psp57030416.tif Journal of Personality and Social Psychology Copyright 1989 by the American Psychological Association, Inc. 1989, Vol. 57, No. 3, 416--425 0 0 2 2 - 3 5 1 4 / 8 9 / $ 0 0 . 7 5 Aging and Susceptibility to Attitude Change Jon A. Krosnick Duane E Alwin O h i o State University University o f Michigan Two hypotheses about the relation between age and susceptibility to attitude change were tested. The impressionable years hypothesis proposes that individuals are highly susceptible to attitude change during late adolescence and early adulthood and that susceptibility drops precipitously immediately thereafter and remains low throughout the rest of the life cycle. The increasing persistence hypothesis proposes that people become gradually more resistant to change throughout their lives. Structural equation models were applied to data from the 1956-1960, 1972-1976, and 1980 National Election Panel Studies in order to estimate the stability of political attitudes and unreliability in measures of them. The results support the impressionable years hypothesis and disconfirm the increasing persistence hypothesis. A decrease in the over-time consistency of attitude reports among 66- to 83- year-olds was found to be due to increased random measurement error in their reports, not to in- creased attitude change. A g r e a t deal o f research has e x p l o r e d individual difference variables t h a t d e t e r m i n e the susceptibility o f attitudes to change. For e x a m p l e , recent meta-analyses have shown t h a t women a r e m o r e easily influenced t h a n m e n u n d e r certain cir- c u m s t a n c e s (Cooper, 1979; Eagly & Carli, 1981). Other re- search has f o u n d relations between personality and resistance to a t t i t u d e change (Hovland & Janis, 1959; Newcomb, 1943). A n d still other research indicates t h a t attitudes t h a t are m o r e central or i m p o r t a n t to individuals are m o r e resistant to change t h a n are n o n c e n t r a l o r u n i m p o r t a n t attitudes (Krosnick, 1988). I n this article, we explore a n o t h e r possible d e t e r m i n a n t : age. It is widely believed t h a t susceptibility to a t t i t u d e change varies as people progress t h r o u g h the life cycle. Two p r i n c i p a l h y p o t h - eses have b e e n proposed, b o t h o f which assert t h a t susceptibility to a t t i t u d e change is greatest d u r i n g adolescence a n d early a d u l t h o o d and decreases thereafter. I n this article, we evaluate these hypotheses in two studies o f political attitudes. We begin b y describing the two hypotheses a n d highlighting the short- comings o f the l i m i t e d existing evidence testing them. We then r e p o r t new evidence, using i m p r o v e d d a t a and analytic methods. This research was supported by Grant 2-R01-AG04743-03 from the National Institute on Aging. Portions of this article were presented at the 1988 annual meeting of the American Association for Public Opin- ion Research in Toronto, Canada. We acknowledge the Inter-University Consortium for Political and Social Research for providing access to the National Election Study data analyzed in this article, and we thank Frank Mierzwa and Lynn Diel- man for conducting the data analysis. We also thank David Sears, Philip Converse, and Richard Petty for very helpful comments and sugges- tions. Correspondence concerning this article should be addressed to Jon A. Krosniek, Department of Psychology, Ohio State University, Colum- bus, Ohio 43210, or to Duane E Alwin, Institute for Social Research, University of Michigan, 426 Thompson Street, Ann Arbor, Michigan 48106. T w o P e r s p e c t i v e s o n A g e a n d A t t i t u d e C h a n g e There are two m a j o r perspectives on the relation between age and a t t i t u d e flexibility (see Alwin, Cohen, & Newcomb, in press). According to the impressionable years hypothesis, the socializing influences individuals experience when they are young have a p r o f o u n d i m p a c t on their t h i n k i n g t h r o u g h o u t their lives (Cutler, 1974; Dennis, 1973; Easton & Dennis, 1969; Greenstein, 1965; Hess & Torney, 1967; Sears, 1975, 1981, 1983). The historical e n v i r o n m e n t in which a young person be- comes an active p a r t i c i p a n t in the a d u l t world shapes the basic values, attitudes, and world views f o r m e d d u r i n g those years. O n c e the p e r i o d o f early socialization has passed, this hypothe- sis argues, its residuals are fixed within individuals, and these core orientations are unlikely to change.I As a result o f this process, m e m b e r s o f each b i r t h c o h o r t are p r e s u m a b l y likely to r e s e m b l e one a n o t h e r closely in t e r m s o f their basic attitudes ( M a n n h e i m , 1952). F u r t h e r m o r e , differ- ences between b i r t h cohorts in t e r m s o f the economic, social, and political conditions and the formal socialization practices they experience p r o d u c e differences between cohorts in t e r m s o f attitudes (Cutler, 1974, p. 441). For e x a m p l e , increases in the rate o f formal education in the U n i t e d States have provided new b i r t h cohorts with different socializing experiences from those o f preceding generations. C o h o r t r e p l a c e m e n t may be responsi- ble for recent changes in a variety o f the A m e r i c a n p u b l i c ' s so- cial and political attitudes. Cohorts who have had one sort o f early socialization experience die off as cohorts with different socialization experiences enter a d u l t h o o d ( M a n n h e i m , 1952). A second view o f the relation between age and susceptibility to attitude change is called the increasing persistence hypothe- There is no universal agreement on precisely which are the impres- sionable years. Some authors see political socialization as largely com- pleted by age 18; others see it as continuing up to approximately age 25. Consistent with this latter view, Newcomb and colleagues (Newcomb, 1943; Newcomb, Koenig, Hacks, & Warwick, 1967) found evidence of significant political socialization between ages 18 and 25. 416 Jon's Assistants Text Box Krosnick, J. A., & Alwin, D. F. (1989). Aging and susceptibility to attitude change. Journal of Personality and Social Psychology, 57, 416-425 AGING A N D SUSCEPTIBILITY TO ATTITUDE C H A N G E 417 sis. According to this perspective, individuals are eminently flexible and responsive to social circumstances when they are young, but as they age, their flexibility decreases gradually (Glenn, 1974, 1980; Jennings & Niemi, 1981; Markus, 1979; Sears, 1981, 1983). There are a number o f justifications for the argument that attitude flexibility decreases steadily through the life span (see, e.g., Glenn, 1980). First, because o f a decline in energy and loss o f brain tissue, capacities for information pro- cessing and memory decline as life progresses, making incorpo- ration o f new information difficult. Second, the experiences one has as one ages leave a gradually growing body o f attitude-rele- vant knowledge within the individual that serves as a source o f psychological stability (Wood, 1982). Third, aging entails social disengagement and a decrease in interest in events distant from one's immediate life (Gergen & Back, 1966; Glenn, 1969), which reduces the likelihood that attitude-challenging informa- tion will be carefully considered by an individual (Petty & Caci- oppo, 1979). Finally, a corresponding increase in social support for one's attitudes takes place as one accumulates friends who share similar social background experiences and world views as a result o f friend selection and social influence processes (New- comb, Koenig, Hacks, & Warwick, 1967). This, too, may in- crease resistance to attitude change. As a result o f these pro- cesses, people's attitudes presumably become less responsive to developments in their environments and thereby become less likely to change. According to this view, gradual changes in ag- gregated society-level attitudes reflect changes among the youn- gest members o f society in response to major national events and personal experiences. A third view o f the relation between age and susceptibility to attitude change is the life-long openness hypothesis (Sears, 1981, 1983). According to this view, individuals are highly flexible throughout their lives and constantly alter their attitudes in re- sponse to changing life circumstances (see, e.g., Brim & Kagan, 1980, p. 1). This view is not necessarily incompatible with the impressionable years and increasing persistence hypotheses be- cause susceptibility to attitude change could decrease across the life cycle, as described by either o f these hypotheses, although it may never reach a very low level. On the other hand, suscepti- bility to attitude change could, in theory, be near zero through- out the life cycle, a hypothesis that we view as the relevant null hypothesis for these others. P r e v i o u s Tests o f T h e s e H y p o t h e s e s Two principal research approaches have been taken to test these hypotheses. The first involves tracking birth cohorts' aver- age attitudes over time. Perhaps the best known o f these studies was Newcomb's classic Bennington study (Newcomb, 1943; Newcomb et al., 1967). Newcomb measured the social and po- litical attitudes o f a cohort o f female students before, during, and after their years at Bennington College. These women had been raised in conservative, socioeconomically advantaged households and were confronted with a very liberal environ- ment at Bennington. During their college years, the students' mean political orientation moved from the conservative end o f the political spectrum toward the liberal end. The longer stu- dents stayed at Bennington, the more liberal they became on average. Newcomb et al. (1967) reinterviewed many o f these students 25 years after they graduated from Bennington and found that they had remained quite liberal on average in terms o f political orientations. And when they were interviewed again in 1984, the Bennington women continued to evidence highly liberal po- litical orientations on average (Alwin et al., in press). This evi- dence is limited because o f the unrepresentativeness o f the sam- ple, but studies o f aggregate-level attitude change in representa- tive national samples have also found large shifts in young cohorts and negligible shifts in middle-aged and older cohorts (Converse, 1976; Crewe, Sarlvik, & Air, 1977; Glenn, 1980; Markus, 1979; Mueller, 1973; Nunn, Crockett, & Williams, 1978). Thus, this body o f evidence is consistent with the im- pressionable years hypothesis and challenges the increasing per- sistence hypothesis. Assessing attitude change by tracking changes in a group's average attitude is almost certain to underestimate the amount o f attitude change that actually occurred. High levels o f aggre- gate stability can coexist with high levels o f individual attitude change that cancel out at the aggregate level. Therefore, more definitive tests o f the impressionable years and increasing per- sistence hypotheses require analyses o f individual-level attitude stability. This is the second principal research method that has been used in this literature. To assess the Bennington women's iudividual-level stability, Newcomb et al. (1967) computed test-retest correlations be- tween attitude measures. Because correlations between atti- tudes in 1938 and attitudes in 1960 were slightly larger (about .50 on average) than were correlations between attitudes in 1935 and attitudes in 1938 (about.42), Newcomb concluded that the Bennington women's attitudes became more stable as they got older. Jennings and his colleagues reported similar evidence on the consistency over time o f attitude reports made by represen- tative national samples o f adults (Jennings, 1988; Jennings & Markus, 1984; Jennings & Niemi, 1978; Markus, 1979). Al- though Newcomb et al. did not assess attitude change after age 40, Jennings and Markus (1984) did; they found no consistent evidence o f an increase in the over-time consistency o f attitude reports between the ages o f 25 and 63. Therefore, this evidence is most supportive o f the impressionable years hypothesis. The only investigation that examined political attitude stabil- ity across the entire life span reached a different conclusion. Sears (1981) found that test-retest correlations between mea- sures o f racial prejudice gradually increased across 21-28-year- olds, 29-44-year-olds, and 45-60-year-olds, a result consistent with the increasing persistence hypothesis. Sears also found un- expectedly low levels o f over-time consistency among individu- als older than age 60, a result he viewed as evidence that an increase in susceptibility to change occurs in later years. C o n f o u n d i n g A t t i t u d e C h a n g e W i t h A t t i t u d e M e a s u r e m e n t U n r e l i a b i l i t y As a group, these studies are problematic partly because o f the inconsistencies among their findings. However, they are also problematic because o f a common flaw in their analytic ap- proaches. It is well known that test-retest correlations, such as those computed by Newcomb et al. (1967), Jennings and Mar- kus (1984), and Sears (198 l), are not pure measures o f attitude 4 1 8 JON A. KROSNICK AND DUANE E ALWIN stability (see, e.g., W h e a t o n , M u t h e n , Alwin, & S u m m e r s , 1977). These correlations reflect b o t h the a m o u n t o f a t t i t u d e change t h a t takes place d u r i n g a given t i m e p e r i o d a n d the a m o u n t o f r a n d o m m e a s u r e m e n t e r r o r in r e p o r t s o f attitudes. Differences between age groups in t e r m s o f over-time corre- lations may reflect differences in the extent o f either o r b o t h processes, because b o t h a t t e n u a t e such correlations. Jennings a n d M a r k u s (1984) r e c o g n i z e d this confounding (see pp. 1004-1005, f o o t n o t e 3). In t h e i r analysis o f p a r t y iden- tifieation, they used s t r u c t u r a l equation m o d e l i n g to separate a t t i t u d e change from a t t i t u d e m e a s u r e m e n t unreliability. Jen- nings and M a r k u s f o u n d t h a t t a k i n g m e a s u r e m e n t e r r o r i n t o a c c o u n t in this fashion d i d n o t alter their evidence o f increased stability in early a d u l t h o o d . However, Jennings and colleagues d i d n o t p e r f o r m s i m i l a r analyses for the other attitudes t h e y m e a s u r e d , and it is i m p o s s i b l e t o be certain t h a t doing so would n o t have altered t h e i r general conclusion. Sears ( 1981) dealt with m e a s u r e m e n t e r r o r in a different fash- ion. H e assessed the r e l a t i o n between age a n d r a n d o m m e a s u r e - m e n t e r r o r b y c o m p u t i n g C r o n b a c h ' s a l p h a coefficients for his 10-item index o f racial attitudes. Because the a l p h a coefficients for all o f t h e age g r o u p s were " i n the s a m e ballpark," Sears con- cluded t h a t the v a r i a t i o n in t e s t - r e t e s t correlations t h a t he ob- served reflected only differences in a t t i t u d e change. Treating C r o n b a c h ' s a l p h a as a m e a s u r e o f reliability as Sears (1981) d i d entails m a k i n g the a s s u m p t i o n t h a t the 10 i t e m s c o m p o s i n g his index a r e all univocal m e a s u r e s o f a single, u n i - d i m e n s i o n a l o r i e n t a t i o n t o w a r d racial integration. However, re- c e n t evidence suggests t h a t the i t e m s Sears analyzed actually m e a s u r e m o r e t h a n one latent a t t i t u d e d i m e n s i o n (Bobo, 1983). I f this is so, Sears's a l p h a s c o n f o u n d r a n d o m m e a s u r e m e n t er- r o r with the strength o f the relations a m o n g the dimensions. Variation in the latter across age g r o u p s c o u l d have masked variation in m e a s u r e m e n t reliability across age groups. F u r t h e r m o r e , although Sears (1981) viewed his C r o n b a c h ' s a l p h a analysis as revealing n o relation between age a n d a t t i t u d e m e a s u r e m e n t reliability, a n u m b e r o f previous investigations have f o u n d t h a t the a m o u n t o f r a n d o m m e a s u r e m e n t e r r o r in a t t i t u d e r e p o r t s is greater a m o n g older respondents (Andrews, 1984; A n d r e w s & Herzog, 1986; Stember, 1951-1952). Indeed, a close inspection o f Sears's (1981) reliability coefficients re- veals t h a t the oldest respondents' a l p h a was lower t h a n those o f younger respondents, enough so to e x p l a i n the decrease in over- t i m e correlations. 2 Therefore, the decline Sears discovered in over-time correlations a m o n g the oldest age g r o u p may reflect an increase in r a n d o m m e a s u r e m e n t e r r o r i n s t e a d o f an in- crease in the r a t e o f a t t i t u d e change. foreign policy issues, attitudes toward social groups, political p a r t y identification, a n d liberal-conservative ideological orien- tations. T h i r d , we i m p l e m e n t e d a n analytic a p p r o a c h t h a t per- m i t t e d us to estimate attitude change a n d r a n d o m m e a s u r e m e n t e r r o r separately in a way t h a t d i d n o t require t h a t a s s u m p t i o n s be m a d e a b o u t the u n i d i m e n s i o n a l i t y o f sets o f attitude items. Study 1 was based on d a t a collected f r o m two panels o f re- spondents, who were each interviewed on t h r e e occasions, 2 years apart. Each p a n e l ' s interviews s p a n n e d two presidential elections a n d one congressional election. Respondents in each p a n e l were asked certain a t t i t u d e questions d u r i n g each inter- view, which allowed us to generate separate estimates o f a t t i t u d e stability a n d a t t i t u d e m e a s u r e reliability, using structural equa- tion modeling. Study 2 used the same analytic m e t h o d b u t was based on d a t a collected over only 8 months, d u r i n g a single pres- idential election c a m p a i g n . Taken together, these d a t a sets per- m i t t e d us to e x a m i n e political a t t i t u d e change b o t h within and between presidential election campaigns. S t u d y 1 M e t h o d Samples. The data analyzed in this study came from two panel sur- veys in the National Election Study series. For the first survey the Uni- versity of Michigan's Center for Political Studies interviewed a repre- sentative, national cross-section of 1,132 adults living in private house- holds in the United States at the time of the 1956 presidential election. They were interviewed again at the time of the 1958 midterm election and again at the time of the 1960 presidential election. In the second panel survey that we examined, a nationally representative sample of 1,320 adults was interviewed at the time of the 1972 U.S. presidential election. They were interviewed again at the time of the 1974 midterm election and again at the time of the 1976 presidential election. In order to permit testing of hypotheses regarding aging, we divided the respondents in each panel into the following age groups on the basis of their age at the first wave of the survey: 18-25, 26-33, 34-41, 42-49, 50-57, 58-65, and 66-83. The groups were based on 8-year age spans in order to generate as many approximately equal-sized age groups as possible that were each large enough to yield reliable parameter esti- mates. The resulting age-group sizes for the 1950s panel were 94, 228, 241, 187, 138, 101, and 99, respectively. For the 1970s panel, the age group sizes were 208, 227, 190, 204, 167, 147, and 161, respectively) Measures. Our analyses required that attitudes be measured on three occasions. In the 1956-1960 study, respondents were asked identical questions in each year that measured attitudes toward school integra- tion, federally guaranteed employment for all Americans, U.S. involve- ment in the affairs of other nations, U.S. international financial aid pro- grams, maintenance of U.S. troops abroad, and federal financial aid to schools. Also, during all three waves, respondents were asked which ma- T h e P r e s e n t I n v e s t i g a t i o n I n sum, p a s t investigations o f the r e l a t i o n between age and a t t i t u d e stability have h a d significant methodological limita- tions. In this article, we will r e p o r t two studies o f the relation between age and a t t i t u d e change t h a t overcome these p r o b l e m s in t h r e e p r i n c i p a l ways. First, the d a t a we a n a l y z e d were col- lected f r o m nationally representative s a m p l e s o f a d u l t s in the U n i t e d States who ranged in age f r o m 18 to 83 years. Second, we e x a m i n e d a range o f social a n d political attitudes, including t r u s t in government, political efficacy, attitudes on d o m e s t i c and 2 When Sears (1981) controlled for educational differences between age groups, the average drop from the 45-60-ycar-olds to the 61 +-year- olds in over-time consistency was. 16, whereas the average drop in reli- ability was .08. Because unreliability in a measure reduces a test-retest correlation twice, at both Wave 1 and Wave 2 (Alwin, 1973), this differ- ence in reliability would fully account for the difference in over-time consistency. 3 These ns do not add up to the total Ns for the panels because of missing data on the age variable. In the 1950s panel, the ages of 44 respondents were not ascertained, and in the 1970s panel, the ages of 16 respondents were not ascertained. AGING A N D SUSCEPTIBILITY TO ATTITUDE C H A N G E 4 1 9 jor political party they identified with and how strongly they did so. In the 1972-1976 panel, respondents were asked identical questions in each year measuring seven categories o f attitudes: (a) political party identification, (b) liberal-conservative ideological orientation, (c) atti- tudes toward social groups, such as labor unions, the military, and the police, (d) attitudes on racial policy issues, such as school bussing pro- grams, (e) attitudes on nonracial domestic policy issues, such as feder- ally guaranteed employment and protecting the rights o f people accused o f committing crimes, (f) political efficacy and alienation, and (g) atti- tudes toward prominent politicians. In total, 50 items were analyzed from these two panel surveys. 4 Some o f the attitude measures allowed respondents to indicate (a) that they did not know what their attitude was on an issue or (b) that they had not thought enough about an issue to form an attitude. Re- spondents who gave either o f these answers to a given question at any wave were eliminated from the analyzed sample for that question. 5 Analysis. We wished to estimate the reliability o f these measures and the stability o f the underlying attitudes for each o f the seven age groups. A number o f analytic approaches have been proposed for doing so using structural equation modeling all o f which require data on a panel o f respondents collected on at least two occasions (see Alwin, 1988). When only two waves o f data are available, these approaches require that the assumption be made that a set o f questions are congeneric measures o f a single attitude. Because we had measures o f the same attitudes on three occasions, we were able to estimate the parameters o f a model that does not entail this assumption. The model we estimated is composed o f two structural equations: X t = ¢, + ~t, ( 1 ) a n d ¢, = / 3 , , , - 1 ~ , - 1 + ~',. ( 2 ) Equation 1 specifies the measurement model, which describes the rela- tion between the latent attitude being measured and verbal reports o f it. It decomposes variance in reports Oft, where t = 1, 2, 3) into two components, one due to the latent attitude (rt) and the other due to random measurement error (~t). Equation 2 specifies the structural model, which describes the relations among the latent attitudes./~t.t-~ is the stability of the latent attitude during the period from t - 1 to t • ~'t represents sources o f change over time in the latent attitude. This model has been discussed extensively by Heise (1969), Wheaton et al., (1977), and Wiley and Wiley (1970). As it appears in Equations 1 and 2, the model is underidentified. There are six independent pieces o f information available (o12, 0,32, 0,31, ~l 2, 0,22, and 0,32) with which to estimate eight parameters (/~21,/~32, ¢~12, ¢~e, 0,~32, #~12, ¢r22, and 0,~32). In order to identify the model, the attitude measure's reliability is assumed to remain constant across waves (#1 e - O'M2)/0,12 = (O"22- 0,E22)/0,22 = ( 0 , 3 2 - 0,e32)/0"32; Heise, 1969). 6 Given this constraint, the model is just-identified, so the parameter estimates will always fit the observed data perfectly. For each attitude measure, the parameters o f the model were esti- mated separately for each o f the seven age groups using LISREL VI (Jores- kng & Sorbom, 1978). We then analyzed the resulting parameter esti- mates in two ways. First, we were interested in the stability o f the latent attitude, so we examined the estimates of/32~ and/~32. Second, we exam- ined the single reliability estimate generated for each attitude measure. And finally, we were interested in comparing the conclusions reached by our analysis with those that would have been reached had we ignored the distinction between attitude instability and attitude measure unreli- ability. We therefore examined Pearson product-moment test-retest correlations between the attitude measures at adjacent waves: rl2 and r2a. For the sake o f simplicity, all o f the following statistical tests involv- ing stability are based on a single combined analysis o f all the stability coefficients (/~2~ and/~32), and all tests involving over-time consistency are based on a single combined analysis o f all the test-retest correlations (r12 and r23 ). Results Zero-order correlations. T h e t o p p a n e l o f T a b l e I displays t h e average t e s t - r e t e s t c o r r e l a t i o n s f o r e a c h age g r o u p . A s e x p e c t e d , a o n e - w a y analysis o f v a r i a n c e (ANOVA) o f t h e figures i n t h e t h i r d c o l u m n i n d i c a t e d t h a t t h e r e were statistically significant differences a m o n g t h e age g r o u p s , F ( 6 , 693) = 2.86, p = .01. Tests o f a p r i o r i c o n t r a s t s r e v e a l e d t h a t , c o n s i s t e n t w i t h t h e t w o a g i n g hypotheses, t h e y o u n g e s t age g r o u p ' s average c o r r e l a t i o n was significantly s m a l l e r t h a n t h e average o f t h e 2 6 - 6 5 age g r o u p s , t(693) = 2.18, p < .03. C o n s i s t e n t w i t h t h e c l a i m t h a t a t t i t u d e m e a s u r e m e n t r e l i a b i l i t y d e c r e a s e s i n o l d age, t h e oldest age g r o u p ' s average c o r r e l a t i o n was significantly s m a l l e r t h a n t h e average o f t h e 2 6 - 6 5 age g r o u p s , t(693) = 2.97, p < .01. T a k e n together, t h e s e results p r e c i s e l y r e p l i c a t e Sears's (1981) findings. 7 Attitude stability. T h e m i d d l e p a n e l o f T a b l e 1 displays esti- m a t e s o f a t t i t u d e stability. T h e i m p r e s s i o n a b l e years h y p o t h e s i s p r o p o s e s t h a t t h e s e stability coefficients s h o u l d b e relatively s m a l l i n t h e y o u n g e s t age g r o u p a n d s h o u l d b e large a n d a p p r o x - i m a t e l y e q u a l i n t h e r e m a i n i n g age groups. T h e i n c r e a s i n g - p e r - sistence h y p o t h e s i s p r o p o s e s i n s t e a d t h a t w e s h o u l d o b s e r v e a consistent, g r a d u a l i n c r e a s e i n t h e stability coefficients a c r o s s t h e age g r o u p s . 4 For two other items in the 1950s data set and six other items in the 1970s data set, either LISREL was unable to estimate the structural equation model parameters for one age group, or one age group's pa- rameter estimates were far out o f range. These problems probably oc- curred because o f random fluctuation in the zero-order correlations be- tween the attitude measures that resulted from small sample sizes and sampling error. In order to maintain comparability o f items across age groups, we did not include these items in any o f the following analyses. However, when we repeated the analyses deleting only those single data points that were not estimable, the results were equivalent to those re- ported here (and in fact supported our conclusions even more strongly). The first-wave National Election Study item numbers for the attitude measures included in these analyses are 32, 35, 41, 53, 56, 74, and 88 in the 1950s and 112, 140, 202, 232, 259, 253, 255,258, 264, 269-274, 570-574, 576, 578, 621,652, 707, 709, 714, 717-722, 724, 737, 744, 745,749, 752, 753, 755,756, and 1067 in the 1970s. 5 All o f the following analyses were also run recoding "don't know" and "haven't thought much about this issue" responses to the midpoint o f the attitude scales. The results o f these analyses are essentially identi- cal to those reported here. Levels o f " d o n ' t know" and "haven't thought much about this issue" responses increase consistently across the life cycle, from a low o f about 12% for 18-25-year-olds to about 23% for 66-83-year-olds, with most o f the increase occurring after age 49. 6 An alternative approach to identifying the model described in Equations 1 and 2 was proposed by Wiley and Wiley (1970). All o f the following analyses were recomputed using their identifying assumption. The results o f these analyses were comparable to those reported in the text. 7 Here and throughout the remainder o f the results, we found no sta- tistically significant differences between attitude--object categories in the relation between age and attitude stability and attitude-report reli- ability. For more detailed results in this regard, see Alwin and Krosnick (1988, 1989) and Krosnick and Alwin (1989). 420 JON A. KROSNICK AND DUANE E ALWIN Table 1 Study 1: Zero-Order Correlations, Attitude Stabilities, and Attitude-Measure Reliabilities by Age Group Time period" Wave 1- Wave 2 - Age group Wave 2 Wave 3 Combined b Zero-ordercorrelations 18-25 .42 .46 .44 26-33 .46 .47 .46 34-41 .48 .52 .50 42-49 .49 .50 .50 50-57 .44 .48 .46 58-65 .46 .48 .47 66-83 .41 .44 .43 Standardized stability coefficients 18-25 0.82 0.89 0.85 26-33 0.86 0.90 0.88 34-41 0.90 0.98 0.94 42-49 0.90 0.96 0.93 50-57 0.83 0.97 0.90 58-65 0.90 0.96 0.93 66-83 0.91 0.94 0.92 Reliabilities 18-25 .55 26-33 .54 34-41 .54 42-49 .55 50-57 .58 58-65 .54 66-83 .51 a Each entry in these two columns is an average of 50 coeffi- cients, b Each entry in this column is an average of 100 coefficients, except for the reliabilities, which are averages of 50 coefficients. As expected, the youngest age group's average stability co- efficient (shown i n the last c o l u m n ) is significantly lower t h a n the average of the r e m a i n i n g age groups, t(693) = 2.36, p < .02. Furthermore, the 26-33-year-olds' average stability is margin- ally significantly lower t h a n the average of the 34-83-year-olds, t(693) = 1.52, p = . 13. However, there are n o other statistically significant differences between the coefficients. Furthermore, there is no correlation between age a n d stability among the 3 4 - 83-year-olds (r = - . 0 2 , ns). This evidence is therefore d e a r l y consistent with the impressionable years hypothesis a n d is in- consistent with the increasing persistence hypothesis. Further- more, these results refute Sears's ( 1981 ) claim that attitude sta- bility decreases toward the end o f the life cycle. Attitude measurement reliability The figures i n the bottom panel of Table 1 are estimates of the average reliability o f the attitude measures for each age group. As expected, the average reliabilities reveal a trend toward decreased reliability toward the end of the life cycle. The average reliability for the 18-65- year-olds is .55, which drops to .51 for the 66-83-year-olds. However, the average reliability of the 66-83-year-olds is n o t significantly smaller t h a n the average of the r e m a i n i n g age groups, t(343) = 1.21, p = .23, and n o other statistically signifi- cant differences appeared among the other age groups. These results are therefore weak b u t consistent with the claim that attitude-measurement reliability decreases at the end of the life cycle a n d that the drop i n over-time consistency of older respon- dents' attitude reports is d u e to decreased measurement reli- ability, n o t decreased attitude stability. S t u d y 2 The results of Study 1 are based on data collected over 4-ycar periods, including two presidential elections and a congres- sional election. Therefore, the estimates of attitude stability re- ported in Study 1 assess attitude change that occurred between elections. Because the dynamics o f between-election attitude change may be different from the dynamics of attitude change that occurs within a single campaign, we conducted a second study, i n which we examined the relation of age to within-cam- paign attitude change. Method Samples. For the 1980 National Election Panel Study, the Center for Political Studies interviewed a representative national cross-section of 769 American adults living in private households. Each respondent was interviewed three times, first in January or February of 1980, again in June, and again in September or October. These respondents were again divided into seven age groups on the basis of their a ~ at the first wave of the survey: 18-25, 26-33, 34--41, 42-49, 50-57, 58-65, and 66-83. The resulting age group sizes were 127, 163, 114, 68, 93, 87, and 88, respectively, s Measures and analysis. Our analyses again focused on attitudes that were measured on three occasions. Respondents in this study were asked identical questions in each year measuring attitudes in five catego- ries: (a) political party identification, (b) liberal-conservative ideologi- cal orientation, (c) attitudes toward social groups, (d) attitudes on non- racial domestic policy issues, and (e) attitudes toward prominent politi- cians. In total, 18 items were included in our anaiyses. 9 Again, respondents who indicated (a) that they did not know what their attitude was on an issue or (b) that they had not thought enough about an issue to form an attitude at any wave were eliminated from the analyzed sam- ple for that issue. Estimates of the stability of these attitudes and of the reliability of these measures for each of our seven groups were generated using the same methods as were used in Study 1. Results Zero-order correlations. The top panel of Table 2 displays the average zero-order test-retest correlations for the seven age groups. A one-way ANOVA of the averages i n the last c o l u m n again indicated that there were statistically significant differ- ences among the age groups, F(6, 245) = 2.90, p = .01. Consis- tent with both aging hypotheses, the youngest age group's aver- age correlation was smaller than the average of the 26-57-year- s Again, the total of these age group ns does not equal the total sam- ple, because the ages of 29 respondents either were not ascertained or fell outside of the 18-83 range. 9 Because of inestimable or out-of-range parameters, five other items were not included in the analyses reported here. The first-wave National Election Study item numbers for the included attitude measures are 425, 426, 428--431,433, 438-440, 442, 801,944, 1114, 1189, 1278, 1282, and 1284. AGING AND SUSCEPTIBILITY TO ATrITUDE CHANGE 421 Table 2 Study 2: Zero-Order Correlations, Attitude Stabilities, and Attitude-Measure Reliabilities by Age Group Time period" Wave 1- Wave 2 - Age group Wave 2 Wave 3 Combined b Zero-ordercorrelations 18-25 .57 .63 .60 26-33 .64 .67 .66 34-41 .64 .66 .65 42-49 .66 .69 .68 50-57 .64 .66 .65 58-65 .59 .63 .61 66-83 .52 .59 .56 Standardized stability coefficients 18-25 0.83 0.90 0.87 26-33 0.88 0.93 0.91 34-41 0.93 0.97 0.95 42-49 0.90 0.97 0.94 50-57 0.96 0.98 0.97 58-65 0.95 1.04 1.00 66-83 0.86 1.02 0.94 Rcliabilities 18-25 .75 26-33 .73 34-41 .69 42-49 .74 50-57 .72 58-65 .63 66-83 .66 a Each entry in these two columns is an average of 18 coeffi- cients, b Each entry in this column is an average of 36 coefficients, except for the reliabilities, which are averages of 18 coefficients. old age groups, t(245) = 2.06, p = .04. A n d consistent with the claim that attitude-measurement reliability decreases in old age, the 26-57-year-old age group's average correlation was sig- nificantly larger than those o f the 58-65-year-olds, t ( 2 4 5 ) = 1.94, p < .06, and the 66-83-year-olds, t(245) = 3.72, p < .01. Thus, these figures replicate Sears's ( 1981) results again. Attitude stability. The middle panel o f Table 2 displays the structural equation model estimates o f attitude stability. Con- sistent with both the impressionable-years hypothesis and the increasing-persistence hypothesis, the youngest age group's av- erage stability coefficient (shown in the last column) is signifi- cantly smaller than the mean o f the 26-83-year-olds, t(245) = 2.05, p < .05. Therefore, young adults appear to have been more susceptible than older adults to attitude change during the 1980 presidential election campaign. Again, there is n o correlation between age and stability among the 34-83-year-olds (r = .02, ns), which disconfirms the increasing persistence hypothesis. A n d although there is some indication here that the attitude stability o f 66-83-year-olds is lower than that o f 58-65-year- olds, these two groups' stability coefficients do n o t differ sig- nificantly, t(245) = 1.20, p = .23. Furthermore, the stability co- efficient for the 66-83-year-olds is n o t significantly different from the average o f that o f the 34-65-year-olds, t(245) = 0.65, ns. Thus, there is n o evidence o f a decrease in attitude stability at the end o f the life cycle. Attitude-measurement reliability. The figures in the b o t t o m panel o f Table 2 estimate the average reliability o f the attitude measures for each age group. The relation between age and reli- ability is again quite clear. The 58-83-year-olds' average reli- ability is marginally significantly smaller than that o f the 18- 57-year-olds, t(119) = 1.87, p < .07, and n o other statistically significant differences appeared among the other age groups. Therefore, these results further reinforce the claim that atti- tude-measurement reliability decreases at the end o f the life cycle. D i s c u s s i o n The analytic strategy used in this investigation represents an improvement over previous methodologies in a n u m b e r o f re- spects. First, we used methods that attempted to unconfound attitude stability and attitude-measurement reliability. As a re- sult, our comparisons o f different age groups in terms o f atti- tude stability presumably do n o t reflect differences in reliability. Second, instead o f using multiple-indicator structural equation models, we used single-indicator models. Therefore, the validity o f our conclusions does not depend on the validity o f the as- sumption that a set o f indicators is congeneric, measuring only a single latent attitude. Finally, our methods improve on those used in past studies because we aggregated attitude stability and reliability estimates over large samples o f attitude measures. Therefore, o u r conclusions can reasonably be generalized broadly across political attitude domains. Aging and political attitude stability. The results o f Studies l and 2 provide strong support for the impressionable years hy- pothesis. According to this perspective, people are most suscep- tible t o political attitude change during their early adult years, and susceptibility drops off immediately thereafter. This is pre- cisely the pattern o f stability coefficients we found in Studies l and 2, although the differences we found were n o t huge. Con- trary t o the increasing-persistence hypothesis, we found n o evi- dence o f gradually increasing attitude stability after age 33. Studies 1 and 2 differed from one another in terms o f the type o f attitude change that was assessed. The focus o f Study 1 was on attitude change that occurred between elections, whereas Study 2 assessed change within the course o f a single campaign. Although between-election attitude change and within-election attitude change may result from very different social processes, o u r results indicate that both are most likely t o occur among young adults and less likely to occur among older adults. This conclusion is satisfying partly because it is consistent with the results o f all previous research on the relation o f age t o political attitude stability (Alwin et al., in press; Converse, 1976; Crewe et al., 1977; Glenn, 1980; Jennings, 1988; Jennings & Markus, 1984; Jennings & Niemi, 1978; Markus, 1979; Mueller, 1973; Newcomb, 1943; Newcomb et al., 1967; N u n n et al., 1978; Sears, 198 l). This finding also partly parallels those o f previous studies o f the effects o f aging on other aspects o f political cognition and behavior. For example, Wolfinger and Rosenstone (1980, p. 48) found that, once education and other demographic variables are controlled, the likelihood that a citi- zen will vote in a particular presidential election is relatively 422 JON A. KROSNICK AND DUANE E ALWlN low at age 18, increases dramatically up to age 30, and continues to increase therealter, although more slowly. Converse (1976) found that the strength of political-party identification follows the same pattern across the life span. These findings are consis- tent with our evidence of increases in political-attitude stability between the ages of 18 and 30, so they may be produced by similar mechanisms. However, our failure to find evidence of increased stability thereafter conflicts with Wolfinger and Ro- senstone's (1980) and Converse's (1976) results and raises the possibility that different mechanisms may be operating. Wolfinger and Rosenstone's (1980) research suggests an inter- esting possible relation that may underlie our results. They found that the increase in voter turnout with age was powerfully related to educational attainment. Aging led to dramatically in- creased turnout among people who had not graduated from high school, but aging had almost no effect on turnout among people with at least some graduate school training (p. 59). Thus, advanced education apparently instigates a high level of turnout during early adulthood that remains crystallized throughout later adulthood. Advanced education may also solidify young people's political attitudes, so that the relation we found be- tween age and attitude stability may disappear among well-edu- cated respondents. Unfortunately, our samples were too small to permit simultaneous analysis o f age and education. Future studies that do this simultaneous analysis may make an impor- tant contribution to our understanding o f the mechanisms through which aging affects political cognition and behavior and o f the limiting conditions of those effects. It is interesting to note that the magnitude of the effect of aging on influenceability found here is comparable with the ap- parent effect on influenceability of another frequently studied individual difference variable: gender. Eagly and Carli's (198 l) mean effect size for gender, calculated in terms o f d, was .26. In terms o f d, the effects of age identified here are .21 in Study l and .20 in Study 2. The methods used in the present study are very different from those used in the studies examined by Eagiy and Carli, so direct comparison o f these ds is probably not war- ranted. Nonetheless, they suggest that age and gender may have comparable effects on persuasibility, although the effects o f both variables are not especially strong. Aging and attitude measurement unreliability. Contrary to Sears's ( 1981 ) conclusion, Studies 1 and 2 revealed no evidence o f a drop-offin attitude stability at the end of the life cycle. Our results suggest that his finding in this regard reflected a slight increase in the amount of random measurement error in atti- tude reports provided by the oldest respondents in surveys. Consistent with previous work on survey-measurement reli- ability (e.g., Andrews, 1984; Andrews & Herzog, 1986), Studies I and 2 produced evidence of this relation between age and atti- tude-measurement reliability. We suspect that this relation oc- curs because, as people age, they become less able or less willing, or both, to make precise reports of internal states by means of closed-ended survey questions. These results demonstrate how equating attitude stability with zero-order test-retest correlations between attitude mea- sures can lead to incorrect conclusions. Unreliability in attitude reports is partly the result o f respondent characteristics (Alwin & Krosnick, 1988), so between-group differences in over-time consistency confound differences in attitude stability with differences in unreliability. Given the clear conceptual distinc- tion between these two processes, researchers making compari- sons of any psychological parameters across groups defined by age or by any other individual difference variable should take into account the possibility o f measurement-error differences between groups. Aging effects versus birth cohort differences. It is well known that comparisons o f age groups using cross-sectional data may reflect the effects o f aging~ but they may also represent other processes instead (see Riley, 1973); most important, they may represent cohort-specific differences. In our case, we have as- sumed that the lower stability we observed in young adults' atti- tudes increases as they age. However, the unique socialization experiences of these individuals may have made them unusually susceptible to attitude change, and this unusual susceptibility may be maintained throughout their lives. We cannot fully test and rule out this possibility. However, the fact that similar re- lations of age to attitude stability were found in the 1950s, 1970s, and 1980s panel data when they were analyzed sepa- rately suggests strongly that this result does not reflect the unique socializing experiences of one young birth cohort. In principle, our finding that the oldest adults evidenced greater unreliability in their attitude reports may also reflect birth cohort differences instead of aging differences. In fact, Sears (1981) suggested that this relation may appear because older birth cohorts acquired less formal education on average than did younger birth cohorts, and education is positively re- lated to attitude-report precision. Although education is indeed negatively related to random measurement error in attitude re- ports (Alwin & Krosnick, 1988; Andrews, 1984; Andrews & Herzog, 1986), studies that controlled for educational differ- ences between age groups found that older respondents still evi- denced heightened random-measurement error (Andrews, 1984; Andrews & Herzog, 1986). Sears's (1981) own analysis revealed that controlling for education reduced but did not eliminate the drop in over-time consistency or reliability of atti- tude reports among respondents ages 61 years and older. And a recent cohort analysis found that the reliability of attitude re- ports decreased as each o f a series o f birth cohorts aged, thus demonstrating aging effects with education held constant (see Alwin & Krosnick, 1989). The survey samples we analyzed were too small to permit a simultaneous analysis of age and education, because this would require dividing already small age groups into even smaller edu- cation subgroups. However, given previous research, we suspect that controlling for education would eliminate some but not all o f the relation we observed here between age and the reliability o f attitude reports. And again, because the relation of age to reliability appeared in the 1950s, 1970s, and 1980s data when they were analyzed separately, it does not seem to be due to a particular birth cohort's unique experiences. Sample attrition. Another possible explanation for our find- ings of decreased attitude report reliability late in life is differ- ential sample attrition. As a birth cohort ages, some members die off. It is conceivable that the most reliable members of our oldest birth cohorts were the most likely to have died. There- fore, what appears to be an effect of aging on attitude-measure- ment reliability may in fact be the result of differential attrition. Again, we cannot test or rule this out using the present data. AGING AND SUSCEPTIBILITY TO ATTITUDE CHANGE 423 This can only be accomplished using longitudinal data, col- lected over very long time periods, which permit assessment o f the relation between reliability and mortality. However, if reli- ability and mortality are related, we can imagine that they could be so in exactly the opposite direction: The most unreliable members o f a birth cohort may be the first to die. I f this is so, we have underestimated the relation o f age to unreliability. A more important potential problem involves the sampling practices used for the National Election Studies. It is conven- tional for the National Election Studies to omit from their sam- ples individuals who live in dormitories and prisons and those serving in the military (Converse, 1976, pp. 49-50). It seems likely that greater proportions o f young adults ages 18-25 fall into these categories (particularly the dormitory category) com- pared with older people. I f individuals who fall into these cate- gories have unusually high attitude stability, this could account for the apparently lower level o f attitude stability o f 18-25-year- olds. This is a plausible and more troubling alternative explana- tion for our results that we cannot rule out. Determinants o f attitude change levels. Throughout this arti- cle, we have treated differences between age groups in terms o f attitude stability levels as indicating differences in susceptibility to attitude change. However, it is important to acknowledge that these levels are in fact joint products o f susceptibility to change and the frequency with which individuals encounter attitude- change-inducing stimuli (see, e.g., Converse, 1962; Zaller, 1987). Therefore, young people may evidence higher levels o f attitude change not because o f greater susceptibility but rather because o f greater exposure to change-inducing events. Indeed, this is quite plausible: People probably experience more major life shifts between ages 18 and 25 than at any other point in the life cycle. These changes may include graduating from high school, going to college, graduating from college, starting a first job, getting married, becoming a parent, moving from one place o f residence to another, and so on. Without measuring and con- trolling for the frequency o f such attitude-change-inducing events, it is impossible to be certain that the higher levels o f attitude change among 18-25-year-olds observed here can be attributed to greater susceptibility to attitude change. The validity of our estimation method. Separating test-retest correlations into components due to stability and unreliability is an inherently tricky business; therefore, it seems appropriate to be cautious about the validity o f our conclusions in this re- gard. Although our approach to this problem is an improve- ment over previous efforts, the structural-equation model we used does make certain assumptions that may be invalid. For example, the model assumes that attitude change occurs ac- cording to a simplex process (see Alwin, 1988), and some skep- ticism has been expressed about this assumption (Ragosa, 1988). However, a careful review o f our test-retest correlations revealed that nearly all o f them had a simplex structure; there- fore, we are not particularly troubled by this assumption. The model also assumes that measurement reliability remains con- stant across repeated interviews, which may not be the case. For example, a Socratic effect may produce a decrease in random measurement error across waves o f panel surveys (Jagodzinski, Kuhnel, & Schmidt, 1987; but see Saris & Putte, 1988). How- ever, the reliabilities generated by our estimation method are essentially averages computed across the three interviews, which can be reasonably compared across age groups even if the assumption of constant reliability is violated. Although these assumptions, therefore, do not seem prob- lematic, the model does make other assumptions that may be untenable and that may complicate a clean separation o f stabil- ity from unreliability. For example, the model assumes that there is no correlated measurement error across interviews, and violation o f this assumption would lead to inappropriately low estimates o f unreliability (although not inappropriate estimates of random measurement error; see Alwin, 1974). We therefore await the development o f more effective analytic techniques or the collection o f multiwave panel data sets that will eliminate the necessity o f making these assumptions. In the meantime, howevc~ the method we have used here seems to us to be the best one available for accomplishing our goals. Laboratory studies of attitude change. Assuming that we have identified age-based differences in susceptibility to attitude change, our findings have important implications for social psy- chologists' traditional approach to studying the determinants o f attitude change: laboratory experiments involving college undergraduate subjects. Our results make it clear that these sub- jects, most o f whom fall in the 18-25 age range, are likely to evidence higher levels o f attitude change than are older adults. Therefore, the levels o f attitude change typically found in labo- ratory studies are likely to be higher than would be produced by more representative samples o f adults (see Hovland, 1959; Sears, 1986). Do our results suggest that social psychologists should no longer conduct studies o f college sophomores? We believe not. Social psychologists are generally interested in documenting the processes by which attitude change occurs, not in assessing gen- eral levels o f flexibility. Whether the processes by which attitude change occurs are any different among young people than they are among older people remains a matter for empirical evalua- tion (although Sears, 1986, has suggested that such differences are likely). Unless process differences are uncovered, our evi- dence suggests that social psychologists' traditional approach is probably a highly sensible one. Studies o f any process can most efficiently be conducted using subjects who are most likely to manifest that process; thus, studies o f attitude change may be most efficiently conducted using subjects who are most likely to be susceptible to it: those between ages 18 and 25 years. Clearly though, assessing the generalizability o f attitude change pro- cesses across the life span is a worthy focus for future research. Impressionable years or lifelong openness? A cursory exami- nation o f our results reveals that we have reported very high stability coefficients, ranging from 0.82 to 1.04. One might be tempted to conclude from this that the political attitudes we examined are nearly perfectly stable after early adulthood and that political attitude change is likely to be an extremely rare event after age 25 (see, e.g., Achen, 1975). This conclusion is supported by the only study that tracked attitude stability across nearly the entire life span, the Bennington study (Alwin et al., in press). It found remarkably high levels o f attitude sta- bility (betas o f approximately 0.86) across approximately 23- year periods (Alwin et al., in press; Newcomb et al., 1967). When this finding is considered in light o f Study l ' s average stability coefficient for 2-year periods (0.91) and Study 2's aver- age stability coefficient for 4-month periods (0.94), it appears 4 2 4 JON A. KROSNICK AND DUANE E ALWIN t h a t stability is a s y m p t o t i c a t a relatively high level. I° Thus, this evidence would offer strong s u p p o r t for the impressionable- years hypothesis a n d w o u l d suggest rejection o f the lifelong- openness hypothesis ( B r i m & Kagan, 1980; G l e n n , 1980; Sears, 1981, 1983). However, it is possible to use these d a t a t o argue for a different conclusion. The women who were the subjects o f the Benning- t o n study a r e u n c o n v e n t i o n a l in n u m e r o u s ways a n d may there- fore evidence u n u s u a l l y high levels o f a t t i t u d e stability (see AI- win et al., in press). F u r t h e r m o r e , the overall average stability coefficients r e p o r t e d here a r e n o t exactly 1.0, a n d t h e y s u m m a - rize only the a m o u n t o f change t h a t t o o k place d u r i n g 2-year p e r i o d s in Study 1 a n d 3 - 4 m o n t h p e r i o d s in Study 2. W h e n m u l t i p l i e d out t o p r o d u c e an e s t i m a t e o f stability over a 10-year period, a 2-year stability coefficient o f 0.90 becomes a b o u t 0.60. Given t h a t the average stability coefficients from Study 2 are n o t m u c h larger t h a n those f r o m Study 1 (which involved longer t i m e intervals between waves), such a s i m p l e - m i n d e d multi- plication exercise m a y n o t a c c u r a t e l y r e p r e s e n t the processes involved. Nonetheless, one c o u l d use this evidence to argue t h a t political a t t i t u d e change m a y n o t be a r a r e exception, b u t rather m a y be a c o m m o n event over the long haul. T h i s is precisely the conclusion r e a c h e d b y S e a r i n ~ Wright, a n d R a b i n o w i t z (1976) on the basis o f a n analysis o f aggregate a t t i t u d e change in the U.S. p u b l i c over a 16-year period. A n d it is a conclusion consistent with Jennings and colleagues' (Jen- nings & Markus, 1984; Jennings & N i e m i , 1981) evidence o f only low-to-moderate levels o f over-time consistency in a t t i t u d e r e p o r t s over 8- to 9-year periods. II These figures u n d e r e s t i m a t e stability because t h e y c o n f o u n d a t t i t u d e change with m e a s u r e - m e n t unreliability, b u t it seems unlikely t h a t correcting for un- reliability would yield stability coefficients very close to 1.0. Thus, this evidence is consistent with the n o t i o n o f lifelong openness to change in beliefs, attitudes, a n d behavior. I f this is true, it has at least two i m p o r t a n t i m p l i c a t i o n s for the study o f political attitudes. First, it w o u l d call i n t o question the widely shared b e l i e f t h a t the m o s t p o t e n t political attitudes, such as p o l i t i e a l - p a r t y identifications a n d racial attitudes, are f o r m e d d u r i n g c h i l d h o o d o r early a d u l t h o o d a n d persist u n a l - t e r e d t h r o u g h o u t the r e m a i n d e r o f the life cycle (e.g., C a m p b e l l , Converse, Miller, & Stokes, 1960; Easton & Dennis, 1969; Sears, 1983). A n d second, it would suggest t h a t political-atti- t u d e change occurs frequently enough in the n o r m a l course o f daily life to justify extensive study o f its d y n a m i c s . Some re- search is b e g i n n i n g to identify the forces t h a t p r o d u c e n a t u r a l l y o c c u r r i n g political-attitude change (e.g., F r a n k l i n , 1984), a n d we l o o k forward t o m o r e such work. We l o o k forward as well to long-term studies o f a t t i t u d e stability across the entire life cycle in representative s a m p l e s t h a t will allow a m o r e definitive test o f the lifelong-openness hypothesis. U n t i l such studies a r e con- d u c t e d , the validity o f the lifelong-openness hypothesis r e m a i n s uncertain. ~o We thank David Sears for pointing this out. ~m Jennings and Niemi's (1981) test-retest correlations vary substan- tially in magnitude. These correlations appear to depend partly on the particular attitude object, perhaps because of different formal proper- ties of the attitude measures (see, e.g., Kxosnick & Alwin, 1989). R e f e r e n c e s Achen, C. H. (1975). Mass political attitudes and the survey response. American Political Science Review, 69, 1218-1231. Alwin, D. E (1973). Making inferences from attitude-behavior corre- lations. Sociomet~ 36, 253-278. Alwin, D. E (1974). Approaches to the interpretation of relationships in the multitrait-multimethod matrix. In H. L. Costner (Ed.), Socio- logical methodology 1973-74 (pp. 79-105). San Francisco: Jossey- Bass. Alwin, D. E (1988). Structural equation models in research on human development and aging. In K. W. Schaie, R. T. Campbell, W. M. Mer- edith, & S. C. Rawlings (Eds.), Methodological issues in aging re- search (pp. 71-170). New York: Springer Publishing. Alwin, D. E, Cohen, R. L., & Newcomb, T. M. (in press). The mgmen o f Bennington: A study o f political orientations over the life span. Mad- ison: University of Wisconsin Press. Alwin, D. E, & Kxosnick, J. A. (1988, June). The reliability ofattitudi- nod survey data. Paper presented at the International Conference on Social Science Methodology, Dubrovnik, Yugoslavia. Alwin, D. E, & Krosnick, J. A. (1989). Aging, cohort stability, and change in sociopolitical attitudes: Exploring the generational-persis- tence model. Unpublished manuscript, Institute for Social Research, Ann Arbor, Michigan. Andrews, E M. (1984). Construct validity and error components of sur- vey measures: A structural modeling approach. Public Opinion Quarterly, 48, 409-442. Andrews, E M., & Herzog A. R. (1986). The quality of survey data as related to age of respondent. Journal o f the American Statistical Association, 81, 403-4 10. Bobo, L. (1983). Whites' opposition to busing: Symbolic racism or real- istic group conflict? Journal o f Personality and Social Psychology, 45, 1196-1210. Brim, O. G., & Kagan, J. (1980). Constancy andchange in human devel- opment. Cambridge, MA: Harvard University Press. Campbell, A., Converse, P. E., Miller, W. E., & Stokes, D. E. (1960). The American voter. Chicago: University of Chicago Press. Converse, P. E. (1962). Information flow and the stability of partisan attitudes. Public Opinion Quarterly, 26, 578-599. Converse, P. E. (1976). The dynamics o f party support: Cohort-analyz- ing party identification. Beverly Hills, CA: Sage. Cooper, H. M. (1979). Statistically combining independent studies: A meta-analysis of sex differences in conformity research. Journal o f Personality and Social Psychology, 37, 131-146. Crewe, I., Sarlvik, C., & Alt, J. (1977). Partisan dealignment in Britain 1964-1974. British Journal o f Political Science, 7, 129-190. Cutler, N. E. (1974). Aging and generations in politics: The conflict of explanations and inference. In A. R. Wilcox (Ed.), Public opinion and political attitudes (pp. 440-462). New York: Wiley. • Dennis, J. (1973). Socialization to politics: A reader. New York: Wiley. Eagly, A. H., & Carli, L. L. (1981). Sex of researchers and sex-typed communications as determinants of sex differences in influence.abil- ity: A meta-analysis of social influence studies. Psychological Bulle- tin, 90, 1-20. Easton, D., & Dennis, J. (1969). Children in the political system: Ori- gins o f political legitimacy, New York: McGraw-Hill. Franklin, C. H. (1984). Issue preferences, socialization, and the evolu- tion of party identification. American Journal o f Political Science, 28, 459-478. Gergen, K. J., & Back, K. W. (1966). Communication in the interview and the disengaged respondent. Public Opinion Quarterly, 30, 385- 398. Glenn, N. D. (1969). Aging, disengagement, and opinionation. Public Opinion Quarterly, 33, 17-33. AGING AND SUSCEPTIBILITY TO ATTITUDE CHANGE 4 2 5 Glenn, N. D. (1974). Aging and conservatism. Annals o f the Academy o f Political and Social Science, 415, 176-186. Glenn, N. D. (1980). Values, attitudes, and beliefs. In O. G. Brim & J. Kagan (Eds.), Constancy and change in human development (pp. 596-640). Cambridge, MA: Harvard University Press. Greenstein, E I. (1965). Children and politics. New Haven, C'12. Yale University Press. Heise, D. R. (1969). Separating reliability and instability in test-retest correlations. American Sociological Review, 34, 93-101. Hess, R. D., & Torney, J. V. (1967). The development o f political atti- tudes in children. Chicago: Aldine. Hovland, C. I. (1959). Reconciling conflicting results derived from ex- perimental and survey studies of attitude change. American Psycholo- gist, 14, 8-17. Hovland, C. I., & Janis, I. L. (1959). Personality andpersuasibility. New Haven, CT: Yale University Press. Jagodzinski, W., Kuhnel, S. M., & Sehmidt, P. (1987). Is there a "So- cratic effect" in nonexperimental panel studies? Sociological Meth- ods and Research, 15, 259-302. Jennings, M. K. (1988, April). Educational stratification and political development in the Vietnam generation. Paper presented at the annual meeting of the Midwest Political Science Association, Chicago. Jennings, M. K., & Markus, G. B. (1984). Partisan orientations over the long haul: Results from the three-wave political socialization panel study. American Political Science Review, 78, 1000-1018. Jennings, M. K., & Niemi, R. G. (1978). The persistence of political orientations: An overtime analysis of two generations. British Journal o f Political Science, 8, 333- 363. Jennings, M. K., & Niemi, R. G. (1981). Generations and politics. Princeton, NJ: Princeton University Press. Joreskog, IC G., & Sorbom, D. (1978). LISREL: Analysis o f linear struc- tural relationships by the method o f maximum likelihood. Chicago: National Educational Resources. Krosniek, J. A. (1988). Attitude importance and attitude change. Jour- nal o f Experimental Social Psychology,, 24, 240-255. Krosnick, J. A., & Alwin, D. E (1989). Symbolic versus non-symbolic political attitudes: Is there a distinction? Unpublished manuscript, Ohio State University, Columbus. Mannheim, K. (1952). The problem of generations. In P. Kecskemeti (Ed.), Essays on the sociology o f knowledge (pp. 276-322). London: Routiedge & Kegan Paul. Markus, G. B. (1979). The political environment and the dynamics of public attitudes: A panel study. American Journal o f Political Science, 23, 338-359. Mueller, J. E. (1973). Wars, presidents, and public opinion. New York: Wiley. Newcomb, T M. (1943). Personality and social change: Attitude forma- tion in a student community. New York: Holt. Newcomb, T. M., Koenig, K. E., Hacks, R., & Warwick, D. P. (1967). Persistence and change: Bennington College and its students after 25 years. New York: Wiley. Nunn, C. Z., Crockett, H. J., & Williams, J. A. (1978). Tolerance for nonconformity. San Francisco: Jossey-Bass. Petty, R. E., & Cacioppo, J. T. (1979). Issue involvement can increase or decrease persuasion by enhancing message-relevant cognitive re- sponses. Journal o f Personality and Social Psycholog?z, 37, 1915- 1926. Ragosa, D. (1988). Myths about longitudinal research. In K. W. Schaie, R. T Campbell, W. M. Meredith, & S. C. Rawlings (Eds.), Method- ological issues in aging research (pp. 171-209). New York: Springer. Riley, M. W. (1973). Aging and cohort succession: Interpretations and misinterpretations. Public Opinion Quarterly, 37, 35-49. Saris, W. E., & Putte, B. V. D. (1988). True score or factor models. Sociological Methods and Research, 17, 123-157. Searing, D., Wright, G., & Rabinowitz, G. (1976). The primacy princi- ple: Attitude change and political socialization. British Journal o f Po- litical Science, 6, 83-113. Sears, D. O. (1975). Political socialization. In E I. Greenstein & N. W. Polsby (Ede.), Handbook o f political science (pp. 96-136). Reading, MA: Addison-Wesley. Sears, D. O. (1981). Life stage effects on attitude change, especially among the elderly. In S. B. Kiesler, J. N. Morgan, & V. K. Oppenhei- mer (Eds. ), Aging: Social change (pp. 183-204). New York: Academic Press. Sears, D. O. (1983). The persistence of early political predispositions: The roles of attitude object and life stage. In L. Wheeler (Ed.), Review o f personality and social psychology (Vol. 4, pp. 79-116). Beverly Hills, CA: Sage. Sears, D. O. (1986). College sophomores in the laboratory: Influences of a narrow data base on social psyehology's view of human nature. Journal o f Personality and Social Psychology,, 51, 515-530. Stember, H. (1951-1952). Which respondents are reliable? Interna- tional Journal o f Opinion and Attitude Research, 5, 474--479. Wheaton, B., Muthen, B., Alwin, D. E, & Summers, G. E (1977). As- sessing reliability and stability in panel models. In D. R. Heise (Ed.), Sociological methodology 1977 (pp. 85-136). San Francisco: Jossey- Bass. Wiley, D. E., & Wiley, J. A. (1970). The estimation of measurement error in panel data. American Sociological Review,, 35, 112-117. Wolfinger, R. E., & Rosenstone, S. J. (1980). Who votes? New Haven, C12. Yale University Press. Wood, W. (1982). Retrieval of attitude-relevant information from mem- ory: Effects on susceptibility to persuasion and on intrinsic motiva- tion. Journal o f Personality and Social Psychology, 42, 798-810. Zaller, J. R. (1987). Diffusion of political attitudes. Journal o f Personal- ity and Social Psychology 53, 821-833. Received N o v e m b e r 14, 1988 R e v i s i o n received M a r c h 27, 1989 A c c e p t e d M a r c h 27, 1989 •