College and Research Libraries Research Notes Analyzing Library Survey Data Using Factor Analysis Linda L. Phillips and William Lyons Factor analysis was used to investigate the di- mensions of faculty attitudes about library poli- cies. Initially developed in the context of psy- chology, factor analysis attempts to place variables in groups. This statistical technique is well suited for managing large data sets such as those collected in a survey. Factor analysis of nineteen policy questions from a library survey at The University of Tennessee, Knoxville, re- vealed that faculty think of library policies in four dimensions-computerized access, circu- lation, specialized collection/services, and gen- eral collection. A characterization of each cate- gory is followed by suggested applications. The purpose of this paper is to illustrate the use of factor analysis to analyze and in- terpret large data sets. Factor analysis is particularly well suited for the examination of library survey data because surveys typi- cally generate a great amount of raw data that must be interpreted before meaningful conclusions can be reached. Unfortu- nately, reports of survey research are fre- quently simplistic and descriptive because the raw data elude interpretation. Appro- priately managed survey data, however, reveal relationships among variables, ex- plain differences among responses, and permit understanding of apparent conflict. Factor analytic procedures attempt to place variables in groups. The groups are named, permitting variables in the group to be considered together. In effect, group- ing of the data makes interpretation possi- ble. For example, in a survey that contained thirty variables, factor analysis might place the variables into three sets. Subsequently, the researcher would draw relationships and reach conclusions by considering the three sets, rather than by comparing each of the thirty variables with all of the others in the survey. A review article by F. William Summers concludes that surveys are "clearly the oldest and most enduring method of re- search on libraries. " 1 However, little pub- lished evidence indicates that factor analy- sis has been used to interpret library survey data. Michael Halperin and Mau- reen Strazdon described the application of a similar technique, conjoint analysis, to measure students' preferences for refer- ence service. 2 Like factor analysis, con- joint analysis places variables in catego- ries. However, the procedures are based on designation of categories prior to ad- ministration of the survey. One of the ad- vantages of factor analysis is that the cate- gories or groups emerge through the application of the statistical program. Re- lationships that are identified among vari- ables may differ remarkably from an origi- nal hypothesis. Linda L. Phillips is Associate Professor and Head, Cooperative Information Services, John C. Hodges Library, and William Lyons is Professor of Political Science and Director of Survey Research in the Social Science Research Institute at The University of Tennessee, Knoxville. 483 484 College & Research Libraries William McGrath et al. applied factor analysis to a formula for determining li- brary collection allocations to academic departments. 3 Twenty-two variables rep- resented questions that should be ad- dressed when developing an allocation formula. Data on each variable were ana- lyzed first by multiple correlation and then by factor analysis. The data were re- duced to three groups or factors. Al- though names had been assigned to the three factors prior to analysis, the results showed that the variables that came to- gether in each of the factor categories did not fit the preselected names chosen by the researchers. Factor analysis drew at- tention to relationships that had not been considered previously. The technique permitted creation of the allocation for- mula using three factors, where previ- ously there would have been twenty-two. BACKGROUND The University of Tennessee, Knoxville, has received national attention for its lead- ership in the area of assessment of educa- tional outcomes. One UTK evaluative pro- gram, the Academic Program Review, is sponsored by the Office of the Provost and has three major components. First, the de- partment being reviewed prepares a de- scriptive self-study. Second, a review team composed of internal and external reviewers visits the department. Third, a series of reports are written by the review team, the department head, and the uni- versity administration. The reports are used to evaluate the department, to estab- lish departmental goals, and to provide a foundation for subsequent strategic plan- ning efforts. In spring 1987, the university library participated in its first program review. As a supplement to the usual review compo- nents, a survey of faculty opinion about the library was conducted. Information about faculty library use, evaluation of the collections and physical facilities, atti- tudes toward services and policies, and assessment of library priorities were in- tended to contribute to the program re- vieW and to be used as a basis for plan- ning. The survey was designed and adminis- September 1990 tered by one of the authors, a social scien- tist with extensive survey research experi- ence. A product of discussions among the faculty senate library committee, the li- brary administration, and campus admin- istration, the instrument was drafted in re- sponse to a list of topics solicited from the faculty senate library committee. Follow- ing the initial draft, the library administra- tion added topics and gave advice on the organization of the questionnaire. The questionnaire addressed a number of points. In addition to overall evaluation of the library, specific judgments were sought regarding the quality of the collec- tion, facilities, and various services pro- vided by the library staff. The instrument also solicited preferences for a range of ac- tual or desired library policies. Finally, a variety of background measures was ob- tained for each respondent to facilitate comparison of the results across colleges and divisions and by faculty rank. Every faculty member and graduate teaching assistant at UTK was sent a com- prehensive survey instrument in April 1987. Sponsored by the UTK Office of In- stitutional Research, the survey was sent in three waves, yielding a response rate of over 75 percent. Of 1,690 questionnaires distributed, 932 were returned by faculty (76 percent) and 343 were from graduate teaching assistants (74 percent). The sur- vey produced considerable data-over 1,200 responses to approximately 140 questions. 4 The authors chose factor anal- ysis as a relevant technique to identify re- lationships among nineteen of the ques- tions that asked faculty to select among priorities regarding various library ser- vices and collections. The following dis- cussion demonstrates how factor analysis may be applied in the interpretation of re- sponses to policy preference questions. Because there were no significant differ- ences between faculty and graduate teach- ing assistant responses on this portion, re- sponses of both groups are referred to by the generic term faculty. METHOD Respondents were asked nineteen ques- tions concerning library policies and prior- ities. (Policy and priority questions from the survey are shown in appendix A). The policy questions followed a Likert format, in which faculty were asked whether they agreed or disagreed with various state- ments. For eleven possible "Improve- ments" to library collections or services, each person was asked to assign priorities, ranging from "not a priority" to "one of the highest priorities." The investigators had particular interest in determining fac- ulty preferences in reduced terms, that is, to find the underlying dimensions of pol- icy preference. Factor analysis works well for such a data reduction task. Basically, factor ana- lytic procedures try to "fit" a matrix of correlations of variables into the smallest number of dimensions. The analysis alwa- sys starts with the same number of dimen- sions as variables. The number of dimen- sions is gradually reduced until the amount of variance left unexplained amount the correlation matrix exceeds ac- ceptability. At that point new composite variables may be created (one for each of the reamining dimensions), and existing variables are correlated with these new composite variables. These new compos- ite variables, or factors, derive their iden- tity from the context of the variables with which they are strongly correlated. Thes correlations are often called factor load- ings. To reveal the multiple dimensions pro- duced by factor analysis, factor loadings · are rotated according to various criteria, the most common of which is a V arimax Rotation. The rotation procedure groups variables that tend to correlate highly with some factors and considers them together. This allows factors to be defined in terms of the variables with which they are highly correlated. In most rotation procedures, the rotated factors are treated as though they were statistically independent. Sim- ply put, this independence allows the ana- lyst to deal with factors that do not statisti- cally overlap with other factors. As factor analysis was initially devel- oped in the context of psychology, the fol- lowing hypothetical situation shows an example of its application. Suppose one is analyzing a number of variables dealing with creativity and finds a distinct, two- Research Notes 485 factor solution. What conclusion could be drawn? The first factor is characterized by high correlation among variables measur- ing artistic ability, and the second factor is a composite of measures tapping abilities in abstract reasoning. We could infer that creativity is two dimensional and that any discussion implying only one dimension would overlook the true nature of the larger concept. FINDINGS Respondents' answers to nineteen questions were subjected to factor analy- sis. The criterion used by the package em- ployed for the analysis, SPSSPC (Kaiser's Criterion) selected a six-factor solution (ei- genvalue= 1). However, we decided tore- linquish a small amount of explanatory power for the advantage of dealing with fewer dimensions and further reduced the number of factors to four. 11There is always some aspect of the variance among all respondents that is specific to a variable and some re- maining variance that represents the underlying dimension of thinking.'' The four factors explained 46 percent of the variance among the policy variables. This means that slightly less than half of the variance among the variables is ''com- mon variance,'' subject to explanation by the four-factor solution. Conversely, slightly more than half of the variance among the variables is "unique variance" not accounted for by the solution. This di- vision of variance into common and unique is important. The logic of the divi- sion is as follows. Even though the re- sponses to a question about, say, the de- sirability of increasing library hours taps what users may think specifically about hours (unique), it also may well reflect some broader, underlying dimension of respondent thinking, such as overall feel- ing about a more accessible library (com- mon). There is always some aspect of the variance among all respondents that is . specific to a variable and some remaining 486 College & Research Libraries variance that represents the underlying dimension of thinking. Thus, by reducing the numbers of dimensions from nineteen to four, we simplify our consideration of that set of questions. This simplification is at the expense of working with a solution that captures only about half of the varia- tion with which we started. When we deal with the common proportion of the vari- ance, we lose the unique variance that is captured by the individual variables. However, the underlying assumption in- herent in the decision to work with the composite factor rather than the unique variable is that the underlying dimension- ality of thinking is of paramount interest rather than the specifics of the individual question. The four-factor rotated solution is shown in table 1. The name provided to each factor reflects the underlying com- posite that summarizes those variables September 1990 with high loadings. Variables associated with each factor are grouped together in table 1 to facilitate interpretation and dis- cussion. DISCUSSION The variables with high loadings on the first factor are: 1. Agreement with the desirability of the library's maintaining access to na- tional databases 2. Higher priority status to subsidized computer literature searches 3. Higher priority status to capacity to send and receive library messages on the personal computer 4. Higher priority status to computer- ized reference service These four variables appear to tap a common dimension: computerized access to information beyond the confines of the local collection. The fact that they load on TABLEl LIBRARY POLICIES AND PRIORITIES: FOUR-FACTOR SOLUTION Computerized 5C~ll~oe; General Factors Access Circulation Collection Computerized Access Increase computerized reference services .82636 .01943 .05073 .03044 Subsidize computerized literature searches .73488 .14812 .06689 -.01884 Increase PC message capabilities .71655 .02657 .03337 .04614 More national database access * .55707 .18675 .06249 .07876 Circulation Open 24 hours* .06470 .78066 .04433 -.01405 Increase hours -.01109 .76204 .18924 .00957 Open every day* .04974 .69938 -.05723 .11997 Check out nonprint* .08587 .48072 -.05263 -.08605 Check out periodicals* .03203 .46269 -.07152 .08381 Specialized Collection Increase AIV acquisitions -.00100 -.00002 .68036 -.07796 Increase staff .04098 .02193 .66165 -.00594 Increase rare books .02514 .00645 .53434 .28309 Increase reference holdings .19651 -.07335 .48778 .33361 Increase user education .40880 -.01958 .47638 -.20624 General Collection Increase book acquisitions -.06574 .01055 .22104 .74490 Increase journal acquisitions .13433 .01551 .07171 .73245 Spend on collection* .02404 .09266 -.33818 .52575 % Explained variance 16.1% 11% 9% 7.7% Cumulatilve % of variance explained = 43.8% *Policy question: High value indicates agreement . Other questions refer to priorities, and high values indicate assignment of high priority . one factor suggests that there is common- ality among these variables and that there is statistical justification to the discussion of this commonality. In such a discussion, it is helpful to use a name for the dimen- sion that would reflect what is shared among these variables, Computerized Ac- cess. Finally, the fact that the variables can be grouped statistically suggests that sur- vey respondents tend to think of them to- gether and apart from other variables. The factor analysis, however, does not reveal anything about the degree to which re- spondents hold one variable or the other in higher or lower priority. It only indi- cates that these items vary together, i.e., that they may be placed together in the Computerized Access category. The second factor was named Circula- tion because items relating to that factor- longer hours, more days open, and more liberal check-out procedures for periodi- cals and audiovisual materials-involved access to and use of library collections. It is interesting that these items come together on a single factor, for it would not be read- ily apparent that respondents think of the hours and days of library operation along the same dimension as increased check- · out policies. The third and fourth factors tap dimen- sions primarily dealing with the collec- tion. The third was labeled Specialized Collection because it is defined by such types of materials in the collection as refer- ence sources, rare books, and audiovisual programs. Two items concerning staff and staff functions also appear on this factor- increased education of the user commu- nity and increased staff. How would one interpret this seeming anomaly? One ex- planation is that all of the variables in group three serve a specialized clientele. As faculty tend not to seek assistance in their use of information resources, the availability of user education and staff as- sistance appears to be considered a spe- cialized feature of the library. The final dimension underlying the col- lective thinking of the faculty on library policies had to do with the general collec- tion of books and journals. Obviously, re- spondents think of these two items to- gether and apart from the other collection variables. Of further note is that when the Research Notes 487 respondents think of spending on the col- lection in general, they may be thinking of the aggregate size of the collection rather than its composition. In fact, the negative loading (- .338) for this variable on the specialized collection factor suggests a weak negative relationship between the priority for spending on the entire collec- tion and the priority assigned for develop- ment of the specialized collection. The above discussion suggests five pos- sible applications. First, when library staff and/ or administrators communicate with faculty on library policies, they can con- sider the policy dimensions used by the faculty. Some policy matters might be ad- dressed collectively, in packages consis- tent with those identified in this analysis. Second, the fourfold policy categorization may be useful in assessing internal library organization. The divisions in faculty con- ''The underlying assumption inher- ent in the decision to work with the composite factor rather than the unique variable is that the underly- ing dimensionality of thinking is of paramount interest rather than the specifics of the individual ques- tion.'' ceptualization among computerized ac- cess, circulation, special collections/ser- vices, and general collection do not closely mirror present traditional organizational structures. The linking of user education and staff resources with specialized collec- tions/services provokes thought and dis- cussion about the relationship of the li- brary organization to faculty perceptions. Third, the emergence of the computer- ized access category is a gratifying sign that faculty, who are traditionally inter- ested primarily in local library holdings, may be thinking in more global terms for access to information. In fact, if the instru- ment had included questions related to in- terlibrary borrowing policy or services that had loaded highly on Factor 1, the cat- egory might have been named External Access. A fourth application of factor anal- ysis is to improve future research instru- 488 College & Research Libraries ments and methods through further exploration and specification of the under- lying dimensions of response. Variables related to externally-focused services should be included in future surveys, ei- ther to test the validity of the external ac- cess interpretation or to improve the preci- sion of the survey instrument. Finally, the factor analytic solution may be used as a point of comparison for other studies, possibly about student perceptions, at this or other institutions. As technological changes permit the delivery of innovative and unforeseen services, a replication of this study would be advantageous. CONCLUSIONS Factor analysis was a useful technique to investigate the dimensionality underlying September 1990 the attitudes of the faculty at the Univer- sity of Tennessee, Knoxville, with regard to library policies. In general, faculty think of library policies in four dimensions. These dimensions-computerized access, circulation, specialized collection/ ser- vices, and general collection-summarize faculty attitudes on nineteen policy ques- tions. The utility of factor analysis as a data- reduction strategy is well documented throughout the social sciences. As statisti- cal techniques become more prevalent in the assessment of library survey data, the use of this along with other methods, should facilitate the understanding of atti- tude structures of various library clientele. REFERENCES 1. F. William Summers, "History and Development of the Survey Model for Planning," Drexel Li- brary Quarterly 21:41 (Fall1985). 2. Michael Halperin and Maureen Strazdon, "Measuring Students' Preferences for Reference Ser- vice: A Conjoint Analysis," Library Quarterly 50:208-24 (Apr. 1980). 3. William E. McGrath,Ralph C. Huntsinger, and Gary R. Barber, "An Allocation Formula Derived from a Factor Analysis of Academic Departments," College & Research Libraries 30:51-62 Oan. 1969). 4. William Lyons and Linda Phillips, Library Survey Summary (Bethesda, Md.: ERIC Document Re- production Service, 1989; ED 297 742, IR 052 414). APPENDIX A. LIBRARY POLICY QUESTIONS FROM UTK LIBRARY SURVEY Libraries at UTK: A Survey of Faculty Opinion Please circle the number which corresponds to the response which best reflects how you feel. Space for additional comments is provided at the end of the survey. 17. In this section, we would like to know how you feel about various library policies. [In the actual instrument the following scale was placed to the right of each statement.] STRONGLY STRONGLY DISAGREE DISAGREE NEUTRAL AGREE AGREE 1 2 3 4 5 a. The library should spend any available money on its collection, even if service declined. b. When acquisition funds are limited, the library should purchase books "on approval." c. It is very important that the library increase the accessibility of national databases. d. The main library should be open twenty-four hours a day. e. The main library should be open every day, even when classes are not in session. f. Periodicals should be allowed to be checked out. g. Nonprint materials should be allowed to be checked out. h. The library should provide a pickup and delivery service to academic units. Research Notes 489 18. Below is a list of improvements that have been suggested for the library. Given the fact that funds are limited, please indicate the priority status that you feel should be assigned to each. [In the actual instrument the following scale was placed to the right of each statement.] NOTA PRIORITY 1 a. Increased hours ALOW PRIORITY 2 b. Increased book acquisitions c. Increased journal subscriptions d. Increased staff e. Increased audiovisual acquisitions A HIGH PRIORITY 3 f. Increased manuscript and rare book acquisitions g. Increased reference holdings h. Increased education of the user community i. Subsidized computer literature searches j. Ability to send messages to and from library via personal computer k. Increased access to computerized reference services ONEOFTHE HIGHEST PRIORITIES 4 Introducing the New Standard in Research Indexes Composite Index for CRC Handbooks 3rd Edition Now, consult a single index containing 600,000 entries to locate data from over 140,000 pages of scientific information . Introducing the Composite Index for CRC Handbooks, 3rd Edition, a merged, unified three-volume index representing over 300 CRC Handbook volumes. Never before has such a comprehen- sive and useful reference resource been offered to the scientific community. CRC Handbooks include topics from the fields of biomedical sciences, biology, chemistry, engineering and computer science, physics and mathematics. By including subjects, compounds, and names of organisms, the Composite Index helps you locate information within these topics quickly, saving you valuable research time . COw ROM The entire Composite Index has been transferred to a CD-ROM disk that is included with the printed index volumes at no additional charge. The CD-ROM's search program offers superior search capabilities that make finding information in the Composite Index faster, easier, and more efficient than you ever thought possible. Electronic bookmarks, on-line program help, customization options, and outputcapabilityare only a few of the added features included with the CD-ROM. Menu-driven and designed to run on IBM PC/XT/ AT, PS/2 or 100% IBM-compatible computers, the CD-ROM version of the Composite Index is simple to use and maintain. One .. Day FAX Page Delivery Service Need an article from a CRC Handbook fast? CRC Press's One-Day FAX Service puts Handbook information in your hands within 24 hours. This unique service provides libraries with access to every CRC Handbook- over 140,000 pages of scientific data-without purchasing all of the Handbooks. The cost per FAX request is a mere $5 .00 for the first two pages and $1 .50 for every page thereafter (telecommunication charges included). Your first year's annual user's license fee is included in the initial purchase price. Obtaining research information has never been so easy! Annual Supplements To keep the Composite Index in pace with the rate of scientific developments, a yearly supplement is available on a subscription basis. This supplement will consist of a 200-300 page hardbound volume, a CD-ROM disk containing the entire updated Composite Index, as well as a renewal of the user's license for the One-Day FAX Service. Whether you are a scientist or student, librarian or engineer, the Composite Index for CRC Handbooks, 3rd Edition provides fast, easy access to information from CRC Handbooks. Libraries should consider the Composite Index for CRC Handbooks, 3rd Edition an indispensable reference acquisition. Pre·Publication Special SAVE $2001 Order now through December 15, 1990 and receive $200.00 off with our special pre-publication price of $995 .00- Catalog No. 284ZV. • - • CRC Press, Inc., 2000 Corporate Blvd ., N .W., Boca Raton, Fl 33431 TO ORDER: Call toll free 1-800-272-7737 Monday through Friday © IBM is a registered trademark of International Business Machines (Continental U.S. only) . In Florida, call collect 407-994-0563.