jost_j_t_kay_a_c_thorisdottir_n_social_and_psychological_bas
.pdf
272 |
PERSONALITY AND INDIVIDUAL DIFFERENCES |
|
|
has documented the relationship between motivation and ideology without examining potential boundary conditions or moderators, such as citizens’ awareness of elite discourse. Of course, the strength and robustness of the relationship between epistemic motivation and ideology in previous studies suggest that a sizable number of citizens do manage to connect motivation with politics, regardless of their individual characteristics or features of the situations they find themselves in. Nevertheless, these results tell us little about which segments of the population account for this relationship and why. Moreover, the few studies that have looked at moderators of the relationship between epistemic motivation and political attitudes have not directly addressed the hypothesis advanced here. For example, recent studies on moderators of the relationship between the need for closure and attitudes toward political conflict have not focused on ideology as a dependent variable (e.g., Federico, Golec, & Dial, 2005; Golec & Federico, 2004). Similarly, a recent analysis of archival data on American foreign policy officials by Kemmelmeier (2007) found that cognitive rigidity is more strongly associated with conservatism among those with a greater interest in politics. However, this study focused on a small, self-selected sample of individuals who were already far more involved in politics than the average citizen (i.e., government officials) and relied on indirect measures of the need for closure (i.e., cognitive rigidity) and political expertise (i.e., frequency of political discussion with friends and family). Thus, the question of whether of the relationship between epistemic motivation and ideology is moderated by attention to and understanding of elite discourse (as provided by political expertise) remains a largely open one.
OVERVIEW
In this chapter, we provide a test of the hypothesis that political expertise should moderate the relationship between epistemic motivation and ideology. Using data from two relatively recent surveys, we look at whether the political impact of the epistemic motivation most often linked to ideology— the need for closure (Jost et al., 1999, 2003a)—varies as a function of political expertise. In each data set, we expected the relationship between the need for closure and a composite index of general ideological orientation to be stronger among those individuals presumably most attuned to and likely to have absorbed elite political discourse—that is, those high in political expertise.
DATA AND METHODS
The first of the two data sets used in this study is a 2000 student survey. The data for this study came from a study of undergraduates at the Univer-
Motivated Social Cognition and Ideology |
273 |
|
|
sity of California, Los Angeles (UCLA; N = 221). Participants were recruited through the undergraduate subject pool, and received partial course credit for their participation. However, since much of the extant research on the relationship between epistemic motivation and ideology has relied on student samples (see Jost et al., 2003a), we wanted to replicate our analyses in a sample taken from a population more likely to vary extensively in political expertise. Therefore, the second data set we used comes from a national survey of adults: the 2006 Core Predispositions Study (CPS). The 2006 CPS (N = 1201) interviewed respondents using a representative sample of the U.S. population in the 48 contiguous states between February 1 and April 26 of that year. All interviews for this survey were conducted via telephone.1 The measures derived from each data set are described below.
Variables from the 2000 UCLA Student Survey. In the 2000 UCLA student survey, our key epistemic motivation—the need for closure—was operationalized using the full 42-item Need for Closure Scale (Webster & Kruglanski, 1994). Examples of items included “I don’t like situations that are uncertain,” “Any solution to a problem is better than remaining in a state of uncertainty,” and “I prefer activities where it is always clear what is to be done and how it needs to be done.” Participants responded to each item on a 6-point scale ranging from “Strongly Agree” to “Strongly Disagree.” All items were recoded, so that high scores indicated a high need for closure (a = .85; M = 3.59, SD = .51).
In turn, our moderator variable, political expertise, was measured using standard factual items, which are regarded as the most valid indicators of ex- pert–novice differences in political cognition and awareness of elite discourse (Delli Carpini & Keeter, 1996; Fiske, Lau, & Smith, 1990; Zaller, 1992). Sixteen items were used: “What job or political office does Al Gore currently hold?”; “What job or political office does William Rehnquist currently hold?”; “What job or political office does Tony Blair currently hold?”; “What job or political office does Dennis Hastert currently hold?”; “Which political party currently has the most members in the Senate in Washington D. C.?”; “Which political party currently has the most members in the House of Representatives in Washington D. C.?”; “How long is the term of office for a U.S. Senator?”; “How many justices are there on the U.S. Supreme Court?”; “How many times can a president be re-elected?”; “Whose responsibility is it to nominate judges to the federal courts—the President, the Congress, or the Supreme Court?”; and
1 In the 2006 CPS, the response rate was 11.6%. Although low by historical standards, this rate is similar to those reported for other recent surveys (Pew Research Center for the People and the Press, 2004). All response rates are AAPOR standard definition RR1.
274 |
PERSONALITY AND INDIVIDUAL DIFFERENCES |
|
|
“Which political party do each of the following figures belong to,” with six target individuals: “Bill Clinton,” “John McCain,” “Newt Gingrich,” “George W. Bush,” “Richard Gephardt,” and “Ted Kennedy.” All items were scored on a 0/1 basis, with 0 indicating no answer or an incorrect answer and 1 indicating a correct answer. Each participant’s scores on the 16 items were then summed to create a single index (KR-20 = .75; M = 9.37, SD = 2.96).
To operationalize the key dependent variable, two ideological orientation measures were used. Each of these variables is linked to the general left–right ideological dimension associated with epistemic motivation in previous work (Jost et al., 2003). Liberalism–conservatism was assessed using respondents’ self-placement on a 7-point ideology scale, similar to the one used in the National Election Studies (NES) (M = 3.32, SD = 1.31). Partisanship was assessed using respondents’ self-identification on a comparable 7-point measure of party identification, also similar to the one used in the National Election Studies (M = 3.34, SD = 1.65). Both of these items were coded, so that a higher score indicated a greater ideological tilt to the right and then averaged to form a composite (a = .68; M = 3.32, SD = 1.30).
Finally, four demographics were considered: age (in years), family income (in raw dollar amounts), race (not White = 0, White = 1), and gender (female = 0, male = 1).
Variables from the 2006 Core Predispositions Study. In the 2006 CPS, the need for closure was operationalized using a short six-item form of the Need for Closure Scale (Webster & Kruglanski, 1994). The items were: “In case of uncertainty, I prefer to make an immediate decision, whatever it may be,” “I get very upset when things around me aren’t in their place,” “When I need to solve a problem, I generally do not waste time in considering diverse points of view about it,” “Generally, I do not search for alternative solutions to problems for which I already have a solution available,” “Any solution to a problem is better than remaining in a state of uncertainty,” and “I prefer activities where it is always clear what is to be done and how it needs to be done.” All items used a 4-point scale, ranging from “Strongly Agree” to “Strongly Disagree.” All responses were recoded, so that higher scores indicated a higher need for closure, and the responses were averaged to form a scale. Unfortunately, the reliability of this short measure was considerably lower than that of the full scale (a = .49; M = 2.50, SD = .60); we return to this problem in the analyses below.
Political expertise was again measured using standard factual items. In this survey, the items had a multiple-choice format, with an explicit “don’t know” option. Six items were included, based largely on recommendations made by Delli Carpini and Keeter (1996): “Who has the final responsibility
Motivated Social Cognition and Ideology |
275 |
|
|
to decide whether a law is constitutional?”; “If the president vetoes a law passed by Congress, what kind of a majority is required by Congress to override the veto?”; “What political office is currently held by Dick Cheney?”; “What political office is currently held by Bill Frist?”; “Do you happen to know which political party has the most members in the House of Representatives in Washington D.C.?”; and “Do you happen to know which political party has the most members in the U.S. Senate in Washington D.C.?” These items were scored and summed just as they were in the 2000 UCLA survey. Again, though, the scale produced a lower reliability than was found for the larger scale used in the UCLA data (KR-20 = .54; M = 4.38, SD = 1.39); as with the Need for Closure Scale, we return to this problem below.
Finally, several ideological orientation measures were used. Liberalism– conservatism was measured using respondents’ self-placement on a 7-point ideology scale similar to the one used in the NES (M = 4.43, SD = 1.95), while partisanship was measured using respondents’ self-identification on a 7-point party identification scale similar to the one used in the NES (M = 4.00, SD = 2.16). These two measures were coded so that a higher score indicated a greater tilt to the right and then averaged to form a composite (a = .76; M = 4.21, SD = 1.85).
Finally, five demographics were considered. Four of these were identical to those used in the 2000 student survey: age (in years), income (in $10,000 units), race (not White = 0, White = 1), and gender (female = 0, male = 1). However, since the adult population drawn on in the 2006 CPS obviously differed more in its level of educational attainment, we also created a dummy variable indicating completion of a college degree (no = 0, yes = 1).
RESULTS
In both data sets, we tested our hypothesis using regression methods. More specifically, our analysis involved looking at whether the interaction between the need for closure and political expertise explained variance in respondents’ ideological orientation over and above the main effects of the need for closure and political expertise. In each data set, we began by examining our hypothesis using a series of ordinary least-squares regression models. However, since measures of our key independent variables—the need for closure and political expertise—did not always attain optimal levels of reliability, we supplemented these basic analyses with a series of latent-variable regressions conducted in a structural-equation modeling framework.
Results: 2000 UCLA Student Survey
Intercorrelations between key variables. As a preliminary step, we examined the correlations between our three key variables: the need for closure,
276 |
PERSONALITY AND INDIVIDUAL DIFFERENCES |
|
|
political expertise, and ideological orientation. These analyses indicated that the two main independent variables, the need for closure and political expertise, were not significantly correlated, r(221) = –.04, p > .50. Consistent with prior findings (e.g., Jost et al., 2003a; Jost, 2006), the need for closure had a significant positive correlation with conservatism, r(220) = .27, p < .001. Expertise was also associated with greater conservatism, albeit weakly and nonsignificantly, r(220) = .10, p > .10.
Regression analyses. As noted previously, our key hypothesis is that a high need for closure should be more strongly associated with a right-wing ideological orientation among those individuals who are most familiar with the elite discussions in which ideology is formulated—that is, those high in political expertise. We conducted an initial test of this hypothesis in the UCLA data using a series of ordinary least-squares regressions. In the actual regressions, composite ideological orientation was regressed on the need for closure, political expertise, and Need for Closure × Political Expertise interaction. To control for additional background factors, age, income, race, and gender were also included in the regressions. To guard against possible effects of heteroscedasticity—which may be introduced by the effect of differing levels of expertise on the precision of political choice (e.g., Alvarez & Brehm, 2002)—HC3 robust standard errors were used in the analyses (as generally recommended by Long & Ervin, 2000, for all regression models). Finally, all predictors were centered, as suggested for interactive models (see Aiken & West, 1991).
The results are summarized in Table 11.1. Model 1 examined the main effects of the need for closure and expertise, while controlling for the aforementioned demographics. Replicating the traditional finding in this area, the estimates revealed that the need for closure was strongly and significantly associated with a greater tilt to the right (b = .70, p < .001). Moreover, higher levels of expertise were marginally associated with a greater tilt to the right (b = .05, p < .10). None of the other effects reached significance (all p > .10).
In turn, Model 2 added the critical Need for Closure × Political Expertise Interaction to the equation. As expected, this interaction was significant and in the predicted positive direction (b = .12, p < .05). To probe this interaction, simple slopes for the relationship between the need for cognition and composite ideological orientation were computed at expertise levels one standard deviation above and below the latter variable’s mean, using Aiken and West’s (1991) method. These analyses indicated that the relationship between need for closure and composite ideological orientation was positive and highly significant among those high in expertise (b = 1.04, SE b = .27,
Motivated Social Cognition and Ideology |
277 |
|
|
Table 11.1 Interactive effects of the need for closure and political expertise on composite ideological orientation: Regression analysis (2000 UCLA Student Survey).
|
|
Composite Ideological Orientation |
|
|
|
MODEL 1 |
MODEL 2 |
||
|
|
|
|
|
Predictor |
b |
SE b |
b |
SE b |
|
|
|
|
|
Age |
–.01 |
(.04) |
.002 |
(.04) |
Income |
.0000005 |
(.000001) |
.0000004 |
(.000001) |
Race (1 = white) |
.07 |
(.23) |
.11 |
(.23) |
Gender |
–.02 |
(.18) |
–.06 |
(.18) |
Need for closure |
.70*** |
(.21) |
.69*** |
(.19) |
Political expertise |
.05+ |
(.03) |
.05+ |
(.03) |
Need for closure × Political |
|
|
|
|
expertise |
– |
– |
.12* |
(.06) |
Constant |
3.46*** |
(.87) |
3.38*** |
(.83) |
F (df) |
2.27 (6, 202)* |
2.69 (7, 201)* |
|
|
R2 |
.088 |
|
.106 |
|
N |
209 |
|
209 |
|
|
|
|
|
|
Note. Entries are unstandardized OLS regression coeffi cients and HC3 robust standard errors.
+p < .10. *p < .05. **p < .01. ***p < .001.
p < .001), but nonsignificant among those low in expertise (b = .35, SE b = .25, p > .10). A graph of this interaction pattern, based on the preceding ordinary least squares (OLS) estimates, is shown in the left panel of Figure 11.1. In sum, then, the regression analyses confirm our primary hypothesis: although the need for closure is indeed associated with a stronger tilt to the right, this relation is more pronounced among those high in political expertise.
Latent variable regressions. Although these results are instructive, OLS regression estimates can be biased by measurement error. This concern is particularly relevant when relationships are compared across levels of expertise, a variable that can affect the level of error with which key predictors are measured (see Goren, 2004; Judd, Krosnick, & Milburn, 1981; Judd & Milburn, 1980). Thus, we opted to replicate our OLS results using a series of interactive latent-variable regression models implemented via structural-equation modeling. To this end, we used Little, Bovaird, and Widaman’s (2006) technique for estimating latent variable models that include interaction terms. To minimize the number of parameters we needed to estimate, we began by combining the need for closure items and the expertise items into three “item
278 |
PERSONALITY AND INDIVIDUAL DIFFERENCES |
|
|
2000 UCLA Student Survey
Orientation |
5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||
4 |
|
|
|
|
|
|
|
|
|
Ideological |
3 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
|
|
|
|
|
|
|
|
|
|
|
Low need |
High need |
|
|
|
|
for closure |
for closure |
|
|
|
|
|
|
|
|
|
|
High Expertise |
Low Expertise |
|
|
|
|
||
|
|
|
|||
|
|
|
|
|
|
2006 Core Predispositions Study
Orientation |
5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||
4 |
|
|
|
|
|
|
|
|
|
Ideological |
|
|
|
|
|
|
|
|
|
3 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||
|
|
|
|
|
|
|
|
||
|
|
|
|
|
|
|
|
|
|
|
2 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Low need |
High need |
|
|
|
|
for closure |
for closure |
|
|
|
|
|
|
|
|
|
|
High Expertise |
Low Expertise |
|
|
|
|
||
|
|
|
|||
|
|
|
|
|
|
Figure 11.1 Moderation of the relationship between the need for closure and ideological orientation by political expertise in the 2000 UCLA Student Survey and the 2006 Core Predispositions Study (predictions based on OLS model estimates).
parcels” each, which were then mean-centered.2 These centered item parcels were then used as the three respective indicators of the need for closure and expertise. Next, indicators for the latent interaction were generated by
(a) computing each pairwise combination of the centered need for closure and expertise indicators, (b) regressing each of the nine resulting product terms on all centered first-order latent variable indicators (i.e., the three need for closure parcels and the three expertise parcels), and (c) saving the residuals from each of these nine regressions as the “orthogonalized” scores for each individual on the nine latent-interaction indicators. This procedure produces a series of latent-interaction indicators that are purged of any variance shared with the indicators of the first-order term, eliminating the need for complex nonlinear constraints in the actual structural-equation model (Little et al., 2006).
Using these indicators, we estimated a series of structural equation models in which ideological orientation was regressed on latent variables corresponding to the need for closure, expertise, and the Need for Closure × Expertise interaction. The estimates were obtained in Mplus 4.1 (Muthen & Muthen, 2007), using Satorra and Bentler’s (1994) mean-corrected maximum
2 The three UCLA need for closure parcels were created by (1) averaging items 1–14, (2) averaging items 15–28, and (3) averaging items 29–42, whereas the three expertise parcels were created by (1) averaging items 1–6, (2) averaging items 7, 8, 9, 10, and 11, and (3) averaging items 12, 13, 14, and 15. Particular sets of items were chosen so as to maximize the internal consistency of each.
Motivated Social Cognition and Ideology |
279 |
|
|
likelihood technique and robust standard errors to correct any non-normal- ity in the observed indicators. The measurement scale of each latent independent variable was established by fixing the variance of each to unity. The first-order latent variables for the need for closure and expertise were allowed to correlate, but the correlations between the interaction and each first-order latent variable were fixed to zero since the indicators for the interaction are already purged of variance shared with the first-order latent variable indicators. Finally, since we have only two available indicators of ideological orientation, the composite ideological orientation variable was entered into the model as an observed variable in order to avoid convergence problems (Kline, 1998).
The results of these analyses are summarized in Table 11.2; estimates for the measurement models are not shown. Paralleling the OLS analyses, two models were run. The first constrained the effect of the latent interaction to zero, approximating a “main effects” model; the second freely estimated the path coefficient for the effect of the interaction as well. As the results indicate, both models produced showed an excellent global fit (with nonsignificant chi-squares and CFIs equaling 1). Replicating the OLS results, Model 1 indicated a significant first-order effect of the need for closure (γ = .39, p < .001), with those high in the need for closure showing greater conservatism; and a marginal effect of expertise (γ = .18, p < .10), with those high in expertise also showing a slightly stronger tilt to the right. More importantly, the Model 2 estimates indicated a significant effect of the latent interaction (γ = .21, p < .05). To probe this interaction, methods described by Aiken and West (1991) were used to compute simple slopes for the relationship between the need for closure and ideological orientation at high and low levels of expertise.3 These analyses indicated that the relationship between need for closure and ideological orientation was significant and relatively large among those high in expertise (γ = .60, SE γ = .14, p < .001), but smaller and nonsignificant among those low in expertise (γ = .18, SE γ = .11, p > .10). Thus, our structural-equation analyses—which disattenuate the effects of the independent variables for measurement error—perfectly replicate our
3 In Aiken and West’s (1991) method, values for “high” and “low” levels of a moderator are usually obtained by taking values one standard deviation above and below the mean of the centered moderator (i.e., 0). In the present case, equivalent values were obtained by taking the square root of the estimated variance of the latent variable corresponding to expertise. This variance was fixed to 1 in the models, meaning that the standard deviation of latent expertise also comes to 1. Thus, values of 1 (high expertise) and –1 (low expertise) were used in the calculations.
280 |
PERSONALITY AND INDIVIDUAL DIFFERENCES |
|
|
Table 11.2 Interactive effects of the need for closure and political expertise on composite ideological orientation: Structural equation model (2000 UCLA Student Survey).
|
Composite Ideological Orientation |
|||
|
MODEL 1 |
MODEL 2 |
||
|
|
|
|
|
Predictor |
γ |
SE γ |
γ |
SE γ |
|
|
|
|
|
Need for closure |
.39*** |
(.08) |
.39*** |
(.08) |
Political expertise |
.18+ |
(.09) |
.18+ |
(.09) |
Need for closure × Political expertise |
– |
– |
.21* |
(.10) |
Satorra-Bentler χ2 (df ) |
17.75 (84), ns |
14.37 (83), ns |
||
CFI |
1.00 |
|
1.00 |
|
R2 for ideological orientation |
.107 |
|
.134 |
|
N |
209 |
|
209 |
|
|
|
|
|
|
Note. Entries are unstandardized mean-corrected maximum-likelihood estimates with robust standard errors.
+p < .10. *p < .05. **p < .01. ***p < .001.
OLS analyses.4 Taken together, the UCLA survey results provide a clear pattern of support for our hypothesis.
Results: 2006 Core Predispositions Study
Although the results from the UCLA student study are thus supportive, we also wanted to provide evidence for our hypothesis using data from a sample of adults drawn from the general population. Here, our goal was to obtain greater confidence about the generalizability of our results to the citizenry at large. To do so, we turned to the 2006 CPS data. In this data set, we pursued an analytic strategy similar to the one used with the UCLA data. Unfortunately, the CPS data confronted us with far more serious measurement problems than the UCLA data. As noted previously, the need for closure and political expertise items did not form scales that were
4 To ensure that our structural-equation results were robust with respect to estimation procedure, we replicated this analysis using a second method for the testing of interactions between latent variables: Klein and Moosbrugger’s (2000) latent moderated structural (LMS) model approach, as implemented in Mplus 4.1 (Muthen & Muthen, 2007). This method also revealed a significant interaction between the need for closure and expertise (γ = .22, SE
γ = .09, p < .05), suggesting that our original result was not an artifact of the procedure used.
Motivated Social Cognition and Ideology |
281 |
|
|
adequately reliable according to conventional criteria. Therefore, our dualtrack strategy—in which both OLS and latent-variable techniques were used to run the regressions—was particularly useful in this case.
Intercorrelations between key variables. Prior to further analysis of the 2006 CPS data, we again examined the correlations between our three key variables. In contrast to what we found in the 2000 UCLA data, the need for closure and political expertise were significantly and negatively correlated, r(1197) = –.20, p < .001, such that those who were high in need for closure tended to be lower in expertise. As before, though, the need for closure had a significant positive correlation with conservatism, r(1182) = .13, p < .001, whereas expertise was essentially uncorrelated with ideological orientation, r(1186) = .04, p > .15.
Regression analyses. As before, we began by examining our hypothesis using ordinary least squares regressions. These regressions were similar to those run using the UCLA data: ideological orientation was regressed on the need for closure, political expertise, and Need for Closure × Political Expertise interaction, with age, income, race, gender, and completion of a college degree included as demographic controls. Again, HC3 robust standard errors were used, and all predictors were centered prior to the analyses. The results are summarized in Table 11.3. As before, Model 1 looked at the main effects of the need for closure and expertise. Again confirming earlier results, the estimates indicated that the need for closure was strongly and significantly associated with a greater tilt to the right (b = .43, p < .001), despite the low reliability of the need for closure measure.5 Of the other predictors, income (b = .04, p < .01), White racial identification (b = .94, p < .001), and male gender (b = .22, p = .05) were significantly associated with greater conservatism. None of the other first-order effects reached significance (all p > .10).
Model 2 added the Need for Closure × Political Expertise interaction, which proved to be significant and in the predicted positive direction (b = .19, p < .001)—again, despite the error involved in measuring the need for closure and expertise. This interaction was broken down by computing simple slopes for the relationship between the need for closure and composite ideological orientation one standard deviation above and below the mean of the expertise variable. Replicating the pattern found in the UCLA data, these estimates indicated that the relationship between need for closure and ideological
5 The effect of measurement error in independent variables is to bias OLS coefficient estimates downward, meaning that the error actually works against us finding large effects of the need for closure or its interaction with expertise (see Cohen, Cohen, Aiken, & West, 2003).