18.7	348	18.4
	No estimate recorded	273	24.2	2	50.0	275	24.4
	Totals	542		650		1192	
33.	Prison Official's Judgment of Prison Conduct: Deportment						
	Good	249	19.7	90	17.8	339	19.2
	Fair or( average	251	27.9	521	18.8	772	21.6
	Poor or unsatisfactory	34	41.2	39	30.8	73	35.6
	No judgment recorded	8	12.5	0	0.0	8	12.5
	Totals	542		650		1192	
34.	Punishment Record in .	Institution					
	None	151	13.9	184	9.2	335	11.3
	0-3 marks	251	25.1	274	16.4	525	20.6
	4-7 marks	72	36.1	113	29.2	185	31.9
	8 or more marks	18	44.4	62	38.7	80	40.0
	Loss of grade	6	50.0	10	40.0	16	43.7
	Solitary confinement	40	30.0	6	50.0	46	32.6
	No information	4	25.0	1	0.0	5	20.0
	Totals	542		650		1192	
35.	Months in Prison Before Paroled						
	16 months or less	122	27.0	99	14.1	221	21.3
	17-24 months	147	16.3	349	19.8	496	18.7
	25-32 months	133	30.8	121	21.5	254	26.4
	33-40 months	77	23.4	53	17.0	130	20.8
	41-48 months	33	36.4	22	27.3	55	32.7
	49 months and over	30	20.0	6	33.3	36	22.3
	Totals	542		650		1192	
36.	Whether Aid was Given Dependents			While Man		Was	
	in the Institution						
	No dependents	201	27.9	475	18.3	676	21.2
	No record of aid given	172	21.5	88	19.3	260	20.8
	Record of county aid only 9		44.5	6	16.7	15	33.3
	Record of state aid only 31		19.4	24	25.0	55	21.8
	Both county and.						
	state aid	123	24.4	55	27.3	178	25.3
	No information	6	16.7	2	0.0	8	12.5
Totals
542
650
119248
PREDICTION METHODS AND PAROLE
6.—Factors Associated with the Period on Parole
Table XVII gives the results for this group of factors. Of special interest are the categories dealing with the place of parole in relation to the place where the offense was committed, or the place where the inmate was brought up. It appears that men who are paroled outside of the community in which they were brought up are considerably less likely to violate parole than those paroled into the home community. Categories 38, 39, and 40 show uniformly consistent results of this kind.
As one might expect, there is a pronounced relationship between an /unsatisfactory work record on parole and violation. The category itself, however, has little meaning. The records contain too little information bearing directly on this question to permit accurate classification. Only in cases of flagrant discontent does correspondence indicating that fact develop and become a part of thei record. This no doubt accounts for the large proportion of the cases entered as “steady, satisfied” under category Number 41.
TABLE XVII
Total Number for Each and Percentage Violating Parole Shown in Each Subclass of Nine Different Categories Dealing with Factors Involved in the Period on Parole for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27.
		Prison		Reformatory		Pris. & Ref.	
			% who		% who		% who
Category		N ■	violate	N	violate	N	violate
		(1)	(2)	(3)	(4)	(5)	(6)
Average Rate of Violation			24.7		19.4		21.8
37.	Size and Type of Community Into Which Inmate was Paroled						
	Farm or open country 112		17.0	177	10.2	289	12.8
	Country town (up to 5,000)	87	18.4	152	13.8	239	15.5
	Small city (5-25,000) Larger city	. 48	22.9	52	21.2	100	22.0
	(over 25,000)	290	30.3	262	28.2	552	29.3
	No information	5	0.0	7	28.5	12	16.7
	Totals	542		650		1192	
38.	Where Living on Parole in		Relation	to Where Brought up			
	In same neighborhood or community	208	30.3	447	20.8	655	23.8
	In adjoining community 129		19.4	136	14.7	265	17.0
	In distant community	193	23.3	60	18.3	253	22.1
	No information	12	8.3	7	28.5	19	15.7
	Totals	542		650		1192	CHAPTER FOUR
49
TABLE XVII (continued)
	Prison	Reformatory	Pris. & Ref.
	% who	% who	% who
Category	N violate	N violate	N violate
	(1) (2)	(3) (4)	(5) (6)
Average Rate of Violation		24.7		19.4		21.8
39. Where Living on Parole in		Relation to Where				
Offense was Committed						
In same community	370	27.0	474	19.8	844	23.0
In adjoining						
community	156	21.8	155	18.1	311	19.9
In distant community	10	0.0	14	14.3	19	15.7
No information	6	0.0	7	28.5	13	15.4
Totals	542		650		1192	
40. Inter-relation of Place Where Brought Up, Where Offense was Committed, and Where placed on Parole in Terms of Communities or Neighborhoods
All in the same
community Brought up and offense in same community;	195	28.7	417	19.9	612	22.7
parole different	33	15.2	73	15.1	106	15.1
Brought up and parole in same community; of-						
fense different	11	45.5	41	24.4	52	28.8
Offense and parole in same community; brought up different	179	24.6	50	22.0	229	24.0
All in different communities	118	20.3	62	14.5	180	18.3
No information	6	0.0	7	28.5	13	15.4
Totals	542		650		1192	
Work Record on Parole						
Steady, satisfied Dissatisfied, many	488	21.7	600	16.8	1088	19.0
changes	21	57.1	30	50.0	51	52.9
Frequently idle, refuses work on some pretext	8	87.5	9	77.8	17	82.4
No information	25	36.0	11	27.2	36	33.3
Totals	542		650		1192	
Wage Received on Parole						
Less than $40 per month	103	18.4	233	18.5	336	18.5
$40-$84 per month	196	28.6	254	22.4	450	25.1
$85-$129 per month	146	24.5	120	17.5	266	21.4
$130 and over No entry	37	18.9	8	0.0	45	15.6
((works own farm etc.)	» 60	26.6	35	14.3	95	22.1
Totals
542
650
1192PREDICTION METHODS AND PAROLE
m
TABLE XVII (Continued)
Category		Prison % who N violate (1) (2)		Reformatory % who N violate (3) (4)		Pris. & Ref. % who N violate (5) (6)	
Average Rate of Violation			24.7		19.4		21.8
43.	Number of Visits to Parolee by State Parole Agent						
	None recorded	268	26.1	299	19.7	567	22.8
	One visit	123	17.1	165	13.3	288	14.9
	Two visits	83	28.9	101	17.9	184	25.0
	Three visits	50	28.0	46	28.3	96	28.1
	Four or more	18	27.7	39	25.6	57	26.3
	Totals	542		650		1192	
44.	At Which Hearing Before the Board of Parole						
	the Parole was Granted						
	First hearing	76	26.3	41	9.8	117	20.5
	Second hearing	124	22.6	207	19.3	331	20.5
	Third hearing	115	21.7	185	20.5	300	21.0
	Fourth hearing	91	28.6	124	20.2	215	23.7
	Fifth hearing	74	29.7	50	20.0	124	25.8
	Sixth or more	57	21.1	41	19.5	98	20.4
	No information	5	20.0	2	50.0	7	28.5
	Totals	542		650		1192	
Category Number 42, “Wage Received on Parole” is included here only as an interesting side-light on the parole situation. The low wages indicated reflect a common situation though they can not be taken at face value. As the figures stand, no allowance has been made for items such as board and room which are sometimes given in addition to the wage indicated. Information was not sufficiently complete to determine with any degree of accuracy the extent to which such additional accommodations were granted. All that the figures represent are rough averages of the monthly income as reported on the “monthly report blanks”.
7.-—Miscellaneous Factors of Interest in Connection with Parole
Information was secured on the following four factors of general interest in connection with parole, which are not, however, particularly related to the question of violation or non-violation: (a) the amount of time by which the maximum sentence was shortened for those discharged from parole; (b) the length of time served oh parole; (c) length of time on parole before violation took place; (d) the number discharged from parole who developed subsequent criminal records. These are taken up in order in the discussion that follows.CHAPTER FOUR
51
(a)—Time by which Maximum Sentence was Shortened for those Discharged from Parole
This item is of special interest in connection with the charge that parole is a form of “jail delivery” on a wholesale scale,—that is to say, a device whereby large numbers succeed in concluding their sentences in much shorter time than they otherwise would. The distributions for the institutional groups are indicated in Table XVIII.
TABLE XVIII
Time Interval Between Discharge from Parole and Expiration of Maximum Sentence for the Group of Non-violators in the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27.
Time	Prison		Reformatory		Pris. & Ref.	
	N	%	N	%	N-	%
Interval						
None (on expiration)	103	25.2	71	13.5	174	18.7
Less than 1 year	80	19.6	37	7.1	117	12.6
1 year	76	18.6	40	7.6	116	12.4
2 years	52	12.7	116	22.1	168	18.0
3 years	37	9.1	64	12.2	101	10.8
4-6 years	38	9.3	87	16.6	125	13.4
7-9 years	10	2.5	79	15.1	89	9.5
10 years or more	12	2.9	30	5.7	42	4.5
Totals	. 408	99.9	524	99.9	932	99.9
It will be seen that for the combined group of Prison and Reformatory nearly one-half (43.7%) are among those who have their maximum sentence shortened by less than two years. Figures for comparison with those discharged without being placed on parole are not available but it is doubtful if they serve any larger proportion of their full term.
Attention should be called to the discrepancy between the above figures for discharge from parole by “expiration of sentence” and those from the published Biennial Reports given on pages 23 and 24 of Chapter III. The figures for this study show 13.5% of the Reformatory group and 25.2% of the Prison men discharged from parole by expiration of sentence as compared with 8.0% and 15.1% respectively for the published reports. The results for the present study represent a follow through until discharge of all the cases paroled during the five year period of study. The figures from the Biennial Reports represent those so discharged from year to year without reference to when they were placed on parole. The result is, of, course, that the two groups are not entirely comparable. This fact probably accounts for the difference though it is surprising that it should be as large as it is.52
PREDICTION METHODS AND PAROLE
(b)-—Length of Time Served on Parole Table XIX gives the results for this item.
TABLE XIX
Number of Months Served on Parole by the Group of Nonviolators in the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27.
	Prison		Reformatory		Pris	& Ref.
Months on parole before discharge	N	%	N	%	N	%
4 months and less	15	3.7	16	3.1	31	3.3
5- 7 months	34	8.3	30	5.7	64	6.9
8-10 months	22	5.4	14	2.7	36	3.9
11-13 months	283	69.4	380	72.5	663	71.1
14-16 months	20	4.9	48	9.2	68	7.3
17-25 months	21	5.1	32	6.1	53	5.7
26 months and over	13	3.2	4	0.8	17	1.8
Totals	408-	100.0	524	100.1	932	100.0
These figures simply reflect in concrete form the accepted policy of the Parole Board to make the period on parole about one year. This policy rests on the assumption that if a man is going to violate parole he is likely to do so within a year; that if he behaves himself for a year it may be presumed that he will continue to “go straight” when discharged from parole. The result^ in the following section throw further light on this question.
(c)—Length of Time on Parole Before Violation The general distributions are summarized in Table XX.
TABLE XX
Number of Months Served on Parole before Violation in the Group of Parole Violators for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1922-27.
Prison Reformatory Pris. & Ref.
Months on parole
before violation	N	%	N	%	N	%
1-3 months	26	21.0	13	11.2	-39	16.3
4-6 months	45	36.3	30	25.9	75	31.3
7-9 months	23	18.5	29	25,0	52	21.7
10 or more months	30	24.2	44	37.9	74	30.8
Totals	124	100.0	116	100.0	240	100.1
Month of violation not noted	10		10		20	
Total violators	134		126		260	CHAPTER FOUR
53
It is apparent from this that violations occur rather uniformly throughout the entire period on parole. Certainly there is no tendency for the violations to reach a high peak of frequency early in the parole period and then drop off to insignificant figures for the later months. This is even more strikingly seen if the monthly distributions with cumulative percentages are taken, as in table XXI.
TABLE XXI
Number of Months Served on Parole before Violation Indicated by Monthly Periods and With Cumulative Percentage Figures for the Combined Group of Minnesota State Prison and State Reformatory, 1922-27.
Prison and Reformatory
Months on parole	Cumulative
before violation	N Distribution Percentages
1 month	12	5.0	5.0
2 months	17	7.1	12.1
3 months	10	4.2	16.3
4 months	25	10.4	26.7
5 months	27	11.3	38.0
6 months	23	9.6	47.6
7 months	14	5.8	53.4
8 months	16	6.7	60.1
9 months	22	9.2	69.3
10 months	20	8.3	77.6
11 months	14	5.8	83.4
12 or more months	40	16.7	100.1
Totals	240	100.1	
Month of violation not	noted 20		
Total violators	260		
The fact that just as large a percentage violate during the tenth and eleventh months as during the first two does not lend much encouragement to the view that the man who does not violate within a year is not likely to violate at all. The facts indicated by the distributions of the present study rather raise the question of whether many more would not be violators if given a little more time. May not the relatively low violation rates for the institutions be due in large part to the short period of time which men are kept on parole? At least there is nothing in the materials that suggests any reason for concluding that one year is a sufficiently long period of time to keep a man on parole.
(d)—Subsequent Criminal Record The cases of all men placed on parole for the two year period from July 1, 1922 to June 30, 1924 were checked for information on any further crime committed by them from the time of discharge from parole to July 1, 1929. This gave a period of from four to six years during which the return to a criminal career might have taken place. Table XXII indicates in concrete form what was found in the official Parole records.54
PREDICTION METHODS AND PAROLE
TABLE XXII
Number and Percentage of Men Placed on Parole Between July 1st, 1922 and June 30th, 1924 Who Had Subsequent Criminal Records by June 30th, 1929 (four to six years after discharge from parole). Information from official parole files for Minnesota State Prison and the State Reformatory, 1922-29.
Status as to subsequent record	Prison	Reformatory	Pris. & Ref.
Number with minor record	16	13	29
Number with major record	20	21	41
Number having both major			
and minor record	2	5	7
Total with subsequent record 38		39	77
Total number paroled July 1st			
1922 to June 30th, 1924	219	221	440
Percent of total with			
subsequent record	17.4	17.6	17.5
Unfortunately the above information is too incomplete to be given much weight. Subsequent crime resulting in a new sentence to a Minnesota state penal institution is of course included, as are the cases of those sentenced in other states where there has been correspondence with the Minnesota parole officials regarding identification and past record. There is also a record of such petty offenses committed in this state as have led to correspondence with the parole officials by the local authorities. The results may perhaps present a fairly true picture of the amount of serious crime committed in this state by men previously discharged from parole in Minnesota. Beyond that, however, no conclusions can be drawn from the present information on subsequent criminal records. A number of men no doubt commit petty offenses in the state, and outside the state, without having that fact recorded in the parole records. No doubt, too, some men commit serious offenses outside the state without ever having the new crime registered in the Minnesota parole office.
8.—The Relative Importance of the Various Factors Involved in Violation or Non-Violation of Parole
It is apparent that the different factors for which distributions are listed in Tables XI, XIV, XV, XVI and XVII are not equally significant in their relation to outcome on parole. Even casual inspection will make clear that factors such as “Previous work habits” (No. 19, Table XIV) or “Prior criminal record” (No. 2, Table XI) have a much more positive and definite relation to outcome on parole than a factor such as “Age at conviction” (No. 22, Table XV.) That there is some degree of association between the subclasses of the categories and the three classes of “outcome on parole” is evident in most of the cases. Yet mere inspection of the distributions does not indicateCHAPTER FOUR
55
the amount- or extent of this association. In order to get a numerical index of this relation it is necessary to apply some statistical measure of association.
The statistical device adopted for this purpose was the contingency method developed by Professor Karl Pearson especially for data entered under descriptive categories or classes, i The result is the coefficient of mean square contingency, expressed by C, as a single numerical index of the degree of association found between the three classes of “outcome on parole” and the subclasses of any category in the list. In order to decrease the routine work of computation, contingency coefficients were worked only for the combined Prison and Reformatory group. This represents roughly a mean of the contingencies that would have been found had the institutions been taken separately. It is not likely that such contingencies would vary a great deal from the one found for the combined group. Taking the combined groups has the advantage of dealing with larger total numbers and consequently with larger numbers in the respective “cells” of the contingency tables.
Contingency coefficients for the relation between “outcome on parole” and forty-four factors on which there was information in the records appear in rank order in Table XXIII. It will be noticed that the factor apparently most definitely associated with outcome on parole is the fact of prior criminal record; likewise, the least important are age at conviction and the receiving of aid by the inmate’s dependents during his incarceration.
For purposes of interpretation the contingency coefficient is usually taken to be roughly equivalent to the Pearsonian r. The upper limit, however, is uncertain, varying with the number of categories in the classification used. Thus, for a four cell contingency table (that is, a 2x2 classification) the highest value possible for C is .707; for a 25 cell table (5x5 classification) the highest value is .894; and for a 100 cell table (10x10 classification) the highest value possible is .949. 1 2 Most of the coefficients in Table XXIII are based on four, five, or six fold classifications. Hence, the highest values for C possible in any given case would be between .82 and .91. Since the values obtained in no case approach their possible upper limits more specific figures for this need not be given.
1	Problems of association and the mean square contingency method of measuring association are discussed at some length in G. U. Yule, An Introduction to the Theory of Statistics, pp. 25-74, especially pp. 64-74, (London 1927 Eight Edition). A shorter and less involved explanation of contingency methods is given by H. E. Garrett, Statistics in Psychology and Education, pp. 195-203 (New York 1926). In the actual computation of C, Garrett’s exposition of the “short method” as given on pages 201-203 of his book, has been followed. The contingency method was originally proposed in Karl Pearson’s, “On the Theory of Contingency and its Relation to Association and Normal Correlation,” Drapers’ Company Research Memoirs, Biometric Series i, Dulau & Company, London 1904.
2	Yule G., op. cit.y p. 66.56
PREDICTION METHODS AND PAROLE
TABLE XXIII
Contingency Coefficients Obtained from the Comparison of the Relation Between “Outcome on Parole” and Each of 44 Other Factors Arranged in Rank Order for the Combined Minnesota State Prison and State Reformatory Group, 1922-27.
Number of C cases
Category compared with “outcome on parole”
1192	.283
1156	.256
1192	.241
1192	.237
1187	.227
1186	.214
1185	.208
1191	.208
1192	.204
1192	.200
1180	.183
1175	.179
1180	.177
1175	.173
1192	.164
1175	.149
1179	.145
1175	.145
1189	.142
917	.139
1093	.103
1192	.092
722	.088
1192	.088
1175	.088
1179	.086
973	.086
1173	.084
1188	.083
1184	083
766	.081
1192	.079
654	.079
1185	.079
558	.079
665	.077
1097	.077
1192	.074
618	.074
983	.069
1184	.068
1179	.060
1192	.060
1184	.050
Previous' criminal record
Work record while on parole
Marital status at time of offense
County from which received
Prison punishment record
Social type of inmate (six place classification)
Work habits prior to conviction
Occupation at or before conviction (six place scale)
Nature of crime of which convicted
Size and type of community in which offense was
committed
Size and type of community in which inmate was brought up
Habits and character: whether ambitions or lazy Size and type of community into which inmate was paroled
Habits and character: whether honest or dishonest
Number of times parole agent visited inmate on parole
Habits and character: use of drugs
Institute of Child Welfare classification of occupation
Habits and character: use of liquor
Mobility of inmate before conviction
Estimate of inmate's mentality (by prison officials)
Home condition (whether parents are living, dead,
separated)
Length of maximum sentence imposed Mental age (from mental tests)
Number of associates in offense for which convicted Habits and character: sexual immorality Inter-relation of communities where brought up, convicted, and paroled Weight (in pounds)
Where living on parole in relation to where brought up Education of inmate before conviction State or country where born Number of siblings in inmate's family Months in prison before parole was granted Whether trial judge favored clemency for inmate Number of hearings before parole board before paroled Number of children (if married)
Intelligence Quotient (from tests)
Wage received on parole Whether inmate pleaded guilty or not guilty Whether county attorney favored clemency for inmate Height (in inches),
Judgment of prison conduct (made by prison officials) Where living on parole in relation to where offense was committed Age at conviction
Whether aid was given dependents while inmate was in prisonCHAPTER FOUR
57
9.—Summary
The mass of detailed material presented in this chapter must not be permitted to obscure the important general facts that appear in connection with the sum total of the results. Attention is especially called to the following points:
1.	No factor on which the records contained information appears to be of outstanding importance in connection with violation or nonviolation of parole. This is apparent from an examination of the distribution and appears even more definitely in the low contingency values indicated in Table XXIII.
2.	As a corollary to the above, it should be noticed that while, no factor is of outstanding importance, yet few, if any, are of no importance. Tables XII and XIII suggest that most of the factors have a relationship to outcome on parole that cannot be accounted for entirely by chance.
3.	There seems to be nothing to justify the belief that most violations of parole occur shortly after release from the institution. Tables XX and 'XXI indicate rather that violations continue at a relatively uniform rate from the time of release and suggest that if the period on parole were longer the proportion violating would presumably also be correspondingly greater.CHAPTER Y
The Question of Reliability or Consistency
1.—The Problem of Reliability in Classification
An important practical question needs to be asked in connection with the materials presented in Chapter IV, namely: How reliable or consistent are the entries under the various categories of' classification?
i. _
It must be remembered that the parole files contain a conglomerate mass of materials including the letters of a varied correspondence, numerous reports, form blanks, recommendations, opinions and contradictions. At best the information is sketchy and incomplete. When this material then is checked against a systematic classification code and items entered according to a scheme of concepts and definite categories, it is inevitable that considerable error in judgment will enter into the result. Material that at one time leads to one characterization and entry under one category will at another time, when the mental set of the investigator is different, lead to a different characterization and entry under a category more or less different. Thus in trying to apply the category of “Social Type” (Table XIV, Number 20, p. 42) the judgment of a case at one time may have led to the characterization “ne’er-do-welL” In rechecking this case at a later date, however, it is entirely possible that the characterization would be “weak character”, or even “respected workman”, or something else. The more highly conceptualized the schedule used, the more subjective and evaluative the entries called for are, the greater will be the errors of this kind.
How serious is this over-lapping of categories and differences in judgment from time to time? To what extent do the results obtained by such a method represent simply chance variation in “guesses” and subjective judgments? Would a rechecking of the original materials give results consistent with the first finding or would there be no appreciable agreement?
2.—Procedure for Checking Reliability in This Study
As partial answers to the above questions, two methods were used in this study to ascertain the reliability or consistency of the general procedure :
58CHAPTER PIVE
59
(a)	An attempt was made to check the writer’s judgment of the records with that of a competent person who had had no previous contact with the study and who presumably would not hold to the same biases and preconceptions. ^ This person’s task was simply to take the present schedule of categories and apply it to the material of the records. 1 After making clear to this investigator the make up of the records, the files of sixty-three cases were gone through foif an independent check. For this purpose the investigator was provided with a typewritten copy of the schedule of categories. These were somewhat elaborated verbally at the time the recheck began, but no further instructions were added after the work began. This group of cases and the results obtained are referred to as “Investigator X” throughout the chapter.
(b)	An attempt was also made to test the writer’s first judgment of a case by a recheck of a group of records at a later time. After an interval of about eight weeks this was done on a group of 198 cases, which included the sixty-three cases also checked by “Investigator X.” This recheck group is referred to as “2nd entry” throughout this chapter. The cases selected for rechecking were obtained by taking the first one hundred released from the prison and the first hundred released from the reformatory for the year 1925 as those names appeared in the official book of “Monthly Releases”. There was no apparent selection in regard to nationality, residence, crime convicted of, length of sentence imposed, etc., except insofar as the year 1925 may have been unusual which seems highly improbable. Since this was merely a recheck to test the consistency of the waiter’s own judgments of the records, the only selective factor that might be important would be the possibility that the records might be notably better or worse in 1925 than in 1922 or in 1927, or any of the other years. This is highly improbable and any error of selection involved would be minor in comparison with the tendency to error in a second judgment of a case originally checked at an earlier date.
In this recheck for reliability only fourteen major categories were used, namely: occupation at time of conviction; Institute of Child Welfare classification of occupation at conviction; social type; type and size of community in which inmate was brought up; type and size of community in which offense was committed; type and size of community into which inmate was paroled; inmate’s previous work habits; outcome on parole; mobility before conviction; and five categories dealing with the inmate’s habits and characteristics, namely: ambition, honesty, use of liquor, use of drugs, and sexual morality. Investigator “X” did not attempt to apply the Institute of Child Welfare classification of occupations.
1 This supplementary investigation had to be paid for out of the writer’s private funds. It was therefore impossible to make it extensive.60
PREDICTION METHODS AND PAROLE
Use of the two methods indicated above gives the following four reliability series on which indexes of comparison may be worked out for each of the fourteen categories:
Series A :
Series B Series C Series D
Own 1st entry with own 2nd entry on 198 cases. Through an oversight two cases were not comparable and so the 200 originally employed for recheck purposes reduce to 198.
Own 1st entry with Investigator “X” on 63 cases.
Own 2nd entry with Investigator “X” on 63 cases.
Own 1st entry with own 2nd entry on the same 63 cases used by Investigator “X”.
3.—The Question of Statistical Methods and Procedures
m
At first glance it would seem that the problem of measuring the extent of consistency or reliability on a recheck of a group of cases should be relatively simple. It is only in the application of present statistical methods to the specific problem that one begins to be aware of their inadequacy. The difficulty in this case lies in the nature of the material. The entries have all been made under more or less specific descriptive categories. No adequate statistical technique has been developed, so far, for measuring the reliability of material so entered. The further implications of this fact will be discussed at some length in this section of the present chapter.
Three procedures suggest themselves as giving measures of the degree of correspondence between 1st and 2nd entry, namely: (a) the Fearsonian coefficient of correlation; (b) the coefficient of mean square contingency; and (c) the analysis of the scatter diagram to see what percent of the whole is in full agreement, what part fails of agreement by one step or class, etc. A brief statement in regard to each will help make clear the problem.
The Pearsonian r as ordinarily used rests on the assumption that the observed phenomena are measurable on a scale the units of which are numerically comparable. In applying the correlation technique to categories of a qualitative sort, it is then necessary to arrange these into a “scale” from the most desirable or “highest” to the least desirable or “lowest,” or vice versa. Such a procedure, however, involves two major difficulties: in the first place, it is necessary to determine the value of the scale steps, which, at best, may be done only by more or less arbitrary assumptions; in the second place, there is never any certain knowledge as to the exact order or sequence of categories that should be utilized in order to form a true scale. There is a consequent practical difficulty in giving any sensible meaning to the resulting regression equations. These objections are fundamental and indicate the serious limitations of the correlation method when applied to materials such as are here involved. Yet, there may be special conditions under which these objections are of less significance than in the usual computations of r.CHAPTER FIVE
61
When the purpose of the correlation is simply to get some measure of consistency, or the degree of correspondence between 1st and 2nd entry of a group of cases using the same classifications, the assumptions as to “step-values” and scale positions may not be so important. In a situation where one is interested merely in checking the consistency of the classifying technique, making no assumptions as to causation and entering into no argument as to the dependence or independence of factors involved, it may perhaps be permissable to use the Pearsonian r. At least, it would seem that if its use is ever permissable on material entered under a schedule of categories, it would be in the restricted sense here suggested.
The coefficient of mean square contingency was devised by Pearson especially for material entered under a system of mutually exclusive categories. 1 As a general rule the C and the r give about the same numerical value when applied to the same data. The contingency measures the departure from the expected probability. When the number of cases is small, however, any random departure tends to affect unduly the size of the C. The particular difficulty with the C for purposes of measuring the degree of consistency in the present problem lies in the fact that there are only a small number of cases and these are scattered in numerous “cells” in the contingency tables. Consequently the numbers in each cell are small and random fluctuations play an undue part in the value for C thus derived. Further implications and limitations of the method will be discussed in connection with the concrete materials.
Analysis of the scatter diagrams in terms of the percentage of the total in full agreement gives the simplest and most “common sense” measure of consistency. If, in material entered under the same scheme of classification, 1st and 2nd entry show full agreement in, say, 90 percent of the cases, it is evident that the procedure has been 'highly consistent. If it is conceded that the confusion of some categories is more serious than that of others, then the whole group of categories may be arranged in a “scale” and the analysis stated in terms of percentage in agreement and percentage failing of agreement by one, two, or more, class or “scale-steps.” This latter procedure amounts to substantially the same thing as the Pearsonian r only the “size” of the step or class interval is not involved. It reflects the amount of deviation from the common category on (1st and 2nd entry, for example), which the C fails to do, but is not so definitely influenced by “scale position” as the r due to the more involved method of computing the latter.
A concrete example will serve to make clear the advantages and limitations of the methods outlined above. For this purpose it will be convenient to consider the category “Occupation at Time of Conviction” from Series A (i. e., own 1st entry with own 2nd entry on
1 See note at the bottom of page 55, Chapter IV for references on the contingency method and the title of Pearson’s original paper proposing it.62
PREDICTION METHODS AND PAROLE
198 cases). The schedule of subclasses used has already been indicated (Table XIV, p. 41, but will be repeated here for convenience:
8. Professional (lawyers, doctors, dentists, teachers)
7. Student (high school or college)
6. Business (proprietors)
5« Clerks and Bookkeepers 4. Farmers (actual farm operators)
3. Skilled and semi-skilled laborers 2. Farm hands 1. General unskilled labor
These were taken to be mutually exclusive. That is, a man would be entered under only one subclass in the occupational classification. Since many men had varied careers with many kinds of work included, the principle was adopted of classifying a man in terms of the kind of work he actually seemed to have been spending most of his time and energy on at or shortly before the offense and conviction. Naturally, in many cases the information in the records was so incomplete that the classification became only an approximation, something of a rough “guess” asi to what he probably spent most of his time doing. It was, however, felt to be best to classify a man in spite of meager information, since full information on previous occupation was scarcely ever available.
Difficulties in classification were numerous. Thus, there was one man who had been employed for several yearsi as a baker who lost his job in a period of business depression, and secured temporary employment as an ordinary laborer on a bridge building gang. While thus engaged he stole a car and was convicted of larceny. His case was entered under the classification “skilled and semi-skilled labor” though the particular job held at the time of1 the offense was “unskilled labor.” Under the term “Farmer” was included only those who actually operated farms; merely working on a farm as a laborer without the responsibility of running it was counted as “Farm hand.” Uhder “Clerks and bookkeepers” were included such jobs as clerking in a store or warehouse, filing clerks, taking care of the books and waiting on customers in small business places, and routine clerical work or bookkeeping in larger establishments. Under “Business” was included only those who were conducting independent enterprises such as stores, tailor shops, restaurants, banks, moving picture theatres, and the like, either as responsible operators or partners in such operation. Salesmen were counted as “clerks and bookkeepers” if of the minor “house to house peddler” type. Securities salesmen and real estate men, who appeared to be doing a considerable volume of business, were classed as “Business.”
Table XXIV summarizes the results from the use of correlation and contingency methods bn the category “occupation at or before conviction” for reliability Series A.CHAPTER FIVE
63
TABLE XXIV
Comparison of r and C Values in Reliability Series A (own 1st entry with own 2nd entry) on the Category “Occupation at or before Conviction,” for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925,
Institution	Number	r	S. E.	C
Prison	98	.607	.065	.786
Reformatory	100	.600	.064	.822
Prison and Reformatory	198	.611	.044	.838
When the method of percentage analysis of the scatter is applied to the combined Prison and Reformatory group for this category in reliability series A, the following results are obtained:
Full Agreement	114 cases	57.6%
1 step off	34 cases	17.2%
2 steps off	40 cases	20.2%
3 steps off	6 cases	3.0%
4 steps off	3 cases	1.5%
More than 4 steps off	1 case	0.5%
	198 cases	100.0
The fact that there is perfect agreement in 57.6% of the cases on a recheck indicates a very positive tendency* to consistency in classification. If it be conceded that it is a more serious error of classification to confuse unskilled labor with business, for example, than with farm hands, then the fact that nearly 75% of the cases fall within 1 scale step (that is, within Vs of the scale) of perfect agreement indicates relatively high accuracy in classification. Similarly, 95% of the cases fall within 2 scale steps (% of the scale) of perfect agreement.
Inspection of the scatter diagrams of Figure 1 (p. 64) indicates considerable discrepancy in the entries. Thus in the first entry group there were 10 classed as in business, in the second only 6; the first found only 44 skilled and semi-skilled, the second 67, etc. Where this decre-pancy involves the confusion of subclasses several “steps” apart on the scale used, the r and the percentage method of analysis are affected. The C, however, is not influenced by this fact. It measures merely the departure from expected probability for each cell and is not influenced by the magnitude of the deviation. That is perhaps the chief reason the value for the C is so much higher than that obtained for the r. Then, too, the very small numbers in the cells and the large number of empty cells tend to make the value for C untrustworthy.
The dependence of the r on scale arrangements can easily be demonstrated. The scale adopted in Figure 1 (Scale I) represents roughly the occupational hierarchy determined from the army intelligence tests. There may be considerable disagreement, however, in regard to the desirability of taking intelligence test performance as a basis for constructing occupational scales. It is for instance, probably true that64
PREDICTION METHODS AND PAROLE
Figure 1
Scatter Diagrams for the Category “Occupation at or before Conviction” (Scale 1) on Reliability Series A (Own 1st entry with own 2nd entry) for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.
Prison and Reformatory
2nd Entry
		1	2	3	4	5	6	7	8	T
	8									0
	7							2		2
>> Vi	6	1		3		1	5			10
«	5			2		12				14
	4		1							1
	3	11		31		1	1			44
	2	14	35	10	4	1				64
	1	29	9	21	1	3				63
	T	55	45	67	5	18	6	2	0	198
Scale I
8. Professional 7. Students 6. Business
5. Clerks and Bookkeepers 4. Farmers
3. Skilled and Semi-skilled 2. Farm Hands 1. General Unskilled Labor
r — .611 C —.838
Highest possible C —.935
Prison
2nd Entry
Reformatory
2nd Entry
	1	2	3	4	5	6	7	8	T			1	2	3	4	5	6	7	8	T
8									0		8									0
7									0		7							2		2
6	1		2			5			8	>» u	6			1		1				2
5					8				8	ts	5			2		4				6
4		1							1	W	4									0
3	6		20		1	1			28	+> m	3	5		11						16
2	14	12	10	4	1				41		2		23							23
1	6	1	4	1					12		1	23	8	17		3				51
T	27	14	36	5	10	6	0	0	98		T	28	31	31	0	8	0	2	0	100
r = .607 C — .786
Highest C possible —.935
r— .600 C = .822
Highest C possible= .935
when the young man in school (most of the men classified as “students” in this study were in high school) seeks industrial employment, he gets a job at about the level of the “farm hand” or of thel “skilled or semiskilled” laborer. It might therefore be argued that the subclass “student” should be placed in a scale position between these two subclasses. Without attempting to decide the question of the validity of such an argument, suppose the point be granted for purposes of illustration. Figure 2 (Scale II) indicates the distribution obtained with the scale changed to conform to this view. The r value, it should beCHAPTER FIVE
65
Figure 2
Scatter Diagrams fori the Category “Occupation at or before Conviction” (Scale II) on Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.
Prison and Reformatory
2nd Entry
		1	2	3	4	5	6	7	8	T
	8									0
	7		1		3	1		5		10
»-1	6		1							1
c	5				2	12				14
	4	11			31	1		1		44
•P m	3			2						2
	2	14	35		10	1	4			64
	1	29	9		21	3	1			63
	T	55	45	2	67	18	5	6	0	198
					Prison					
				2nd		Entry				
		1	2	3	4	5	6	7	8	T
	8									0
	•7	1			2			5		8
>>			1							1
-P «	5					8				8
w	4	6			20	1		1		28
	3									0
rH	2	14	12		10	1	4			41
	1	6	1		4		1			12
	T	27	14	0	36	10	5	6	0	98
r —.470 C= .783
Highest C possible—.935
Scale II
8. Professional 7. Business 6. Farmer
5. Clerks and Bookkeepers 4. Skilled and Semi-skilled 3. Student 2. Farm Hand 1. General Unskilled Labor
r = .481 C — .838
Highest possible C —.935
Reformatory
2nd Entry
		1	2	3	4	5	6	7	8	T
	8									0
	7				1	1				2
p	6									0
"S H	5				2	4				6
	4	5			11					16
-p w	3			2						2
rH	2		23							23
	1	23	8		17	3				51
	T	28	31	2	31	8	0	0	0	100
r = .413 C — .822
Highest C possible —.935
observed, drops from .611 to .481. The percentage distribution of the scatter, on scale II, is indicated below:
Full agreement		114 cases	57.6%
1	scale step off (scale II)	26 cases	13.1%
2	scale steps off	11 cases	5.6%
3	scale steps off	37 cases	18.7%
4	scale steps off	8 cases	4.0%
More than 4 scale steps off		2 cases	1.0%
		198 cases	100/.0%
The greater dispersion of cases, apparent from inspection, is confirmed by the percentages above. Comparison of the results for Figure 1 with those for Figure 2 makes clear the fact that only two values remain unchanged with the change in scale positions, namely, the value for C and for the “percentage in full agreement on the scatter.” The r varies widely and the “step analysis” shows similar changes.66
PREDICTION METHODS AND PAROLE
Without entering into any extended discussion of the relative merits or demerits of the two scales used above, it should be pointed out that they rest upon different assumptions with reference to the importance attached to school work and intelligence test scores as related to the work in the world/ that men do. In adopting a scale arbitrary choice becomes necessary. The important question, then, becomes that of, the assumptions underlying the scale adopted. The difficulty in using the r on material of this kind lies in the fact that the measure of consistency which it gives may reflect primarily the preconceptions and biases back of the scale. If one regards intelligence tests as important indexes of the abilities of the various social and occupational classes, Scale I will probably be selected. If, however, one is inclined to discount the ratings assigned to the various classes by intelligence tests, Scale II may be preferred. But the one who prefers Scale I will have the satisfaction of apparently getting considerably higher consistency in classification than the one who prefers the other scale. It would seem that any measure so directly dependent on the arbitrary assumptions of the investigator adopting the scale can not be considered as particularly desirable in attempts to get an objective statistical technique.
In view of these changing values with changes in the “scale” arrangements of the subclasses of classification, the question is natural and inevitable. Which of the three methods gives the most satisfactory measure of consistency or reliability? No categorical answer can well be given. Important considerations involved in the use of each may, however, be pointed out.
The C is untrustworthy for present purposes due to the small number of cases involved and the further fact that it does not register the magnitude of deviation from the correct subclass in a recheck. It is clearly a more serious error in the reclassification of occupations to confuse, unskilled labor with “professional” or “business” than with “f&rm, hands.” Any statistical index which is insensitive to this sort of error fails to give an adequate measure of the degree of consistency or reliability in reclassification. The r, on the other hand, is sensitive to the magnitude of deviation, but it is probably too directly dependent on the “scale” arrangements adopted to be taken at face value. A further difficulty in the use of fhe r for this material lies in the concentration of cases at the “lower” end of the assumed scale. Thus in Figure 1, in a scatter diagram! of 64 cells, 44 % of the total cases lie in four cells at the lower left hand corner while 81% of the cases lie in nine cells in this same corner. What importance to attach to the value of the r obtained under; such conditions is problematical, to say the least.
In this situation of uncertain values for the r and the C, perhaps the most helpful approach is through the percentage analysis of the scatter. The “percentage in full agreement” is a real and substantial value indicating the amount of actual identity and in no way influenced by “scale position” or “step-values” of the subclasses. As such, it is a simple general measure of consistency and, in the case of limited numbers, probably a more satisfactory measure of extent of agree-CHAPTER FIVE
67
ment than the C. Where it is desired to get some measure of the magnitude of deviation in the use of subclasses which may legitimately be thought of as a progressive series or “scale,” then the “step analysis” in connection with “percentage in full agreement” appears more satisfactory than the Pearsonian r. This method, at least, does not involve so many questionable assumptions and it makes use of much less involved computations.
With these difficulties in mind, the presentation of the “reliability results” for this study has taken the following form:
(a)	The next section of this chapter will present the scatter diagrams for all the categories on which reclassification was attempted in Series A (i. e., 1st entry with 2nd entry). Values for r and C have been computed for each scatter diagram and are included for whatever interpretation the reader may wish to make of them. Explanations and comments on the classifications used are interspersed between the scatter diagrams.
(b)	Following the scatter diagrams will come a section given over to tables summarizing the results from the different methods of measuring consistency. Scatter diagrams are included only for the larger group of Series A and the results from the minor Series B, C, and D are indicated in tabular form.
4.—Scatter Diagrams for “Reliability Series A”
The materials for the category “Occupation at time of conviction” has already been presented in scatter diagram form in Figures 1 and 2. Further comment is unnecessary at this point.
Figure 3 (p. 105) gives the scatter of the series A (Own 1st entry with own 2nd entry) cases oil the Minnesota Institute of Child Welfare classification of occupation. 1 This represents an attempt to make the classification more objective than the relatively subjective categories of the schedule arranged by the writer. The plan of the Institute classification may be seen from the following outline:
Group	I.	Professional (12 professional groups listed by name)
Group	II.	Proprietors and Business Executives (14 specific groups named)
Group	III.	Upper Trades and Specialties (34 groups named)
Group	IV.	Trades and Skilled Workmen (33 groups listed by name)
Group	V.	Semi-skilled (16 groups listed, such as messengers, doormen, guards, night watchman, etc.)
Group	VI.	Laborers (5 groups of laborers named)
1 Acknowledgment of indebtedness to the Minnesota Institute of Child Welfare for permission to use this schedule is gratefully made.68
PREDICTION METHODS AND PAROLE
Figure 3
Scatter Diagrams for the Minnesota Institute of Child Welfare Classification Applied to the Category “Occupation at or before Conviction” on Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Prison and Reformatory, 1925.
Prison and Reformatory
2nd Entry
		6	5	4	3	2	i r	
								1.
	1				1		i	2.
-+J	2	1	3	1		7	12	q
tí	3		3	2	1		6	o. 4
	4	3	7	8	2	1	21	5
	5	8	22			1	31	6.
	6	85	31	5	6		127	
Scale
Group I (Professional)
Group II (Proprietors)
Group III (Upper Trades and Specialties) Group IV (Trades and Skilled Workmen) Group V (Semi-skilled, Messengers) Group VI (Laborers)
T 97 66 16 10	9	0 198
r = .591 C —.690
Highest C possible—.913
Prison
2nd Entry
		6	5	4	3	2	1 T
	1				1		1
	2	1	2			5	8
>»	3		2	2	1		5
W	4	1	5	5		1	12
	5	5	13			1	19
«	6	33	14	1	5		53
	T	40	36	8	7	7	0 98
r — .562 C — .718
Highest C possible —.913
			Reformatory			
				2nd Entry		
		6	5	4	3 2 1	T
	1					0
	2		1	1	2	4
>» t-t	3		1			1
tí	4	2	2	3	2	9
	5	3	9			12
cc rH	6	52	17	4	1	74
	T	57	30	8	3 2 0	100
				Yz	= .598	
				C:	= .644	
Highest C possible —.913
The total number of occupational groups listed by name is 114. In addition to the r and C values indicated in Figure 3, the following percentage analysis of the scatter for the combined group of Prison and Reformatory should be noted:
Full agreement	123	cases	62.1%
1/step off	50	cases	25.3%
2	steps off	14	cases	7.1%
3	steps off	10	cases	5.1%
More than 3 steps off	1	case	0.5%
Total	198 cases	100.1%
It will be observed that all three methods of measuring consistency or reliability give rather uniform and comparable results. The Institute of Child Welfare schedule seems to have given about the sameCHAPTER FIVE
69
Figure 4
Scatter Diagrams for the Category “Social Type of Inmate” on Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.
Prison and Reformatory
2nd Entry
		1	2	3	4	5	6	7	8	9	T
	9			1	3				4	3	11
	8			1	9	3		1	15		29
>>	7			1		12		5			18
4-> a	6				1	1					2
W	5	2			15	41		3	3		64
4J W	4		1		32	2			7		42
1-1	3			2	5				2		9
	2				3	3			4		10
	1	2			5	5			1		13
	T	4	1	5	73	67	0	9	36	3	198
Scale
9. Responsible and substantial Citizen
8. Respected workman 7. Irresponsible youngblood 6. Recent Immigrant 5. Weak character (poor family) 4. Ne’er-do-well
(resident loafer)
3. Transient worker (hobo)
2. Small town or “country bully” 1. City “Tough”
r—.321 C = .727
Highest possible C = .943
Prison
2nd Entry
Reformatory
2nd Entry
		1	2	3	4	5	6	7	8	9	T			1	2	3	4	5	6	7	8	9	T
	9			1	2				4	3	10		9				1						1
	8				7				12		19		8			1	2	3		1	3		10
	7			1		3		1			5	K In	7					9		4			13
	6				1						1	-P ft	6					1					1
ft m	5	1			5	7					13	H	5	1			10	34		3	3		51
-p	4		1		27				6		34	■p M	4				5	2			1		8
iH	3			1	4				2		7		3			1	1						2
	2				1				2		3		2				2	3			2		7
	1	1			3	1			1		6		1	1			2	4.					7
	T	2	1	3	50	11	0	1	27	3	98		T	2	0	2	23	56	0	8	9	0	100
r—.377	r. —205
C —.797	C — .652
Highest possible C = .943	Highest possible C = .943
degree of accuracy in reclassification that the arrangement of categories worked out by the writer did.
Figure 4 gives the scatter in a recheck of the series A cases on the category “social type.” This is the least definite and satisfactory of all the categories used. The classes are all quite subjective. Confusion in reclassification is therefore to be expected. The schedule adopted grew out of the characterizations and descriptive phrases found in the records in a preliminary check of about fifty cases. Reduced70
PREDICTION METHODS AND PAROLE
to categories they led the list given below (arranged as a scale) :
9. Responsible and substantial citizen:	(banker, lawyer,
merchant, farm owner, etc. who occupies a position of trust and responsibility in the community and whose integrity has been unquestioned)
8. Respected workman: (A settled, well established, regular “working man.” Less prominent than the “responsible and substantial citizen,” but presumably of just as unquestioned integrity)
7. Irresponsible youngblood: (youth of good substantial family, well brought up, who needs to- “settle down”)
6. Recent immigrant: (adult immigrant in this country less than 10 years; “greenhorn” who gets into trouble largely through ignorance of ways and language)
5. Weak character: (youth of poor or irresponsible family who has never amounted to anything; not especially vicious, just “nothing there”)
4. Ne’er-do-well: (older than the “weak character”; applied to men of about 30 or more who belong to the nondescript category of “harmless no good’s”; differs from the “transient” in that this group has fairly fixed residence)
3. Transient worker: (hobo or drifter; no permanent connections and no role in any community)
2. Small town or country “bully”: (rural “tough guy” with belligerent attitude towards community, police, and control; fighter involved in minor squabbles of many kinds)
1. City tough: (runs with a bad crowd and has a definite bad record; rather definitely identified with a continuous delinquent and criminal career)
The following percentage distributions of the scatter for the combined group of Prison and Reformatory should be noted in connection with Figure 4:
Full agreement	100 cases	50.5%
1 step off	28 cases	14.1%
2 steps off	20 cases	10.1%
3 steps off	14 cases	7.1%
4 steps off	24 cases	12.1 %
5 steps off	5 cases	2.5%
6 steps off	5 cases	2.5%
More than 6 steps off	2 cases	1.0%
	Ì98 cases	99.9%
Examination of the scatter for the “social type” distribution shows a large amount of overlapping in the categories at the center of the scale. It appears that a man may at one time be classed as a “respected workman” and on a second attempt classed as a “ne’er-do-well” or a “weak character.” In fact, the four categories occupying scale positions 3, 4, 5, and 8 respectively show overlapping with, almost every other category in the scale. This is a serious weakness of the above classification. It is therefore not surprising to see that all of the indexes of reliability are low. Yet the fact that 50.5% of the cases are in full agreement suggests a relationship considerably better than chance. There are eighty-one. cells in the scatter diagram. If chanceCHAPTER FIVE
71
Figure 5
Scatter Diagrams for the Category “Type and Size of Community in Which Inmate was Brought Up,” Reliability Series A, for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.
Prison and Reformatory
2nd Entry
		1	2	3	4	5	T
	5		5	11	3	36	55
+3	4		4	3	8	2	17
£ w	3	2	2	34	0		38
	2	2	27	27	10	1	68
rH	1	16	1	3			20
	T	45	39	61	12	41	198
Prison
2nd Entry
	1	2	3	4	5	T
* 5		4	4	2	15	25
P 4		1		4		5
1 3	2	2	31			35
^ 2	9	13	4			26
,3 1	5	1	1			7
T	16	21	40	6	15	98
r —.752 C — .791
Highest possible C = .894
Scale
5. City over 25,000 4. Small City 5,000-25,000 3. Outside State no other information
2. Small town up to 5,000 1. Farm or open country
r — .755 C — .755
Highest possible C —.894
Reformatory
2nd Entry
	1	2	3	4	5	T
!» 5		1	7	1 _	21	30
P 4		3	3	4	2	12
H 3			3			3
+3 2	18	14	6	1	3	42
X i	11		2			13
T	29	18	21	6	26	100
r -= .758'	;
C —.711. .
Highest possible C — .894
were the only factor involved in the distribution, every cell would be as likely to have a case as every other and the 198 cases wtould be distributed with a number equal to (about) 2.4 cases in each cell. The nine cells along the diagonal (indicating perfect agreement) would then, by chance alone, include (about) twenty-two cases. Actually there are 100 cases along the diagonal. The difference is presumably due to the tendency toward consistency in the use of the subclasses of the classification. Chance would seem to account for agreement in about 10 to 12 percent of the cases but not in 50% of them. This tendency to consistency is reflected in the relatively high C ; the range of confusion in reclassification results in a relatively low r. In this case, therefore, the percentage analysis of the scatter gives a truer picture of the actual situation than either thej r or the C.
One point should, perhaps, be stressed. The scatter diagram is fundamental in presenting the results of these consistency checks. The statistical methods utilized represent merely attempts to give simple numerical indexes of meaning to the data thus recorded. The actual72
PREDICTION METHODS AND PAROLE
Figure 6
Scatter Diagrams for the Category “Type and Size of Community in Which Offense was Committed,” Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.
Prison and Reformatory
2nd Entry
		1	2	3	4	T
>» »-<	4	1	7	3	89	100
P3	3		3	12	2	17
	2	17	84	8	1	60
+» M »-1	1	17	4			21
	T	35	48	23	92	198
		Prison			
		2nd	Entry		
	1	2	3	4	T
E? 4	1	4	1	53	59
tì 3		3	5		8
H 2	7	14	2		23
» 1	4	4			8
iH					
T	12	25	8	53	98
		r=,	.835		
		C =	.729		
Highest C possible —.866
Scale
4. City over 25,000 3. Small city 5,000-25,000 2. Small town up to 5,000 1. Farm or open country
r — .857 C = .746
Highest C possible —.866
Reformatory
2nd Entry
	1	2	8	4	T
>> 4		3	2	36	41
£ 3			7	2	9
W 2	10	20	6	1	37
w 1	13				13
iH					
T	23	23	15	39	100
r — .871 C = .766
Highest C possible = .866
observed facts, showing both agreement and disagreement, are graphically portrayed in the scatter diagrams and no errors or shortcomings of the statistical indexes adopted can change the relationships there indicated.
In keeping with this view, scatter diagrams are presented for the other categories on which a 2nd entry was made in Series A, The list of scatter diagrams included, with the categories to which they apply, follows :
Figure 5
Figure 6
Figure 7
Figure 8
Figure 9 Figure 10
Figure 11
Figure 12
Type and Size of Community in Which Inmate was Brought Up (p. 71)
Type and Size of Community in Which Offense was Committed (p. 72)
Type and Size of Community Into Which Inmate was Paroled (p. 73)
Inmate’s Work Habits Prior to Conviction (P. 74)
Outcome on Parole (p. 75)
Mobility before Conviction, Native and Foreign-born (p. 76)
Mobility before Conviction, Native born only (P. 77)
Five Categories Dealing with “Habits and Character Traits” (p. 78)CHAPTER FIVE
73
Figure 7
Scatter Diagrams for the Category “Type and Size of Community Into Which Inmate was Paroled,” Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.
Prison and Reformatory
2nd Entry
	1	2	3	4	T
>> A u 4		1	7	86	94
13		1	19	1	21
H 2	26	23	4		53
« 1	22	6		2	30
T	48	31	30	89	198
		Prison			
		2nd	Entry		
	1	2	3	4	T
t>> A u 4		1	3	50	54
C 3 ja W 2			11		11
	11	7	2		20
2 1	7	4		2	13
T	18	12	16	52	98
		r—,	.841		
		C:=	.770		
Highest C possible = .866
Scale
4. City over 25,000
3.	City	5,000-25,000	
2.	Town	up	to 5,000
1.	Farm	or	open country
r — 887 C — .769
Highest C possible —.866
Reformatory
2nd Entry
		1	2	3	4	T
>» u	4			4	36	40
c W	3		1	8	1	10
	2	15	16	2		33
w rH	1	15	2			17
	T	30	19	14	37	100
r = .925 C=:.771
Highest C possible —.866
The subclasses of the category “Mobility before Conviction” have been discussed in Chapter IV (p. 42). The other categories are reasonably clear and may be passed over without detailed explanation. It should be noticed that in no case is there perfect agreement, but, on the other hand, in every case there is evident a considerable clustering of cases along the diagonal indicating agreement. This condition is indicated roughly, and with some degree of accuracy, by the r and C values included in each scatter diagram.
5.—Tabular Presentation of Reliability or Consistency Results
It will be recalled that the procedure followed gave material on four reliability series, namely:
Series A: own 1st entry with own 2nd entry on 198 cases.
Series B: own 1st entry with Investigator X on 63 cases.
Series C: own 2nd entry with Investigator X on 63 cases.
Series D: own 1st entry with own 2nd entry on the same 63 cases used by Investigator X.74
PREDICTION METHODS AND PAROLE
Figure 8
Scatter Diagrams for the Category “Inmate’s Work Habits Prior to Conviction,” Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.
Prison and Reformatory
2nd Entry
	1	2	3	4	T
E? 4		3	12	25	40
C 3 H 2	1	15	36	6	58
	6	41	12	2	61
® 1	30	6	3		39
T	37	65	63	33	198
Scale
4. Loafer (seldom works)
3. Odd jobs only 2. Irregularly employed (many changes)
1. Regularly employed (few changes)
r — .767 C = .722
Highest C possible —.866
Prison
2nd Entry
	l	2	3	4	T
►» 4			7	13	20
■§ 3		8	11	3	22
W 2	2	25	6		33
-p 1 in	21	2			23
					
T	23	35	24	16	98
r — .869 C — 764
Highest C possible —.866
Reformatory
2nd Entry
	1	2	3	4	T
& 4		3	5	12	20
3 3	1	7	25	3	36
W 2	4	16	6	2	28
-p 1 m x i—i	9	4	3		16
T	14	3,0	39	17	100
r = .644 C — .661
Highest C possible^ .866
Scatter diagrams have been presented showing the results for Series A for all categories on which a 2nd entry had been made. The various results for all the series and all the categories may be summarized in tabular) form, as follows:
Table XXV:	Gives results in Series A for all categories
except those dealing with “Habits and Character Traits*” r and C values together with “percentage in agreement” and “percentage failing of agreement by one scale step” are indicated.
Table XXVI:	Gives results for all four reliability series in
terms of “percentage in full agreement on the scatter” for all categories except those dealing with “Habits and Character Traits.”
Table XXVII: Gives results for all four reliability series in terms of- “percentage in full agreement on , the scatter” for the five categories dealing ■with “Habits and Character Traits.”CHAPTER FIVE
75
Figure 9
Scatter Diagrams for the Category “Outcome on Parole,” Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.
Prison and Reformatory
2nd Entry
>>	1	2	3	T
f-t fl 3 W 2	2	3	32	37
	1	7	6	14
■5 1	144	2	1	147
*-<				
T	147	12	39	198
Prison
2nd Entry
	1	2	3	T
>>				
1 3		1	14	15
W 2		5	6	11
m 1	69	2	1	72
T	69	8	21	98
r—.904 C — .726
Highest C possible = .816
Scale
3. Major Violator (total failure)
2. Minor Violator (partial failure)
1. Non-violator (success)
r = .904 C = .716
Highest C possible = .816
Reformatory
2nd Entry
	1	2	3	T
t 3	2	2	18	22
H 2	1	2		3
1 m rH	75			75
T	78	4	18	100
r = .921 C = .734
Highest C possible = .816
The C values in Table XXV must be interpreted with the following limitations in mind:1
In a	3	place	contingency	table	the	highest	C	possible	is	.816
In a	4	place	contingency	table	the	highest	C	possible	is	.866
In a.5 place contingency table the highest C possible is .894 In a	6	place	contingency	table	the	highest	C	possible	is	.913
In a	7	place	contingency	table	the	highest	C	possible	is	.926
In a	8	place	contingency	table	the	highest	C	possible	is	.935
In a	9	place	contingency	table	the	highest	C	possible	is	.943
The consistency with which the three different methods of measuring reliability vary should be noticed. Thus for “Size and Type of Community Into Which Inmate Was Paroled” the r is .887, the C is .869, with 75.8% of the cases in full agreement on the scatter and with 98.5% included within one scale step of perfect agreement. But for “Inmate’s Work Habits Prior to Conviction” the r drops to .767, the C to .722, with only 66.7% in perfect agreement and 95.5% within one scale step of perfect agreement. Similar comparisons for the other categories indicate a corresponding tendency to consistency in variation.
Table XXVI gives the results for all four reliability series in terms of “percentage in full agreement on the scatter” for all categories except those dealing with “Habits and Character Traits.”
1 From Yule, G. U., An Introduction to the Theory of Statistics, p. 66.76
PREDICTION METHODS AND PAROLE
Figure 10
Scatter Diagrams for the Category “Mobility before Conviction, Native-born and Foreign-born,” Reliability Series A for The Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.
Prison and Reformatory
2nd Entry
	1	2	3	4	5	6	7	8	T	7.
8								1	1	6.
7						1	3		4	
!» 6						8			8	5.
fi 5 S					10	2			12	4.
W 4		2	3	5					10	3.
® 3	2	5	26	2					35	
^ 2	9	42	11	1					63	2.
1	54	8	3						65	1.
										
T	65	57	43	8	10	11	3	1	198	
Scale
F. B,, Transient, no fixed residence
F. B., Some travel, “bummed”
1-2 years
F. B., Lived several places in or near state
F. B., Always in same neighborhood
N. B., Transient, no fixed residence
N. B., Some travel, “bummed”
1-2 years
N. B., Lived several places in or near state
N. B., Always in same neighborhood r—.924 (1
				Prison							Reformatory						
			2nd Entry									2nd	Entry				
	1	2	3	4	5	6 7	8 T		1	2	3	4		5 6	7	8	T
8							0	8								1	1
7						1	1	7							3		3
6						8	8	„ c									0
^ 5 +->					8	1	9	fi 5						2 1			3
A																	
H 4		1	2	3			6	H 4		1	l	2					4
t 3			6				6	t 3	2	5	20	2					29
£ 2	6	33	3	1			43	^ 2	3	9	8						20
1	20	5					25	1	34	3	3						40
T	26	39	11	4	8	10 0	0 98	T	39	18	32	4		2 ,1	3	1	100
No values for r and C have been computed for series B, C or D because of the small number of cases included. The percentage results presented can be taken only as general approximations considerably influenced, no doubt, by chance fluctuations in the small number of cases. As such, the figures serve to confirm what one would expect, namely, that there is, in general, greater consistency between the writer’s, own 1st and 2nd entry than between either of his entries and those of Investigator X. The surprising thing is rather the relatively high consistency shown in Series B and C where comparisons are made with the classifications of Investigator X. This suggests the probability that, if another person should repeat this study using the same classifications, substantially the same results would be obtained.
1 These are spurious correlations; including foreign born as additional steps on the same scale is simply weighting the correlation in a positive direction.CHAPTER FIVE
77
Figure 11
Scatter Diagrams for the Category “Mobility before Conviction, Native-born only,” Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.
Prison and Reformatory
2nd Entry
	1	2	3	4	T
^ Â h 4		2	3	5	10
£ 3	2	5	26	2	35
4. 2	9	42	11	1	63
X 1	54	8	3		65
T	65	57	43	8	173
	1	Prison 2nd Entry 2 3	4	T
►» 4		1 2	3	6
g 3		6		6
H 2	6	33 3	1	43
-p i to x	20	5		25
tH				
T	26	39 11	4	80
		r — .775		
		C — .756		
Highest C possible = .866
Scale
4. Transient, no fixed residence
3. Some travel; “bummed” one or two years
2. Lived several places m or near state
1. Always in same town or neighborhood
r = .785 C — .722
Highest C possible —.866
Reformatory
2nd Entry
	1	2	3	4	T
i* 4		1	1	2	4
% 3	2	5	20	2	29
H 2	3	9	8		20
■s1 iH	34	3	3		40
T	39	18	32	4	93
r — .733 C = .685
Highest C possible= .866
Table XXVII gives the results for all four reliability series for the five categories dealing with “Habits and Character Traits of the Inmate.” Here, again, agreement with Investigator X is surprisingly consistent. In a two-place classification, such as used here, the chance factor is, of course, relatively very large. It should be noticed, however, that in every case the reliability results obtained are considerably higher than can be accounted for by chance.78
PREDICTION METHODS AND PAROLE
Figure 12
Scatter Diagrams for Five Categories Dealing with “Habits and Character Traits of the Inmate,” Reliability Series A for the Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925.
Subclasses I. b. Good worker, ambitious a. Lazy, shiftless	s>>	Prison Group 2nd Entry a b T		Reformatory Group 2nd Entry a b T >>			Prison and Reformatory 2nd Entry a b T		
	+-> ib 4J a m	4 34	30 34 30 64	p tì b W p a	10 50	25 35 15 65	p b W p a m	14 84	55 69 45 129
	T	38	60 98	T	60	40 100	rH T	98	100 198
II. b. Honest, reliable a. Dishonest,		2nd a	Entry b T	>» u	2nd a	Entry b T	>» Sh	2nd a	Entry b T
	i- j i cl	4 38	32 36 24 62	lb I 3/	13 48	19	32 20	68	P S K W b p a Xfl	17 86	51 68 44 130
reckless	H T	42	56 98	T	61	39 100	T	103	95 198
ill.	>>	2nd a	Entry b T	>>	2nd a	Entry b T	>>	2nd a	Entry b T
b. No record of liquor a. Uses liquor	C K W b -M a m	7 72	12 19 7 79	Ib a p	4 70	12 16 14 84	p b H b p a cc	11 142	24 35 21 163
	T	79	19 98	T	74	26 100	T	153	45 198
IV.	>>	2nd a	Entry b T		2nd a	Entry b T	>.	2nd a	Entry b T
b. No record of drugs a. Drug addict	p fi h p a co	4	90 90 4 8	|b j 8/ «3		98 98 2 2	¡b j -1 E m	4	188 188 6 10
	T	4	94 98	iH T	100 100		T	4	194 198
V. b. No record of sex a. Sexually immoral	>>	2nd a	Entry b T	>»	2nd a	Entry b T	>»	2nd a	Entry b T
	c W b w a r-H	4 42	29 33 23 65	H b ^ a CQ	3 68	19 22 10 78	p 1 b p a	7 110	48 55 33 143
	T	46	52 98	rH T	71	29 100	T	117	81 198CHAPTER FIVE
79
Table XXV
Summary Table of the Reliability Results for Series A (Own 1st entry with, own 2nd entry) on All Categories Except those Dealing with “Habits and Character Traits of Inmate’7 for Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 1
Category Institution		r	C	Xi	X2	X3
Occupation at or	P&R	.611	.838	57.6%	17.2%	74.8%
before conviction	R	.600	.822			
(Scale I; 8x8)	P	.607	.786			
Institute of Child	P&R	.591	.690	62.1%	25.3%	87.4%
Welfare Classification	R	.598	.644			
of Occupation (6x6 scale) P		.562	.718			
Social Type of Inmate	P&R	.321	.727	50.5%	14.1%	64.6%
(9x9 scale)	R	.205	.652			
	P	.377	.797			
Type and Size, of Com-	P&R	.755	.755	61.1%	24.2%	85.3%
munity in Which Inmate was R		.758	.711			
Brought Up (5x5)	P	.752	.791			
Type and Size of Com-	P&R	.857	.746	76.8%	18.7%	95.5%
munity in which Offense	R	.871	.766			
was committed (4x4)	P	.835	.729			
Type and Size of Com-	P&R	.887	.769	75.8%	22.7%	98.5%
munity into Which Inmate R		.925	.771			
was Paroled (4x4)	P	.841	.770			
Inmate’s Work Habits	P&R	.767	.722.	66.7%	28.8%	95.5%
Prior to Conviction	R	.644	.661			
(4x4)	P	.869	.764			
Outcome on Parole	P&R	.904	.716	92,4%	6.1 %	98.5%
(3x3)	R	.921	.734			
	P	.904	.726			
Mobility before Convic-	P&R	.9242				
tion, Native and Foreign	R	.8792				
born (8x8)	P	.9492				
Mobility, Native born	P&R	.785	.722	64.1 %	19.2%	83.3%
only (4x4)	R	.733	.685			
	P	.775	.756			
1	Key: P = Prison — 98 cases; R = Reformatory == 100 cases; P&R = Prison and Reformatory = 198 ases; X1 = Percentage in full agreement on the scatter; X2 = Percentage failing of agreement by one scale step; X3 = sum of X1 and X2.
2	Spurious correlations due to faulty scale.80
PREDICTION METHODS AND PAROLE
Table XXVI
“Percentage in Full Agreement on the Scatter” Shown for All Categories for Each of the Four Reliability Series. Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory* 1925 1
Insti- Category tution under comparison R= Reformatory P = Prison		Series A P&R = 198 R= 100 P= 98	Series B P&R = 62 R = 30 P = 32	Series C P&R = 63 R= 30 P = 33	Series D P&R = 62 R= 30 P = 32
Occupation of Inmate	P&R	57.6%	46.8%	57.2%	58.1%
at or before Conviction	R	63.0	53.3	43.3	70.0
(8x8)	P	52.1	40.6	69.7	46.9
Institute of Child2	P&R	62.1			
Welfare Classification	R	66.0			
of Occupation (6x6)	P	58.2			
Social Type of Inmate	P&R	50.5	41.3	44.4	53.2
(9x9)	R	48.0	43.3	53.3	56.7
	P	53.1	39.4	36.4	50.0
Type and Size of Commun-	P&R	61.1	71.0	71.4	67.8
ity in Which Inmate was	R	53.0	73.3	66.7	56.7
Brought Up (5x5)	P	69.4	68.8	75.8	78.1
Type and Size of Commun-	P&R	76.8	77.4	88.9	77.4
ity in Which Offense	R	76.0	83.3	93.3	86.7
was Committed (4x4)	P	77.6	71.9	84.8	68.8
Type and Size of Commun-	P&R	75.8	82.3	88.9	74.2
ity Into Which Inmate	R	75.0	83.3	96.7	80.0
Was Paroled (4x4)	P	76.5	81.3	81.8	68.8
Inmate’s Work Habits	P&R	66.7	59.7	58.7	72.6
Prior to Conviction	R	62.0	56.7	50.0	73.3
(4x4)	P	71.4	62.5	66.7	71.9
Outcome on Parole	P&R	92.4	87.1	84.1	88.7
(3x3)	R	95.0	90.0	86.3	90.0
	P	89.8	84.5	84.8	87.5
Mobility Before Conviction,	P&R	75.3	72.6	74.6	82.3
Native and Foreign born	R	71.0	73.3	83.3	83.3
(8x8)	P	79.6	71.9	66.7	81.3
1	Key: Series A Own 1st entry with own 2nd entry; Series B Own 1st entry with Investigator X; Series C Own 2nd entry with Investigator X; Series D Own 1st entry with own 2nd entry on the same group of cases used by Investigator X.
2	Investigator X did not check on the Minnesota Institute of Child Welfare Classification of Occupation,CHAPTER FIVE
81
Table XXVII
“Percentage in Full Agreement on the Scatter” Shown for Five Categories Dealing with “Habits and Character Traits of the Inmate” for Each of the Four Reliability Series. Minnesota State Prison, the State Reformatory, and the Combined Group of Prison and Reformatory, 1925 * 1
Insti Series Series Series Series
Category Under Comparison 2 3 4 5 R = P =		tution Reformatory Prison	A P&R —198 R = 100 P= 98	B P&R = 62 R= 30 P= 32	C P&R = 63 R = 30 P = 33	D P&R = 62 R=30 P= 32
1.	Ambition	. P&R	70.2%	74.2%	81.0%	74.2%
		R	75.0	73.3	76.7	83.3
		P	65.3	75.0	84.8	65.6
2.	Honesty	P&R	69.2	79.0	69.8	, 67.7
		R	67.0	83.3	66.7	56.7
		P	71.4	75.0	72.7	78.1
3.	Use of Liquor	P&R	83.8	69.9	76.2	81.0
		R	82.0	60.0	76.7	80.0
		P	85.7	81.3	75.8	84.4
4.	Use of Drugs	P&R	97.0	96.8	100.0	96.8
		R	98.0	96.7	100«0	96.7
		P	95.9	93.8	97.0	96.9
5.	Sexual Immorality P&R		79.8	90.3	82.5	80.6
		R	87.0	90.0	86.7	90.0
		P	72.4	90.6	78.8	71.9
1	In the four-cell scatter diagrams for these categories “chance” would account for agreement up to fifty percent.
2	Subclasses used for these categories :
1.	Ambition
b. Lazy, shiftless
a.	Good worker, ambitious
2.	Honesty
b.	Dishonest, reckless
a.	Honest, reliable
3.	Use of Liquor
b.	Uses liquor
a.	No record of use of liquor
4.	Use of Drugs
b.	Uses drugs: drug addict
a.	No record of the use of drugs
5.	Sexual Immorality
b.	Record of sexual immorality a. No record of sexual immoralty82
PREDICTION METHODS AND PAROLE
6.—Summary
A few of the more important points involved in the materials on reliability presented; in this chapter may be summarized as follows:
1.	Contradictory and incomplete information in the parole records together with the difficulty of getting satisfactory objective classifications make probable a considerable degree of uncertainty and error in reclassification.
2.	Measuring the reliability of the classification used was attempted by reclassifying 198 cases at a later date and by having an independent outside investigator use the writer’s schedule of categories on a small group of these same cases.
3.	No entirely satisfactory statistical indexes for measuring the reliability of a reclassification of material entered under categories has been found.
4.	Three statistical indexes of reliability, none very satisfactory, were Utilized in this study, namely, the Pearsonian correlation, the coefficient of mean square contingency, and the percentage analysis of the scatter. The contingency is untrustworthy due to the small number of cases and the large number of “squares” or “cells” in the contingency tables together with the skewed distributions involved and the further fact that this index fails to register the magnitude of deviation from the correct category. The Pearsonian r is too dependent on the assumed “scale positions” assigned to the categories to be taken at face value, though it does reflect definitely the magnitude of deviation. The “percentage in full agreement on the scatter” gives a definite, positive value not subject to the above criticisms. It is very general, however, and by carrying it further to a “step analysis” one gets something by way of a measure of the magnitude of deviation, without having the result influenced so definitely by the assumptions as to “scale” positions.
5.	The most significant reliability result that appears is the tendency to consistency in variation from category to category on the part of the three different methods used.
6. The category on which the reliability results are the lowest is that of “Social Type of Inmate”; the highest results are for the category “Outcome on Parole.”
7.	In the case of all categories (on which reliability results are available), however, the consistency or reliability results on reclassification are considerably greater than can be accounted for by “chance.”CHAPTER VI
The Possibility of Predicting Outcome on Parole
1.—The Problem of Utilizing Parole Record Information
Social agencies have long been puzzling over the problem of how to make intelligent use of their case histories. They have found their files bulging with miscellaneous information having some bearing on their cases but never indicating in any intelligible manner what action should be taken. The same problem has confronted parole departments. They, too, have found themselves with a mass of information but with no precise device or method for interpreting it beyond simply “common sense” judgments as to possible implications.
It was for the expressed purpose of determining how useful this mass of information was in actual practice that Professor Warner made his study of the parole records of the Massachusetts Reformatory. 1 His work and the important studies that followed it have already been discussed. 1 2 It has been pointed out that the difficulty with Warner’s study lay, not in his facts, but in his interpretations. Because he found no item of information in the records which had any very close relationship to outcome on parole, he concluded that none of the information was significant. He overlooked the fact, as was suggested earlier, that the factors on which he had information might be individually relatively insignificant, yet their cumulative effect might be most important in connection with outcome on parole. Attention was first called to this fact by H. Hart 3 and the methods he suggested have later been developed and applied by Burgess 4 to the Illinois parole records, and by the Gluecks 5 to the Massachusetts Reformatory records. Both of these later studies found it possible to “predict” outcome on parole by taking into account the cumulative effect of many factors, no one of which taken by itself was of any great importance.
1	Warner, S. B., “Factors Determining Parole From the Massachusetts Reformatory,” Preface by Sanford Bates, Commissioner of Correction. Journal of Criminal Law and Criminology, 14: 172-207, August 1923.
2	See pages 12-18 of this study for a review of the procedure followed by Hornell Hart, E. W. Burgess and S. Glueck.
3	Hart, Hornell, “Predicting Parole Success,” Journal of Criminal Law and Criminology, 14: 405-413, November 1923.
4	Burgess, Ernest W., “Factors Determining Success or Failure on Parole” in Parole and the Indeterminate Sentence by A. A. Bruce, E. W. Burgess, A. J. Harno, and J. Landesco, pp. 205-249. Published by Illinois Board of Parole, 1928.
5	Glueck, Sheldon, and Eleanor T., 500 Criminal, Careers. Alfred A. Knopf, New York, 1930. (xiii + 339 + xvi).
8384
PREDICTION METHODS AND PAROLE
Table XXVIII
Contingency Coefficients, in Rank Order, Obtained in Comparing “Outcome on Parole” with 34 Pre-Parole Factors in the Combined Minnesota State Prison and State Reformatory Group, 1922-27.
Number C of obtain-cases ed
Pre-parole Categories compared with “Outcome on Parole”
1192	.283
1192	.241
1192	.237
1187	.227
1183	.214
1185	.208
1191	.208
1192	.204
1192	.200
1180	.183
1175	.179
1175	.173
1175	.149
1179	.145
1175	.145
1189	.142
917	.139
1093	.103
1192	.092
722	.088
1192	.088
1175	.088
973	.086
1188	.083
1184	.083
766	.081
1192	.079
654	.079
665	.077
1192	.074
618	.074
983	.069
1184	.068
1192	.060
Previous criminal record
Marital status at time of offense
County from which received
Prison punishment record
Social type of inmate (six place classification)
Work habits prior to conviction
Occupation at or before conviction (six place scale)
Nature of crime of which convicted
Size and type of community in which offense was
committed
Size and type of community in which inmate was brought up
Habits and character: whether ambitious or lazy Habits and character: whether honest or dishonest Habits and character: use of drugs
Institute of Child Welfare classification of occupation1
Habits and character: use of liquor
Mobility of inmate before conviction
Estimate of inmate’s mentality (by prison officials)
Home condition (whether parents are living, dead, separated)
Length of maximum sentence imposed Mental age (from mental tests)
Number of associates in offense for which convicted Habits and character: sex immorality Weight (in pounds)
Education of inmate before conviction State or country where born Number of siblings in inmate’s family Months in prison before parole was granted Whether trial judge favored clemency for inornate Intelligence Quotient (from tests)
Whether inmate pleaded guilty or not guilty Whether county attorney favored clemency for inmate Height (in inches)
Judgment of prison conduct (made by prison officials) Age at conviction
The data of the present study show consistent similarity with the above named studies in that no factor appears to be of outstanding importance in connection with outcome on parole. Chapter IV has presented the detailed distributions for forty-four factors as related to parole violation. Of these, only thirty-four have reference to traits, circumstances, or conditions of the man’s, life prior to parole. It is evi-
1 The “Institute of Child Welfare classification of occupation” represents merely the application of another schedule of classifications to the category “Occupation at or before conviction.” The table, therefore, really represents only 33 different pre-parole factors.CHAPTER SIX
85
dent, however, that “prediction” tables can utilize only the results from the pre-parole factors. Table XXVIII p. 84 indicates these, arranged in rank order by contingency values when compared with “outcome on parole,” and shows the C values in each case with the number of cases on which they are based. Attention should be called to the fact that information is incomplete on several items. Thus, the category “nature of the crime of which convicted” shows full information in the total group of 1192 cases; but, the category “statement of county attorney in regard to clemency” contains information on only 618 cases,—not much over one-half of the total number.
Efforts to construct prediction tables for the data of the present study raise a number of questions in regard to procedure, such as:
1.	Should such tables be based on all pre-parole factors even though some may be more important than others, or should an attempt be made to select out and use only the more important ones? The Burgess study included all factors available while the Gluecks used the contingency value as a basis for selecting out the ones of greatest importance.
2.	What weight should be given the various factors when they are combined into cumulative tables? The Burgess study weighted all factors equally' and took the “average rate of violation” for the institution as the base on which to construct a scale of “points for number of factors better than average.” The Glueck method adopted a system of weighting whereby each factor was counted in terms of its actual percentage figures for violation. That is, an individual’s failure score was obtained by adding the violation rate percentage values for each category of classification under which the case would fall.
3.	What should be done with the “no information” subclasses in the various categories? Neither the Burgess nor the Glueck studies made use of such a subclass. In the data of the present study, however, information was complete on only a relatively few categories and the “no information” subclasses must therefore be handled in some way. At least three possibilities are apparent: (a) the categories on which information is incomplete might be eliminated and prediction tables constructed on only those with complete information; (b) all categories might be included with the “no information” subclasses treated in exactly the same way as any other subclass (that is, on the basis of the violation rate found for the group so classified under any specific category) ; (c) a middle course might be followed by eliminating from consideration those categories in which a considerable proportion of the cases fall under the “no information” subclass, yet retaining those categories with nearly complete information and treating the small number of “no information” cases like any other subclass.
4.	May not prediction tables be constructed on the combined group of Prison and Reformatory and the results applied to predicting the outcome on parole for each institutional group taken separately? If this is possible much routine effort can be saved in the labor required to construct tables for the separate institutional groups. It would also86
PREDICTION METHODS AND PAROLE
have the added advantage of dealing with larger numbers with a consequent decrease in chance errors in the various percentages obtained.
These represent some of the more fundamental questions in regard to procedure. Answers can only be found by adopting an experimental approach. Try out the various possibilities suggested and see which method gives the most useful and generally satisfactory results. The various methods attempted are outlined in detail in the next section.
2.—Prediction Methods Attempted in The Present Study
In attempting to construct prediction tables for these data, it was felt that the methods adopted should be made to throw some light on the questions raised in the last section. A detailed outline of procedure follows:
A.	The institutional groups were first divided into two randomly selected halves in order that the results from one half might be used to predict the outcome in the other, and in the group as a whole. These predicted values may then be compared with the actual distributions obtained. For convenience in description, these two halves were labelled “operating group” and “control group” respectively, though of course, in any strict sense, they are not what the terms imply. This division gave the following institutional groups:
1.	For the Prison:
(a)	an “operating” group of 272 cases
(b)	a “control” group of 270 cases
(c)	the whole group of 542 cases
2.	For the Reformatory:
(a)	an “operating” group of 325 cases
(b)	a “control” group of 325 cases
(c)	the whole group of 650 cases
3.	For the combined group of Prison and Reformatory:
(a)	an “operating” group of 597 cases
(b)	a “control” group of 595 cases
(c)	the whole group of 1192 cases
B.	In the actual construction of prediction tables, the methods indicated below were used. Tables were constructed based on:
I. The highest 17 pre-parole factors from the rank order contingency list in Table XXVIII. All the categories with a C value of .100 or more were included, giving a total of 17 factors (omitting the duplicate item of “Institute of Child Welfare” classification of occupation) . This method is in keeping with the implied thesis of the Gluecks* study that only the categories showing highest contingency values in comparison with outcome on parole should be used in constructing prediction tames. Both the Burgess and the Glueck methods of scoring were tried:
(a) Using the Burgess method of scoring:
1. For the combined Prison and Reformatory group, with the average violation rate (Prison and Reformatory average, 21.8%) as the base:
Table XXIX:	Whole group, 1192 cases
(p. 90)CHAPTER SIX
Table	XXX:	Operating group,	597
cases (p. 92)
Table XXXVII:	Control group, 595 cases
(Appendix A)
2.	For the Prison group, based on the combined Prison and Reformatory average violation rate (21.8%):
Table XXXVIII: Whole group, 542 cases (Appendix A)
Table	XXXIX:	Operating group,	272
(Appendix A)
Table XL:	Control group,	270
cases (Appendix A)
3.	For the Reformatory group, based on the combined Prison and Reformatory average violation rate (21.8%):
Table	XLI:	Whole group, 650	cases
(Appendix A)
Table XLII:	Operating group,	325
cases (Appendix, A)
Table XLIII:	Control group,	325
cases (Appendix A)
4.	For the Prison group, based on the average violation rate for the Prison (24.7%):
Table	XXXI:	Whole group, 542	cases
(P. 93)
Table XLIV:	Operating group,	272
cases (Appendix A)
Table XLV:	Control, 270 cases
(Appendix A)
5.	For the Reformatory group, based on the average violation rate for the Reformatory (19.4%):
Table XXXII:	Whole group, 650 cases
(P. 94)
Table XLVI:	Operating group,	325
cases (Appendix A)
Table XLVII:	Control group,	325
cases (Appendix A)
(b) Using the Glueck method of scoring:
1.	For the combined Prison and Reformatory group, using the Prison and Reformatory violation rate percentage values:
Table XXXIII:	Operating	group,	597
cases (p. 96)
2.	For the Prison group, using the Prison and Reformatory violation rate percentage values:
Table XLVIII:	Operating	group,	272
cases (Appendix B)
3.	For the Reformatory group, using the Prison and Reformatory violation rate of percentage values:
Table XLIX:	Operating	group,	325
cases (Appendix B)
II. The lowest 17 pre-parole factors from the rank order list of contingency coefficients, Table XXVIII (p. 84) (Count up 17 from the bottom of the list). This pro-88
PREDICTION METHODS AND PAROLE
cedure was tried as a check on the validity of the assumption that only the more important factors in the contingency list should be considered:
(a) Using the Burgess method of scoring:
1. For the Prison group, based on the average violation rate for the Prison (24.7% ):
Table XXXIV:	Whole group, 542 cases
(P. 99)
Table L:	Operating group, 272
cases (Appendix C)
Table LI:	Control group, 270
cases (Appendix C)
III. A selected group of 25 pre-parole factors in which information was nearly complete with only a small proportion of the cases coming under “no informa-tion” subclasses. (The categories included are those with information on over 1000 cases, except that of “Institute of Child Welfare Classification of Occupation,’’ in the list of Table XXVIII (p. 84) :
(a) Using the Burgess method of scoring:
1.	For the Prison group, based on the Prison average violation rate (24.7%):
Table XXXV:	Whole group, 542 cases
(p. 101)
Table LII:	Operating group,	272
cases (Appendix D)
Table LIII:	Control group,	270
cases (Appendix D)
2.	For the Reformatory group, based on the average violation rate for the Reformatory (19.4%);
Table XXXVI : Whole group, 650 cases (p. 102)
Table LIV:	Operating group,	325
cases (Appendix D)
Table LV:	Control group,	325
cases (Appendix D)
The Glueck method of scoring was not used extensively because it involves considerably more work than the Burgess method, and besides, was found to give substantially the same results when applied to the Prison and Reformatory “operating” group. Of the 27 prediction tables listed above, eight (XXIX-XXXVI inclusive) will be presented and discussed in the next section of this chapter. The remaining 19 will be included without comment as appendices.
3.—Some Suggestive Results from the Attempts at Predicting Outcome on Parole
This section will present a few of the more important results of the efforts to construct cumulative “prediction” tables for the data. The sequence of tables and methods indicated in the last section will be followed:
I. Tables based on the 17 highest pre-parole factors in the rank order contingency list. The attempt made in the Glueck study to sift out the unimportant factors before constructing prediction tables wasCHAPTER SIX
89
felt to be sound. Accordingly, only those factors were utilized which had C values above .100. This gave the following factors on which to base the cumulative tables:
Previous criminal record; mafital status at time of conviction; county from which received; prison punishment record; social type of inmate; occupation at or before conviction; work habits prior to conviction; nature of offense;
size, and type of community in which offense was committed;
size and type of community in which inmate was brought up;
whether inmate is ambitious or lazy;
whether honest or dishonest;
whether inmate used, drugs;
whether inmate used liquor;
inmate’s mobility prior to conviction,
prison officials’ estimate of inmate’s mentality;
inmate’s home condition.
To the results from these categories the Burgess method of constructing prediction tables was applied. That is, taking the average rate of violation for the group (Prison and Reformatory, or Prison only, or Reformatory only) as a base, each man was credited with one point for every category under which he was classified in a subclass having a violation rate lower than the average for the institution. If he was classified in a subclass having a rate higher than the average for the group, no credit was given. Since there were 17 factors, the highest credit score possible would be 17; the lowest 0. The different institutional groups were then scored by this method, using the appropriate average violation rate as a base.
Using the average violation rate for the Prison and Reformatory (21.8% as a base, the results indicated in Table XXIX (p. 90) were obtained for the combined Prison and Reformatory group. It will be seen that out of the total group of 1192 men, there were 79 who, on 15-17 categories, were classified in subclasses having a lower violation rate than the average (21.8%) for the Prison and Reformatory. At the other extreme was a group of 89 men who were thus classified on only 2-4 categories. Of the 79 in the group at, the upper end of the scale, 2 cases or 2.5% were parole violators. Of the 89 in the group at the lower end of the scale, 53, or practically 60%, were considered parole violators. While these figures represent the j actual distribution of cases in the sample studied, they may also be taken as indicating the probable future violation rate of any similar Prison and Reformatory group similarly studied. They may also be interpreted as giving the probable chances for parole violation on the part of any inmate so classified at the time of parole. For example, if this scoring device be used and an inmate is found to fall in the score class 15-17, the chances are( only 2.5 in 100 that he will violate parole; but, if he falls in,' the score class 2-4, the chances are 59.6 in 100 that he will violate.Table XXIX
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison and Reformatory, 1922-27: Whole Group of 1192 Cases. Distributions based on the average violation rate for Prison and Reformatory (21.8%) . (Burgess method on 17 highest pre-parole factors)
Points for number of factors	
having violation	
rates lower than the	Number
average	of
rate for	men
the	in
institu-	each
tion	group
15-17	79
13-14	147
11-12	227
9-10
7-8
5-6
2-4
Total
243
218
189
89
1192
Experience and Expectancy Rate of Parole Violation								% Who violated parole in:	
								Prison and	Prison and
								Reform-	Reform-
								atory	atory
		Violators						Operat-	Control
								ing	Group
								Group	of
								of 597	595
						Non		Cases.	Cases.
						Violators		(See	(See
Minor		Major		Total				Table	Table
								XXX	XXXVII
N	%	N	%	N	%	N	%	p. 92	App. A)
		2	2.5	2	2.5	77	97.5	2.5	2.6
		2	1.4	2	1.4	145	98.6	2.7	0
2	0.8	21	9.3	23	10.1	204	89.9	10.2	10.1
10	4.2	30	12.3	40	16.5	203	83.5	14.8	18.5
18	8.3	45	20.6	63	28.9	155	71.1	27.4	30.1
10	5.3	67	35.4	77	40.7	112	59.3	39.4	42.1
16	18.0	37	41.6	53	59.6	36	40.4	52.4	66.0
56		204		260		932			
CO
o
PREDICTION METHODS AND PAROLECHAPTER SIX
91
Table XXX (p. 92) shows similar distributions for half of the Prison and Reformatory group (that is, for the “operating group” of 597 cases). These results may be used to “predict” outcome for the whole group. The percentages in each group “expected” to violate are shown in the columns at the right of Table XXIX. In none of the score classes is there any appreciable discrepancy between the percentages “expected” and those actually found violating. This means that with information on only 597 cases, it is possible to tell (with an error of less than 7% in the above example) how many men will violate in the larger group of 1192 cases. Similar “predictions” may be made from the “control” group, not only for the entire Prison and Reformatory group, but also for the “operating” group. These predicted values are indicated in each table in an appropriately labelled column.
Tables were constructed for the prison and the Reformatory group separately, based on the average violation rate for the combined group. These may be found in Appendix! A, Tables XXXVIII to XLIII inclusive. The results are similar to those indicated in Table XXIX, with, perhaps, a tendency to sharper discrimination between violators and non-violators in the prison group and a less clear-cut distinction for the reformatory cases. The results would seem to indicate the general possibility of applying a prediction technique based on the combined group to the separate institutions with some degree of success. The tables, however, are not so satisfactory as those obtained when the institutions are taken separately.
Besides these attempts to utilize the violation rate for the combined group of Prison and Reformatory for the institutions taken separately, tables were also constructed based on the average violation rates of the respective institutions. Table XXXI (p. 93) gives the results for the Prison group based on the average violation rate (24.7%) of the prison; Table XXXII (p. 94) gives the results for the Reformatory on the reformatory average (19.4%). Corresponding “operating” and “control” group tables may be found in Appendix A, Tables XLIV to XLVII inclusive.
When the average violation rate for the specific institution is taken as a base rather than the rate, for the combined group of Prison and Reformatory, it should be noticed that the two extremes of the scale discriminate conduct on parole much more sharply1 than the tables using the combined group as a base. Thus, for the Prison (Table XXXI), all cases falling in the three score classes at the upper end of the scale are non-violators; in the group at the lower extreme of the scale, 72% are violators. As an instrument for predicting outcome on parole, this table therefore represents a distinct improvement over those presented in the preceding pages. The Reformatory results (Table XXXII) show a similar tendency, though it should be noticed that this table distinguishes much less sharply between the violators and the non-violators. This is presumably a reflection of the fact that the classifications are less meaningful when applied to the younger group of men.Table XXX
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison and Reformatory, 1922-27: Operating Group of 597 Cases. Distributions based on the average violation rate for the Prison and Reformatory (21.8%). (Burgess method on the 17 highest preparole factors)
Points for Number of factors	
having violation	
rates lower than the	Number
average	of
rate for	men
the	in
institu-	each
tion	group
15-17	40
13-14	73
11-12	118
9-10
7-8
5-6
2-4
Totals
135
95
94
42
597
Experience and Expectancy Rate of Parole Violation			
Violators			Non Violators
Minor	Major	Total	
N %	N - %	N %	N %
2 1.7 6 4.4 6	6.3 £ 8.5 7	16.7	1	2.5 2	2.7 10 8.5 14	10.4 20 21.1 29 30.9 15	35.7	1	2.5 2	2.7 12 10.2 20 14.8 26 27.4 37 39.4 22 52.4	39 97.5 71 97.3 106 89.8 115 85.2 69 72.6 57 60.6 20 47.6
29	91	120	477
% Who violated parole in Prison and Reformatory Control Group of 595 Cases. (See Table XXVII, App. A)
2.6
0
10.1
18.5
30.1
42.1
66.0
CO
to
PREDICTION METHODS AND PAROLETable XXXI
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Whole Group of 542 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on the 17 highest preparóle factors.
Experience and Expectancy Rate of Parole Violation
% Who violated parole in:
Po:nts for
Number of
factors
having
violation
rates
lower than the average rate for institution	Number of men in each group
15-17	33
13-14	52
11-12	77
9-10	103
7-8	117
5-6	103
2-4	57
Totals	542
Violators
Minor	Major	Total
N %	N %	N %
4 3.9	17 16.5	21 20.4
10 8.6	22 18.8	32 27.4
8 7.7	32 31.1	40 38.8
11 19.3	30 52.6	41 71.9
33	101	134
Non
Violators
N	%
33	100.0
52	100.0
77	100.0
82	79.6
85	72.6
63	61.2
16	28.1
Prison	Prison
Oper-	Control
ating	Group
Group	of
of 272	270
Cases.	Cases.
(See	vSee
table	table
XLIV,	XLV
App.	App.
A)	A)
18.0	22.6
27.4	27.3
38.9	38.8
66.7	76.7
408
<£>
CO
CHAPTER SIXTable XXXII
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27: Whole Group of 650 Cases. Distributions based on the average violation rate for the Reformatory (19.4%). (Burgess method on the 17 highest pre-parole factors)
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for
the
institu-
tion
15-17
13-14
11-12
9-10
7-8
5-6
2-4
Totals
		Experience and Expectancy Rate of						
				Parole Violation				
			Violators					
Number of men in							Non Violators	
	Minor		Major		Total			
each group	N	%	N	%	N	%	N	%
4							4	100.0
46			1	2.2	1	2.2	45	97.8
74			1	1.4	1	1.4	73	98.6
140	2	1.5	10	7.1	12	8.6	128	91.4
146	5	3.4	19	13.0	24	16.4	122	83.6
122	9	7.4	25	20.5	34	27.9	88	72.1
118	7	6.0	47	39.8	54	45.8	64	54.2
650	23		103		126		524	
% Who violated parole in:
Reform-	Reform-
atory	atory
Operat-	Control
n g	Group
Group	of
of 325	325
Gases.	ases.
(See	(See
Table	Table
XLVI,	XLVII,
App.	App.
A)	A)
4.8
2.7
7.7 16.7 25.9 36.4
9.7
16.2
29.4
54.0
CD
PREDICTION METHODS AND PAROLECHAPTER SIX
95
The Glueck method of scoring was also applied to the results on the seventeen highest pre-parole factors. The individual inmate’s score was obtained by adding the percentage values for violation in each subclass in which the man was classified under the seventeen categories enumerated a few pages back. The lowest score possible on these categories was 159; the highest 614. There were no cases, however, at the extreme ends. Appropriate scale divisions were therefore arranged as indicated in Table XXXIII (p. 96). This method has the advantage of weighting the final score in proportion to the actual percentage of violation represented by any subclass under each category. Thus, for example, in the case of the category “Previous Criminal Record” (Table XI, p. 33) in the Prison group the subclass “arrest, probation, or fine” has a violation rate of 24.9% while that of “one major and one minor term” shows a violation rate of 51.4%. Under the Burgess method a man falling in either of the above two subclasses would receive no credit for that category, since the rate in each case is higher than the Prison average of 24.7%. No account would be taken of the fact that the latter subclass has. a violation rate nearly double that of the former. Since both have rates above the average for the institution, no credit would be given for that category. Under the Glueck method, however, the percentage violation rate of the subclass in which a man is classified would be taken and the to cal score would then be weighted by the actual amount of his violation rate without regard to any “average” for the institution.
The results obtained from applying the Glueck method to the Prison and Reformatory “operating” group appear in Table XXXIII (p. 96). It is apparent from simple inspection that the distribution is very much like that obtained by the Burgess method. Comparison of the two methods by means of the Pearsonian r, gives the following coefficients:
Prison and Reformatory “operating” group, 597 cases, r=+.922 Prison “operating” group, 597 cases, r=+.925 Reformatory “operating” group, 325 cases, r=+.922
Tables for the distributions, using the Glueck method, for the operating group of the Prison and Reformatory taken separately may be found in Apendix B. Figure 13 (p. 97) gives the scatter diagram for the comparison of the two methods on the Prison and Reformatory operating group. Due to the high degree of similarity in the results obtained, in the three groups above, no further use was made of the Glueck method. 1
These various attempts have indicated the possibility of constructing reasonably satisfactory prediction tables for outcome on parole
1 The Glueck method is somewhat more laborious, calling for the use of an adding or calculating machine and really requiring two people to do the scoring,—one to locate and call off the percentages, the other to operate the machine and record the results. With the Burgess method, however, one person can readily score the data rapidly and accurately.CO
05
Table XXXIII
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison and Reformatory, 1922-27: Operating Group of 597 Cases. Distributions based on violation rate percentages for the prison and Reformatory gioup. (Glueck method on the 17 highest preparole factors)
			Outcome	on Parole	
	Number		Violators		
Total Failure Score	of				Non
(Sum of percent-	cases				
ages of Violations	in	Minor	Major	Total	Violators
	each				
unaer eacn					
category)	group	N %	N %	N %	N %
211-260	23				23 100.0
261-310	89				89 100.0
311-360	165	2 1.2	14 8.5	16 9.7	149 90.3
361-410	156	9 5.8	24 15.4	33 21.2	123 78.8
411-460	113	12 10.6	32 28.3	44 38.9	69 61.1
461 and over	51	6 11.7	21 41.2	27 52.9	24 47.1
Totals	597	29	91	120	477
PREDICTION METHODS AND PAROLECHAPTER SIX
97
Figure 13
Scatter Diagram Obtained in Applying the Two Scoring Methods (Burgess and Glueck) to the Data of the Minnesota Prison and Reformatory “Operating” Group of 597 Cases, 1922-27.
Glueck’s Method
211	261-	311-	361-	411-	461-	511-
260	310	360	410	460	510	560	Total
2					1	7 2	10
3					3	8 2	13
4					6	13	19
5				1	38	11	50
TJ 6 S 7				8	33	4	45
			1	21	17		39
■£ 8			3	39	13		55
S 9			18	50	4		72
8 10			36	28			64
g> 11		4	48	11			63
S 12		20	35	1			56
M 13	2	19	19				40
14	7	18	5				30
15	7	20					27
16	6	6					12
17	2						2
Total 24	87	165	159	115	43	4	597
			r = .922	S.E.	= .006		
by taking the highest 17 pre-parole factors from the contingency rank order list. In general, it has been demonstrated that the tables based on the average violation rates for the institutions are more satisfactory for prediction purposes than those based on the combined Prison and Reformatory average. Two further changes in method were experimented with the hope of throwing some light on the question of whether more satisfactory tables could be. worked out. These are discussed in the following pages.
II. Tables based on the 17 lowest pre-parole factors in the contingency rank order list. This variation in technique was adopted to check the soundness of the assumption that he unimportant factors (as measured by the C) should be eliminated before constructing prediction tables. The following are the categories included:
Home condition;
length of maximum sentence;
mental age ;
number of associates in offense for which convicted; whether sexually immoral; weight ;
education of inmate before conviction;98
PREDICTION METHODS AND PAROLE
state or country where born;
number of siblings in inmate’s family;
months in prison before parole was granted;
whether trial judge favored clemency for inmate;
intelligence quotient;
whether inmate pleaded guilty or not;
whether county attorney favored clemency for inmate;
height;
judgment of inmate’s prison conduct made by prison officials; age at time of conviction.
The results for the prison group appear in Table XXXIV (p. 99). Corresponding tables for the Prison “operating” and “control” groups may be found in Appendix C.
This method, apparently, gives results nearly as satisfactory as those obtained by using the 17 highest factors on the combined Prison and Reformatory average as a base. The large difference in the percentage who violate in the groups at the two extremes of the scale in Table XXXIV would seem to justify the conclusion that the factors included have considerable significance as indicators of conduct on parole, regardless of their low contingency values. These results seem to warrant the construction of prediction tables based on all of the pre-parole factors in the contingency list. This was not attempted, however, because of thei large number of cases in the ‘no information” subclasses under some of the categories. It seemed better to include only those categories on which there was information in practically all cases. This led to the procedure described below.
III. Tables based on 25 selected pre-parole factors. Since high contingency values seem not to be a necessary requirement for significant factors, the basis of selection was made that of completeness of information. There were found to be 25 factors with information on over 1093 out of the total 1192 cases. These were selected and made the basis of a further effort to construct a more satisfactory prediction table for outcome on parole* The categories used were:
Previous criminal record; marital status at time of offense; county from which received; prison punishment record; social type of inmate; work habits prior to conviction; occupation at or before conviction; nature of crime of which convicted;
size and type of community in which offense was committed;
size and type of community in which inmate was brought up;
whether ambitious or lazy;
whether honest or dishonest;
whether inmate used drugs;
whether inmate used liquor ;
mobility of inmate before conviction;
home condition;
length of maximum sentence imposed;
number of associates in offense for which convicted;
whether sexually immoral;
education of inmate;
state or country where born;Table XXXIV
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Whole Group of 542 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on the 17 lowest pre-parole factors)
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for
the
institu-
tion
10 and ove] 9 8 7 6 5
4 or less
Total
		Experience and Expectancy Rate of						
				Parole Violation				
			Violators					
Number								
								
of							Non	
men in each group	Minor		Major		Total		Violators	
	N	%	N	%	N	%	N	%
70		0	2	2.8	2	2.8	68	97.2
62	2	3.3	2	3.2	4	6.5	58	93.5
84	5	5.9	14	16.7	19	22.6	65	77.4
97	4	4.1	18	18.6	22	22.7	75	77.3
90	4	4.5	21	23.3	25	27.8	65	72.2
73	8	10.9	17	23.3	25	34.2	48	65.8
66	10	15.2	27	40.9	37	56.1	29	43.9
542	33		101		134		408	
% Who violated parole in:	
Prison	Prison
Operat-	Control
ing	Group
Group	of
of 272	270
Cases.	Cases.
(See	(See
Table L,	Table LI,
App. C)	App. C)
3.0	2.7
10.0	3.1
18.0	29.4
32.6	13.7
26.1	29.5
27.8	40.5
48.4	62.9
CD
CD
CHAPTER SIX100
PREDICTION METHODS AND PAROLE
months in prison before parole was granted; type of plea entered;
judgment of prison conduct of inmate made by prison officials; age of inmate at time of conviction.
Results for the Prison and for the Reformatory, using the average violation rate for the respective institutions as a base, appear in Tables XXXV and XXXVI (p. 101, 102). “Operating” and “control” group tables for this series may be found in Appendix D.
These two tables give the sharpest differentiation so far indicated between the violators and non-violators of parole. In the four score classes at the upper end of the scale are 141 cases (\over 26% of the total), all of whom are non-violators; in the group at the lower extreme are 36 cases of whom over 86% are violatQrs. In other words, if this method were to be used on another group of Prison cases the presumption is (other things remaining the same) that for men falling in the score classes “15-16” or above, the chances of violating parole are less than 1 in 100; for those falling in the group “4 or less” the chances of becoming parole violators are over 86 in 100.
The various attempts reviewed in the preceding pages lead to these general conclusions:	All of the methods attempted have given results
that made some distinctions between violators and non-violators. The least satisfactory results were obtained by constructing tables on the 17 lowest pre-parole factors; the most satisfactory distinctions appear in the tables based on 25 selected factors having some information in nearly all cases. The entire procedure shows high internal consistency. This may be seen in the close similarity between the “predicted” values from the “operating” and “control” groups when compared with the group as a whole. This all argues for the general reliability and validity of the method.
4.—Some Considerations in the Application of Prediction Tables
The use of prediction tables for judging in advance of any individual inmate’s “chances” of parole violation has already been explained. Continued application of such tables would involve many other phases of administrative procedure. A few of these may conveniently be considered here:
1. The use of such tables by parole departments would make possible a more scientific parole policy. With the rate of parole violation known in advance and subject to control, the Parole Board might well adopt an experimental policy. If it deemed it wise, the Board could with relative ease so select the group paroled that there would be practically no violation. The “violation rate” would then be decreased, but it is also likely that the number placed on parole w*ould be smaller. On the other hand, if the Board desired, it could parole a much larger proportion of the inmates than at present and knowingly accept the risk of higher violation rates.Table XXXV
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Whole Group of 542 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on 25 selected pre-parole factors)
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for....
the
institu-
tion
21 and over 19-20 17-18 15-16 13-14 11-12 9-10 7-8 5-6
4 or less
Total
Experience and Expectancy Rate of Parole Violation
Number of ... men in
each
group
15
25
38
63
64 88 81 78 54 36
542
Violators
Minor
N	%
	0
	0
	0
	0
	0
3	3.4
5	6.1
6	7.7
12	22.3
■ 7	19.4
33
Major
N	%
	0
	0
	0
	0
5	7.8
9	10.2
22	27.2
21	26.9
20	37.0
24	66.7
101	
Total
N	%
	0
	0
	0
	0
5	7.8
12	13.6
27	33.3
27	34.6
32	59.3
31	86.1
134
Non
Violators
N	%
15	100.0
25	100.0
00 00	100.0
63	100.0
59	92.2
76	86.4
54	66.7
51	65.4
22	40.7
5	13.9
408 ‘
% Who violated parole in;
Prison Prison Operat- Control ing	Group
Group	of
of 272	270
Cases.	Cases.
(See	(See
Table LILTable LIT App. D) App. D)
6.5	9.1
8.5	19.5
'41.7	26.7
35.7	33.3
55.2	64.0
81.3	90.0
o
CHAPTER SIXTaMe XXXVI
Experience and Expectancy Rate of Parole Violation in the Minnesota Reformatory, 1922-27: Whole Group of 650 Cases. Distributions based on the average violation rate for the Reformatory (19.4%). (Burgess method on 25 selected pre-parole factors)
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for
the
institu-
tion
21 and ovei 19-20
17-	18 15-16
18-	14 11-12
9-10
7-8
6 or less Total
Number
of
men
in
each
group
17
37
59
93
104
122
97
75
46
650
Experience and Expectancy Rate of Parole Violation
Minor
Violators
N	%
1	1.1
2	1.9
6	4.9
5	5.1
4	5.3
5	10.9
23	
Major
N
%
1
3
8
20
19
26
26
103
1.7 3.2
7.7
16.4
19.6
34.7
56.5
Total	
N	%
1	1.7
4	4.3
10	9.6
26	21.3
24	24.7
30	40.0
31	67.4
126	
Non
Violators
N	%
17	100.0
37	100.0
58	98.3
89	95.7
94	90.4
96	78.7
73	75.3
45	60.0
15	32.6
524	
% Who violated parole in:
Reform-	Reform-
atory	atory
Operat-	Control
ing	Group
Group	of
of 325	325
Cases.	Cases.
(See	See
Table	Table
LIV,	L,
App. D)	-PP- D)
3.1	
5.6	2.6
15.8	2.2
19.3	23.1
17.4	31.4
35.5	43.2
57.2	75.6
102	PREDICTION METHODS AND PAROLECHAPTER SIX
103
2.	Prediction methods would seem to have their greatest usefulness in making possible the taking of special measures to prevent violation on the part of men found likely to violate. By providing special supervision for those whose “chances” of violation are high, it might be possible to parole a larger proportion of inmates than at present and stilll keep the actual rate of violation at a low figure.
3.	With a parole department making use of such prediction techniques, it would follow as a matter of course, that such use would lead to further research and experimentation on the problem. Once the method was in operation, new conditions and changed circumstances would make necessary continuous modification and adaptation. This, it would seem, should lead to experimentation with methods of control over the men on parole and result in considerable refinement in the prediction technique.
4.	There would seem to be no reason why a “prediction” technique, comparable to the one discussed in this study, should not be applied to a great many other fields of activity. Thus, criminal courts could presumably apply it at the time of the trial and pronounce sentence accordingly. Probation departments already have the essential information in their files and would only need to systematize it in some such manner as was done for the parole records in this study to apply the method to their problems.
The principle of predicting future human conduct, in the mass, from the record of the past, on the assumption that “other things will remain constant,” seems sound and worthwhile. Insurance companies have built up large business enterprises on this principle of predicting the future from the past. The present study, in common with those of Burgess and of the Gluecks, seems to establish the validity of that general principle for other phases of human conduct as well.CHAPTER VII
Summary and Conclusions
The present study was undertaken in order to discover to what extent the information in the records of the Minnesota Parole Department could be utilized to predict outcome on parole. Past practice in this state has relied upon “common sense” interpretations of the information in possession of the Board in reaching a decision on the question of whether to parole any particular inmate. Would a more scientific method be possible on the basis of the present records? That is the primary question in which this study has been interested. The recent studies by Burgess and by the Gluecks have indicated techniques for this problem and the task of the present study has been to apply these methods to the Minnesota records.
For this purpose the records of 1192 men (542 State Prison and 650 Reformatory), paroled during the five year period from July 1st, 1922 to June 30th, 1927, were carefully studied and the information they contained entered in systematic form according to a schedule of categories with appropriate subclasses. Outcome on parole was classed under three heads, namely, non-violators, major violators, and minor violators. The proportion of men in each subclass for each category who were violators was then computed, giving a percentage “violation rate.” Such rates were computed separately for the two institutions and for the combined group of Prison and; Reformatory. In an effort to secure a single consistent measure of the relation between each factor or category of information and outcome on parole, contingency coefficients were computed for all the categories. By placing these in rank order some measure of the relative importance of each factor was obtained.
It is clear, however, that only those factors having to do with the inmate’s life before parole could be made use of in constructing prediction tables for outcome on parole. The records contained information on thirty-four pre-parole factors. Three series of prediction tables were constructed on the basis of these: (a) on the 17 highest in the contingency list; (b) on the 17 lowest ini the contingency list; (c) on a selected group of 25 pre-parole factors, all with information available in nearly all cases. Both thé Burgess and the Glueck methods of scoring were attempted in the case of the “17 highest” series. Tables were constructed for each institution separately and for the combined group of Prison and Reformatory. In each case the total number of cases was divided into two equal parts, an “operating” group and a “control” group, and the results for one part used to “predict” the
104CHAPTER SEVEN
105
outcome in the other and in the group as a whole. In all, twenty-seven different prediction tables were constructed.
It is clear that the validity of the entire procedure depends on the reliability of the original classification. If, on a second reading of the record, a man should be entered under quite different classifications, it would indicate a serious; error in technique. As a test for this, a recheck of 200 cases was made eight weeks after the original entries had been made. Another investigator was employed to make an independent check on 63 of these same cases. This gave results’ on four reliability Series: (a) own first entry with own second entry; (b) own first entry with Investigator X; (c) own second entry with Investigator. X; (d) own first entry with own second entry on the same cases used by Investigator X.
The results for this whole procedure can be stated here in only the most general way. It has been apparent that no single factor, or small group of factors, on which information was available, had any very important relation to outcome on parole. In this the results of the present study were in entire accord with those of earlier studies. When, however, these results were combined into cumulative tables, significant differences appeared between violators and non-violators in every case. In every one of the prediction tables constructed regardless of method used, distributions were obtained in which the men at one extreme of the “scale” were seldom violators while at the other high violation rates were the rule. The tables based on 25 selected factors gave the sharpest distinctions, those on the 17 highest the next best, with those based on the 17 lowest factors showing the least positive differences between violators and non-violators. The checks for reliability indicate some confusion in classification with considerable overlapping of subclasses in some categories and with very little in others. 1 Classification was most uniform on the category “outcome on parole” and least satisfactory cn that of “social type of inmate.” It must be kept in mind, however, that no very satisfactory statistical device was found for measuring the extent of reliability or consistency in the classifications.
What has really been done in all this, it should be stressed, has been to apply and test for the Minnesota records the methods used by Burgess and by Glueck in their studies. Both of these earlier studies indicate definite procedures with the results obtained, but neither one stops to put to the pragmatic test of a trial any of the numerous questions that the methods necessarily raise. Thus, Glueck used a system of weighted scores on a relatively few factors deemed important because of their high contingency values. Burgess used an unweighted scoring device and did not try to select out the important factors. The assumption^ back of each procedure raise questions that the present study has sought to put to the test of actual comparison on the same group of cases. For the group used in this study there was no very great difference in the results obtained from tables based on the 17 lowest rather than the highest factors on the contingency list. The weighted system of scoring used by Glueck was found to correlate with106
PREDICTION METHODS AND PAROLE
an r of +.922 when compared with the unweighted system used by Burgess. Neither of these earlier studies considered the question of possible error in the application of their classifications. The present study found considerable confusion and overlapping, but nevertheless, a consistency that can not be accounted for by chance alone.
The major question with which this study has been concerned seems, therefore, to have been answered rather definitely. The consistency óf the prediction tables that were constructed would seem to indicate that the probabilities that an inmate will violate his parole agreement can be stated in terms of “chances in 100” at the time of the granting of the parole. It would thus seem that “outcome on parole” can be systematically predicted from the information now in the Minnesota parole records.
In order to test further the validity of the entire procedure, it would be highly desirable to have it applied to a group of current parole cases. If such a test were made and the “predictions” made at the time the parole was granted actually gave a true picture of conduct on parole then the method might fairly be said to have been established. Such a crucial test would seem to be the logical next step in the development of American parole procedure.APPENDIX A
Experience and Expectancy Rate of Parole Violation tables based on the 17 highest pre-parole factors, using the Burgess method of scoring.
107Table XXXVII
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison and Reformatory, 1922-27: Control Group of 595. Distributions based on the average violation rate for Prison and Reformatory (21.8%).	(Burgess method in the 17 highest pre-parole
factors)
			Experience and Expectancy Rate of						
					Parole Violation				
Points for Number of									
factors having									
violation				Violators					
rates lower									
than the	Number								
average	of							Non	
rate for the	men in	Minor		Major		Total		Violators	
institu- tion	each group	N	%	N	%	N	%	N	%
15-17	39			1	2.6	1	2.6	38	97.4
13-14	74							74	100.0
11-12	109			11	10.1	11	10.1	98	89.9
9-10	108	4	3.7	16	14.8	20	18.5	88	81.5
7-8	123	12	9.8	25	20.3	37	30.1	86	69.9
5-6	95	2	2.1	38	40.0	40	42.1	55	57.9
2-4	47	9	19.2	22	46.8	31	66.0	16	34.0
Totals	595	27		113		140		455	
% who violate parole in Prison and Reformtory Operating Group of 595 Cases. (See table XXX, p. 92 of text)
2.5
2.7
10.2
14.8
27.4
39.4
52.4
108	APPENDIX ATable XXXVIII
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Whole Group of 542 Cases. Distributions based on average violation rate for Prison and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors)
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for
the
institu-
tion
15-17
13-14
11-12
9-10
7-8
5-6
2-4
Totals
Number
of
men
in
each
group
14
65
101
109
100
98
55
542
Experience and Expectancy Rate of Parole Violation								% Who violated parole in:	
								Prison	Prison
		Violators						Operat-	Control
								ing	Group
								Group	of
								of 272	270
						Non		Cases.	Cases.
Minor		Major		Total		Violators		(See Table	( See Table
								XXXIX,	XL,
N	%	N	%	N	%	N	%	App. A)	App. A)
						14	100.0		
		1	1.5	1	1.5	64	98.5	3.3	
		7	6.9	7	6.9	94	93.1	5.7	8.3
4	3.7	11	10.1	15	13.8	94	86.2	12.3	15.4
10	10.0	25	25.0	35	35.0	65	65.0	36.4	33.9
6	6.1	33	33.7	39	39.8	59	60.2	40.0	39.6
13	23.7	24	43.6	37	67.3	18	32.7	62.1	73.1
33		101		134		408			
APPENDIX A	109Table XXX-IX
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Operating Group of 272 Gases. Distributions based on the average violation rate for the Prison and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors)
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for
the
institu-
tion
15-17
13-14
11-12
9-10
7-8
5-6
2-4
Experience and Expectancy Rate of Parole Violation
Violators
Number
of
men
in
each
group
9
30
53
57
44
50
29
272
Minor
N	%
2	3.5
3	6.9
4	8.0
5	17.3
14	
Major
N	%
1	3.3
3	5.7
5	8.8
13	29.5
16	32.0
13	44.8
51
Total
N	%
1	3.3
3	5.7
7	12.3
16	36.4
20	40.0
18	62.1
65
Non
Violators
N	%
9	100.0
29	96.7
50	94.3
50	87.7
28	63.6
30	60.0
11	37.9
207
% who violate in Prison Control Group of 270 Cases. (See Table XL, App. A)
8.3
15.4
33.9
39.6
73.1
Totals
110	APPENDIX ATable XL
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Control Group of 270 Cases. Distributions based on the average violation rate for the Prison and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors)
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for
the
institu-
tion
15-17
13-14
11-12
9-10
7-8
5-6
2-4
Totals
Experience and Expectancy Rate of Parole Violation
Violators
of men in each group	Minor		Major	
	N	%	N	%
5				
35				
48			4	8.3
52	2	3.9	6	11.5
56	7	12.5	12	21.4
48	2	4.2	17	35.4
26	8	30.8	11	42.3
270	19		50	
Total
N	%
4	8.3
8	15.4
19	33.9
19	39.6
19	73.1
69	
Non
Violators
N
%
5
35
44
44
37
29
7
201
100.0
100.0
91.7
84.6
66.1
60.4
26.9
% who violated parole in Prison Operating Group of 272 Cases. (See Table XXXIX, App. A)
3.3
5.7
12.3
36.4
40.0
62.1
APPENDIX A	111Experience 1922-27: the Prison
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for
the
institu-
tion
15-17
13-14
11-12
9-10
7-8
5-6
2-4
Table XLI
and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, Whole Group of 650 Cases. Distribution based on the average violation rate for and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors)
		Experience and Expectancy Rate of						
				Parole Violation				
			Violators				1	
Number of men in each group								
	Minor		Major		Total		Non V iolators	
	N %		N	%	N	%	N	%
65			2	3.1	2	3.1	63	96.9
82			1	1.2	1	1.2	81	98.8
127	2	1.6	14	11.0	16	12.6	111	87.4
134	6	4.5	19	14.2	25	18.7	109	81.3
118	8	6.8	20	16.9	28	23.7	90	76.3
91	4	4.4	34	37.4	38	41.8	53	58.2
33	3	9.1	13	39.4	16	48.5	17	51.5
650	23		103		126		524	
% Who violated parole in:	
Reform-	Reform-
atory	atory
Operat-	Control
ng	Group
Group	of
of 325	325
Cases.	Cases.
' See	(See
Table	Table
XLII, App. A)	XLIII,
	App. A)
3.2 2.3	2.9
13.6	11.5
16.7	21.4
19.6	26.9
38.6	44.7
33.3	57.1
Totals
112	APPENDIX ATable XLII
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27: Operating Group of 325 Cases. Distributions based on the average violation rate for the Prison and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors)
Points for
Number of
factors
having
violation
rates
lower
than the	Number
average	of
rate for	men
the	in
institu-	each
tion	group
15-17	31
13-14	43
11-12	66
9-10	78
7-8	51
5-6	44
2-4	12
Totals
325
Experience and Expectancy Rate of Parole Violation
Violators
Minor		Major		Total		Non Violators	
N	%	N	%	N	%	N	%
		1	3.2	1	3.2	30	96.8
		1	2.3	1	2.3	42	97.7
2	3.0	7	10.6	9	13.6	57	86.4
4	5.2	9	11.5	13	16.7	65	83.3
3	5.5	7	13.7	10	19.6	41	80.4
4	9.1	13	29.5	17	38.6	27	61.4
2	16.6	2	16.6	4	33.3	8	66.7
15		40		55 -		270	
% who violated in Reformatory Control Group of 325 Cases. (See Table XLIII, App. A)
2.9
11.5
21.4
26.9
44.7
57.1
APPENDIX A	113Table XLIII
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27: Control Group of 325 Gases. Distributions based on the average violation rate for the Prison and Reformatory (21.8%). (Burgess method on the 17 highest pre-parole factors)
Points for Number of factors	
having violation	
rates lower than the	Number
average	of
rate for	men
the	in
institu-	each
tion	group
15-17	34
13-14	39
11-12	61
9-10
7-8
5-6
2-4
Totals
56 67 47 ..21
325
Experience and Expectancy Rate of Parole Violation
Minor
N
%
2	3.5
5	7.5
1	4.7
Violators
Major
N	%
1	2.9
7	11.5
10	17.9
13	19.4
21	44.7
11	52.4
63
Total
N	%
1	2.9
7	11.5
12	21.4
18	26.9
21	44.7
12	57.1
71
Non V iolators	% who violated in Reformatory Operating Group of 325 Cases. (See Table XLII, App. A)
N %	
33 97.1	3.2 |
39 100.0	2.3
54 88.5-	13.6
44 78.6	16.7
49 73.1	19.6
26 55.3	38.6 |
9 42.9	33.3
254	1
APPENDIX ATable XLIV
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Operating Group of 272 Cases. Distributions based omthe_ average violation rate for the Prison (24.7%). (Burgess method on the 3 7 highest pre-parole factors)
Points for Number of factors having violation rates lower than the		Experience and Expectancy Re Parole Violation		
		Violators		
	Number			
average	nf			
rate for	U1			
the	men	Minor	Major	Total
institu-	in			
tion	each group	N . %	N %	N %
15-17	15			
13-14	26			
11-12	38			
9-10	50	2 4.0	7 14.0	9 18.0
7-8	62	5 8.0	12 19.4	17 27.4
5-6	54	4 7.4	17 31.5	21 38.9
2-4	27	3 11.1	15 55.6	18 66.7
Totals	272	14	51	65
Non	
Violators	
N	%
15	100.0
26	100.0
38	100.0
41	82.0
45	72.6
33	61.1
9	33.3
207	
% who violated in Prison Operating Group of 272 Cases. (See Table XLV, App. A)
22.6
27.3
38.8
76.7
APPENDIX A	115Table" XLV
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Control Group of 270 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on the 17 highest pre-parole factors)
Experience and Expectancy Rate of Parole Violation			
Violators			Non Violators
Minor	Major	Total	
N %	N %	N %	N %
2 3.7 5 9.1 4 8.2 8 26.7	10 18.9 10 18.2 15 30.6 15 50.0	12 22.6 15 27.3 19 38.8 23 76.7	18 100.0 26 100.0 39	100.0 41 77.4 40	72.7 30 61.2 7 23.3
19	50	69	201
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for
the
institu-
tion
15-17
13-14
11-12
9-10
7-8
5-6
2-4
Totals
Number
of
men
in
each
group
18
26
39
53
55
49
30
270
% who violated in Prison Control Group of 270 Cases. (See Table VLV, App. A)
18.0
27.4
38.9
66.7
116	APPENDIX ATable XLVI
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27: Operating Group of 325 Cases. Distributions based on the average violation rate for the Reformatory (19.4%). (Burgess method on the 17 highest pre-parole factors)
Points for Number of factors having violation rates		
lower than the	Number
average	of
rate for	men
the	in
institu-	each
tion	group
15-17	2
13-14	21
11-12	37
9-10	78
7-8	78
5-6	54
2-4	" 55
Totals	325
Experience and Expectancy Rate of Parole Violation
Violators
Minor
N %
1	1.3
3	3.9
5	9.2
6	10.9
Major
N %
1	4.8
1	2.7
5	6.4
10 12.8
9	16.7
14	25.5
Total
N
1	4.8
1	2.7
6	7.7
13	16.7
14	25.9
20	36.4
Non
Violators
N %
2 100.0
20	95.2
36	97.3
72	92.3
65	83.3
40	74.1
35	63.6
% who violated parole in Reformatory Control Group of 325 Cases. (See Table XLVII, App. A)
9.7
16.2
29.4
54.0
15
40
55
270
APPENDIX A	117Table XLVII
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27: Control Group of 325 Cases. Distributions based on the average violation rate for the Reformatory (19.4%). (Burgess method on the 17 highest pre-parole factors)
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for
the
institu-
tion
15-17
13-14
11-12
9-10
7-8
5-6
2-4
Totals
	Experience and Expectancy Rate of			
		Parole Violation		
Number	Violators			
				
of				Non
men in	Minor	Major	Total	Violators
each				
group	N %	N %	N %	N %
2				2 100.0
25				25 100.0
37				37 100.0
62	1 1.6	5 8.1	6 9.7	56 90.3
68	2 3.0	9 13.2	11 16.2	57 83.8 '
68	4 5.9	16 23.5	20 29.4	48 70.6
63	1 1.6	33 52.4	34 54.0	29 46.0
325	8	63	71	254
% who violated parole in Reformatory Operating Group of 325 Cases. (See Table XLVI, App. A)
4.8
2.7
7.7 16.7 25.9 36.4
118	APPENDIX AAPPENDIX B
Experience and expectancy rate of parole violation tables based on the 17 highest pre-parole factors, using the Glueck method of scoring.
119Table XLVIII
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Operating Group of 272 Cases. Distributions based on violation rate percentages for the Prison and Reformatory group. (Glueck method on the 17 highest pre-parole factors)
Total Failure Score (Sum of percentages 3f Violations under each category)	Number of men in each group			Outcome		on Parole			
		Violators						Non Violators	
		Minor		Major		Total			
		N	%	N	%	N	% I	N	%
211-260	15						1	15	100.0
261-310	35							35	100.0
311-360	74	1	1.4	7	9.4	8	10.8	66	89.2
361-410	62	2	2.8	9	14.5	11	17.3	51	82.3
411-460	60	9	15.0	21	35.0	30	50.0	30	50.0
461 and over	26	2	7.7	14	53.8	16	61.5	10	38.5
Totals	272	14 1		51		65		207	
120	APPENDIX BTable XLIX
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27: Operating Group of 325 Cases. Distributions based on violation rate percentages for the Prison and Reformatory Group. (Glueck method on the 17 highest pre-parole factors)
Total Failure Score (Sum of percentages of Violations under each category)	Number of cases in each group			Outcome		on Parole			
		! Violators						Non Violators	
		Minor		Major		Total			
		N		N	%	N	c/o	N	%
211-260	8							8	100.0
261-310	54							54	100.0
311-360	91	1	l.i	7	7.7	8	8.8	83	91.2
361-410	94	7	7.4	15	16.0	22	23.4	72	76.6
411-460	59	4	6.8	12	20.3	16	27.1	43	72.9
461 and over	19	3	15.8	6	31.6	9	47.4	10	52.6
Totals	325	15		40		55		270	
APPENDIX B	121APPENDIX C
Prediction tables based on the 17 lowest pre-parole factors, using the Burgess method of scoring.
123Table L
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Operating Group of 272 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on the 17 lowest pre-parole factors)
Points for	
Number of	
factors	
having	
violation	
rates	
lower	
than the	Number
average	of
rate for	men
the	in
institu-	each
tion	group
10 and over	33
9
8
7
6
5
4 or less Totals
30 50 46 46 36
31
272
Experience and Expectancy Rate of Parole Violation
		Violators			f " ;
Minor		Major		Total	
N	%	N	%	N	%
		1	3.0	1	3.0
1	3.3	• 2	6.7	3	10.0
1	2.0	8	16.0	9	18.0
2	4.3	13	28.3	15;:	3-2.6
2	4.4	10	21.7	12	26.1
3	8.4	7	19.4	10	27.8
5	16.1	10	32.3	15	48.4
14		51		65	
Non
Violators
N
%
32 • 27 41 31 34 26 16
207
97.0
90.0
82.0 67.4 73.9 72.2 51.6
% who violate parole in Prison Control Group of 270 Cases. (See Table LI, App. C)
2.7
3.1
29.4 13.7
29.5
40.5 62.9
124	APPENDIX CTable LI
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Control Group of 270 Cases. Distributions based on the average violation rate for the Prison (24.7.%) (Burgess method on the 17 lowest pre-parole factors)
Points for Number of factors having violation rates lower than the	i ! ! Number
average	of
rate for	men
the	in
institu-	each
tion	group
10 and over	37
9	32
8	34
7	51
6	44
5	37
4 or less	35
Totals	270
Experience and Expectancy Rate of Parole Violation
Minor
Violators
N	%
1	3.1
4	11.8
2	3.9
2	4.5
5	13.5
5	14.3
19	
Major	
N	%
1	2.7
6	17.6
5	9.8
11	25.0
10	27.0
17	48.6
50	
Total	
N	%
1	2.7
1	3.1
10	29.4
7	13.7
13	29.5
15	40.5
22	62.9
69	
Non Violators	
N	%
36	97.3
31	96.9
24	70.6
44	86.3
31	70.5
22	59.5
13	37.1
201
% who violate parole in Prison Operating Group of 272 Cases. (See Table L, App. C)
3.0
10.0
18.0
32.6
26.1
27.8
48.4
APPENDIX C	125APPENDIX D
Prediction tables based on 26 selected pre-parole factors, using the Burgess method of scoring.
127Table LII
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Operating Group of 272 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on 25 selected pre-parole factors)
Points for Number of factors having vi ol ati on	!	|l Experience and Expectancy Rate of Parole Violation			
rates					
lower			Violators		
than the	Number				
average rate for	of				Non
the	men in	Minor	Major	Total	Violators
					
institu-	each				
tion	group	N %'	N %	N %	N %
21 and cvei	6				6 100.0
19-20	15				15 100.0
17-18	20				20 100.0
15-16	30				30 100.0
13-14	31		2 6.5	2 6.5	29 93.5
11-12	47	1 2.1	3 6.4	4 8.5	43 91.5
9-10	36	2 5.6	13 36.1	15 41.7	21 58.3
7-8	42	3 7.1	12 28.6	15 35.7	27 64.3
5-C	29	6 20.7	10 34.5	16 55.2	13 44.8
4 or less	16	2 12.5	11 68.8	13 81.3	3 18.7
Totals	272	14	51	65 1	207
% who violate parole in Prison Control Group of 270 Cases. (See Table LIU, App. C)
9.1
19.5
26.7
33.3
64.0
90.0
128	APPENDIX DTable LIII
Experience and Expectancy Rate of Parole Violation in the Minnesota State Prison, 1922-27: Control Group of 270 Cases. Distributions based on the average violation rate for the Prison (24.7%). (Burgess method on 25 selected pre-parole factors)
	Experience and Expectancy Rate of						
			Parole Violation				
		Violators					
						Non	
Minor		Major		Total		Violators	
N	%	N	%	N	%	N	%
						9	100.0
						10	100.0
						18	100.0
						33	100.0
		3	9.1	3	9.1	30	90.9
2	4.9	6	14.6	8	19.5	33	80.5
3	6.7	9	20.0	12	26.7	33	73.3
3	8.3	9	25.0	12	33.3	24	66.7
6	24.0	10	40.0	16	64.0	9	36.0
5	25.0	13	65.0	18	90.0	2	10.0
19		50		69		201	
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for
the
institu-
tion
21 and ovei 19-20 17-18 15-16 13-14 11-12 9-10 7-8 5-6
4 or less Totals
Number
of
men
in
each
group
9
10
18
33
33
41
45
36
25
20
270
% who violate parole in Prison Operating Group of 272 Cases. (See Table LII, App. C)
6.5
8.5
41.7
35.7
55.2
81.3
APPENDIX D	129Table LÏV
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27Operating Group of 325; Gases. Distributions based on the average violation rate for the Reformatory (19.4%). (Burgess method on 25 selected pre-parole factors)
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for '
the
institu-
tion
21 and over 19-20 17-18 15-16 13-14 11-12 9-10 7-8
6 or less
Totals
Experience and Expectancy Rate of Parole Violation
Number
of
men
in
each
group
9
18
32
54
57
57
46
31
21
325
Minor N %
1-	1.9
2	3.5
3	5.3
3	6.5
2	6.5
4	19.1
15
Violators				Non Violators	
Major		Total			
N	%	N	% •	.N	%
				9	100.0
				18	100.0
1	3.1	1	3.1	31	96.9
2	3.7	3	5.6	51	"94.4
7.	12.3	9	15.8	48	84.2
8	14.0	11	19.3	46	80.7
• 5	10.9	8	17.4	. 38	82.6
9	29.0	11	35.5	20	64.5
8	38.1	12	•57.2	9	42.8
40		55		270	
% who violate parole in Reformatory Control Group of 325 Cases. (See Table LV,
App. D)
2.6
2.2
23.1 31.4
43.2 75.6
130	APPENDIX DTable LV
Experience and Expectancy Rate of Parole Violation in the Minnesota State Reformatory, 1922-27: Control Group of 325 Cases. Distributions based on the average violation rate for the Reformatory (19.4%). (Burgess method on 25 selected pre-parole factors)
Points for
Number of
factors
having
violation
rates
lower
than the
average
rate for
the
institu-
tion
21 and ovei 19-20 17-18 15-16 13-14 11-12
9-10
7-8
6 or less Totals
Number . of men in
each
group
8
19
27
39
47
65
51
44
25
325
	Experience and Expectancy Rate of						
			Parole Violation				
Violators							
						Non	
Minor		Major		Total		Violators	
N	%	N	' % :	N	%	N	%
						8	100.0
						19	100.0
						27	100.0
		1	2.6	1	2.6	38	97.4
		1	2.2	1	2.2	46	97.8
3	4.6	12	18.5	15	23.1	50	76.9
2	3.9	14	27.5	16	31.4	35	68.6
2	4.6	17	38.6	19	43.2	25	56.8
1	3.6	18	72.0	19	75.6	6	24.4
8		63		71		254	
% who violated parole in Reformatory Operating Group of 325 Cases. (See Table LIV,
App. D)
3.1
5.6
15.8
19.3
17.4
35.5 57.2
APPENDIX DAPPENDIX E
Do Parole Prediction Tables Work in Practice?1
During the past few years several studies have been made which have sought to develop methods for predicting the probable conduct of men released on parole. Especially significant are the pioneer studies of Professors Burgess 2 and Glueck. 3 Last year the writer completed a similar study of the Minnesota parole records. 4 The results obtained were highly comparable to those of the earlier studies. Inmates whose score placed them in classes at the upper end of the scoring scale were 100 per cent non-violators; those at the lower extreme violated in over 86 per cent of the cases.
It has been customary to take these actual violation rate percentages—experience tables, as it were—as measures of the expected future violation rates, other things remaining the same by assumption. The experience tables of the past thus become the expectancy tables of the future. This, however, involves the necessity of predicting outcome in the same group of cases on which the experience tables have been constructed, which becomes simply another form of the age-old fallacy of reasoning in a circle. The question of whether the experience tables of the past actually give á measure of the parole violation that may be expected in the future must be answered by more crucial tests than by merely assuming that other things will remain unchanged. This is the problem of the present project.
The workability of parole prediction has been approached from three angles: (a) by applying tables worked out on a group paroled before July 1st, 1927, to a group paroled from the same institution since that date; (b) by employing one of the clerks in the Minnesota Parole Department to score the records of men paroled since July 1st,
1	A paper read before the Division of Social Research at the Annual Meeting of the American Sociological Society, Cleveland, Ohio, December 29-31, 1930. Reprinted with permission from the Publication of the American Sociological Society, Vol. XXV, No. 2 May 1931. Sections b and c of the project sketched in this paper have now (July 1931) reached the stage where most of the essential research has been completed. As soon as computations and interpretations can be made publication as a separate paper is expected.
2	E. W. Burgess, “Factors Determining Success or Failure on Parole,in Bruce, Harmo, and Burgess, Parole and the Intermediate Sentence (Illinois Parole Board, 1928), pp. 205-69.
3	Sheldon Glueck and Elinor T. Glueck, 500 Criminal Careers (New York: Alfred A. Knopf, 1930).
4	G. B. Void, Prediction Methods and Parole (Hanover: Sociological Press, 1931).
133134
APPENDIX E
TABLE I
Comparison of Actual Parole Violation in a Group of 282 Cases Paroled From the Minnesota State Prison 1927-29 and the Expected Violation on the Basis of Experience with a Group of 542 Cases Paroled From the Same Institution 1922-27
Point Scale	Number
Using the	of Men
Burgess	in Each
Method of	Score
Scoring	Class
21 and over	9
19-20	21
17-18	22
15-16	36
13-14	40
11-12	47
9-10	42
7-8	33
5-6	24
4 or less	8
Totals 282
Number and Percentage Expected to Violate Parole on Basis of Experience with 1922-27 Group of 542 Cases
No.	Per Cent
3	7.8
7	13.6
14	33.3
12	34.6
14	59.3
7 57	86.1
Number and Percentage in 1927-29 Group of 282 Cases Who Actually Did Violate Parole No. Per Cent
2	5.6
4	10.0
6	12.8
11	26.2
16	48.5
17	70.8
7 63	87.5
1927, using the schedules worked out by the writer on the earlier group, to see if the same predictions can be made as when the writer does1 the scoring; (c) by having one of the parole agents make a subjective “common sense” rating (on a four-place scale) of probable outcome of the current parole cases at the time the parole is voted, which eases are then scored according to the schedules already worked out and comparisons made between the predicted outcome and the agent’s rating. The predictions thus made, as well as the agent’s rating, can then later be compared with the actual outcome on parole. The present paper is a preliminary report on the first phase of this project.
To date, the records of 282 men, paroled from the Minnesota State Prison between July 1st, 1927, and July 1st, 1929, have been scored on the basis of the same schedules that were used in constructing the Experience-Expectancy tables for the earlier group of 542 cases paroled from the same institution before July 1st, 1927. An attempt has thus been made to determine for each man his chance of parole violation, judged by the experience with men similarly classified in the group paroled before July 1st, 1927. This is in effect doing what the Parole Board might have done had it made use of such prediction tables at the time the parole was granted. The detailed results in Table I.
The violation rate for the 1922-27' group of 542 cases was 24.7 per cent. If the same rate were to hold for the, latter group of 282 cases, one should expect 70 violators of parole. Using the experience table for the 1922-27 group, it is possible to predict 57 violators in the group of 282 cases—an error of 13 cases. Out of the total group of 282 cases this represents a 4.6 per cent error. The actual experience violation rate, however, in the 1927-29 group of 282 cases is 22.3 per cent (or 63 cases) and not 24.7 per cent. The difference between 63, the numberAPPENDIX E
135
actually violating, and 57, the number expected to violate, is 6, or a 2.1 per cent error. This would seem to indicate that the principal error that, appears is due to the changed rate of parole violation for the institution. Parole prediction seems to have worked within the limits of about a 2 per cent error.
Care must be taken not to over-emphasize the results from this limited number of cases. The general trend would seem to point, however, to the conclusion that reasonable accuracy may be expected in appling prediction tables to actual parole practice. If further research should bear this out, it would seem that application of prediction techniques should be among the next important developments in the administration of parole.LIST OF TITLES CITED
Biennial Report of the Minnesota State Board of Parole, 1911-1928.
Biennial Report of the Minnesota State Reformatory,
1921-1928.
Biennial Reports of the Minnesota State Prison, 1921-1928.
Brockway, Z., “The Ideal of a True Prison System for a State,” Twenty-sixth Annual Report of the Executive Committee of the Prison Association of New York, 1870.
Bruce, A. A., Harno, A. J., Burgess, E. W., and Landesco, J., Parole and the Indeterminate Sentence. Illinois State Board of Parole 1928. xiv-277.
Butler, A. W., “What Prisoners Should Be Eligible to Parole and What Considerations Should Govern the »Granting of It,” Journal of Criminal Law and Criminology, 12: 549-553, February 1922.
Clark, W. W., “Success Record of Delinquent Boys in Relation to Intelligence,” Journal of Delinquency, 5: 174-182, September, 1920.
Clark, W. W., “Supervised Conduct Responses of Delinquent Boys,” Journal of Delinquency, 6: 387-401, May, 1921.
Doll, E. A., “Classification of Prisoners for Purposes of Training Work and Parole,” Journal of Criminal Law and Criminology, 14: 110-116, May, 1923.
Edgerton, H. A., and Paterson, D. G., “Table of Standard Errors and Probable Errors of Percentages for Varying Numbers of Cases,” Journal of Applied Psychology, 10: 387-391, No. 3, September, 1926.
Garrett, H. E., Statistics in Psychology and Education. Longmans Green and Co., New York, 1926, xiii-317.
Gault, R H., “The Parole System, A Means of Protection,” Journal of Criminal Law and Criminology, 5: 799-806, March, 1915.
General Statutes of Minnesota, Sections 9273-9279.
Glueck, S. H., and Eleanor T., 500 Criminal Careers. Alfred A. Knopf, New York, 1930. xxvii-365-xvi.
Hart, Hornell, “Predicting Parole Success,” Journal of Criminal Law and Criminology, 14: 405-413, November, 1923.
Heacox, F. L., “A Study of One Year's Parole Violators Returned to Auburn Prison,” Journal of Criminal Law and Criminology, 7: 233-258, July, 1917.
Knox, E. E., “Follow-up of 103 Whittier State School Boys Ten Years after Admission,” Journal of Delinquency, 12: 261-269, September-December, 1928.
137New York:	Prison Survey Committee Report, Albany, New York,
1920.
Pearson, K., “On the Theory of Contingency and Its Relation to Association and Normal Correlation,” Drapers’ Company Research Memoirs, Biometric Series i, Dulau and Co., London, 1904.
Pintner, R., and Reamer, J. C., “Mental Ability and Future Success of Delinquent Girls,” Journal of Delinquency, 3: 74-89, March, 1918.
Spalding, W. F., “Parole and Probation, A Symposium,” Proceedings of the American Prison Association, 1916.
Warner, S. B., “Factors Determining Parole From the Massachusetts Reformatory,” (Preface by Sanford Bates, Commissioner of Correction for Massachusetts). Journal of Criminal Law and Criminology, 14: 172-207, August, 1923.
Yule, G. U., An Introduction to the Theory of Statistics. Charles Griffin and Co., London, 1927. Eighth Edition, xv-422.
138