The Dimensional Structure of Symbolic Ideology: An Experiment on Liberal-Conservative Self-Placements Suggested running head: An Experiment on Liberal-Conservative Self-Placements Abstract It has been argued that political ideology consists of more than one dimension when the concept is used to explain policy preferences. These arguments are based on analyses of policy preferences that utilize dimension-reduction techniques to find at least two dimensions of the liberal-conservative scale at work—most frequently social and economic dimensions. However, no one has demonstrated whether individuals think of their ideological identifications in two dimensions. Do respondents’ provide different self-placements for economic issues as compared to social issues? This paper uses data from a national survey experiment that directly measures the social and economic dimensions of ideology to determine whether respondents think of their ideological views and report their self-placement differently on social and economic issues; and whether the two self-placement measures of social and economic ideology are more accurate predictors of policy preferences than the single measure. The analysis provides evidence that the experimental measures offer some advantages over the unidimensional measure, particularly for explaining the preferences of individuals who are not strong ideologues. On its face, ideology is not a concept that lends itself to easy operationalization and empirical use; it is a complex concept that is not easily measured (e.g., Converse 1964; Luskin 1987). According to the original definition of the term, ideology is an individual’s view of how society should work, and in the specific case of political ideology, it is a view of how politics and government should work (Roucek 1944). As Rosenberg (1988) defined it, ideology is “not simply a set of learned preferences. More basic, it is a way of making sense of politics—of defining who and what is involved, what they do, and how they relate to one another.” The terms most associated with ideologies in American politics, “liberal” and “conservative,” are often used in common language, and the labels carry with them a basic understanding of the preferences and ideals implied by the use of the label. The labels, and corresponding beliefs, are typically thought of as a unidimensional scale with extremely liberal views at one end, extremely conservative views at the other end, and moderate views in the middle. This unidimensional scale, or some variation of the liberal-conservative structure, has been used to evaluate individuals’ political views, policy preferences, and many aspects of political life for decades. Because liberal-conservative ideology is so often used in models of behavior, the measurement of the construct has been an intermittent subject of debate in the literature. One of these critical debates has been whether liberal and conservative views are truly unidimensional. Recently, some scholars have begun to revisit the dimensionality of liberal-conservative views, a debate that had some traction in earlier decades (e.g., Altemeyer 1981; Chong, McClosky and Zaller 1983; Maddox and Lilie 1984) but had not seen much work until a few years ago (e.g., Feldman and Johnston 2009; Treier and Hillygus 2009; Zumbrunnen and Gangl 2008). The basic argument is that political ideology consists of more than one dimension when the concept is used to explain policy preferences. To support this argument, scholars have used 2 factor analysis, item response theory, and a host of other dimension-reduction techniques to derive individuals’ placements on the latent ideological dimensions from a battery of policy preference questions. The general finding is remarkably consistent: there are at least two dimensions of liberal-conservative ideology at work with respect to policy preferences—most frequently social and economic dimensions (e.g., Altemeyer 1998; Feldman and Johnston 2009; Haidt, Graham, and Joseph 2009; Jost et al. 2009; 2003; Layman and Carsey 2002; Swedlow 2008; Swedlow and Wyckoff 2009;Treier and Hillygus 2009; Zumbrunnen and Gangl 2008). This work has established that policy preferences are multidimensional, but what has not been addressed is whether individual-level liberal-conservative self-identification, as measured using the traditional liberal-conservative scale survey question, is multidimensional. This project uses data from a nationwide survey experiment to determine whether respondents will provide different liberal-conservative self-placements when given the opportunity to place themselves on the scale for economic and social issues in two different questions. More substantively, the analysis will address the question of whether the twodimensional self-placements explain individuals’ policy preferences better than the commonlyused unidimensional measure. This research shows that some individuals do place themselves differently on the liberal-conservative continuum for the two dimensions and that these two measures do result in some improvement in preference prediction. The evidence supports the overall conclusion that individuals are able to perceive and report differences in their views on social and economic dimensions. Why should we expect different self-placements? The American politics literature has established two distinct types of liberal-conservative 3 ideology for use in models of political behavior: symbolic ideology, which is what individuals believe their liberal-conservative ideology to be in abstract terms and based on self-reports; and operational ideology, which is a researcher-created concept of what individuals’ ideological placements are based on their measured policy positions (e.g., Abramowitz and Saunders 2008; Campbell et al 1960; Conover & Feldman 1981; Converse 1964; Jacoby 1991, 2002; LewisBeck et al 2008; Peffley and Hurwitz 1985; Sears et al 1979; Sears et al 1980; Swedlow 2008). The traditional spatial configuration of symbolic ideology—as operationalized in survey research—asks a single survey question to get respondents to place themselves on a unidimensional continuum from strongly liberal to strongly conservative. Operational ideology measures individuals’ positions on the liberal-conservative scale using respondents’ reported policy preferences, a method which does not rely on respondents having specific knowledge of what it means to be liberal or conservative. The idea that policy preferences exist along multiple dimensions is not a new one, but the multidimensional structure of ideology has developed almost completely within the parameters of operational ideology. Evidence across many studies shows that in conceptual and factoranalytic terms, political ideology has two distinct dimensions with regard to policy preferences(e.g., Altemeyer 1998; Feldman and Johnston 2009; Haidt et al 2009; Jost et al. 2009; Maddox and Lilie 1984; Swedlow 2008; Swedlow and Wyckoff 20009; Treier and Hillygus 2009; Zumbrunnen and Gangl 2008). Ideology at the mass level is dependent upon the policies about which individuals form preferences and therefore any dimensions of ideology in the mass public will be defined according to the dimensions of the policies that are under consideration (Maddox and Lilie 1984). Since Ellis and Stimson (2009) have demonstrated that how an individual perceives their own ideological placement (symbolic ideology) can be quite different 4 from where their ideological placement is based on their policy positions (operational ideology), these findings cannot be generalized out to the symbolic ideology measured by liberalconservative self-placements. This literature does, however, demonstrate how to approach measuring liberal-conservative self-placement in multiple dimensions. The consensus in the multidimensional discussion is that policy issues, as they relate to liberal-conservative political ideology, are generally divided by social and economic lines. Maddox and Lilie (1984) define these policy dimensions as “attitudes toward government intervention in the economy” and “attitudes toward the maintenance or expansion of personal freedoms.” By these definitions, the dimensions Maddox and Lilie (1984) use are similar to the dimensions of “capitalism” and “democracy” articulated by Chong, McClosky, and Zaller (1983): capitalism corresponds to government intervention in the economy, and democracy corresponds to the maintenance or expansion of personal freedoms. Later work that refers to “liberalism” or “communitarianism” (e.g., Swedlow 2008; Swedlow and Wyckoff 2009), uses those terms to refer to two of the categories created by using two dimensions of ideology in a Cartesian space; the dimensions that divide the space are social and economic ideology. In more recent studies, Feldman and Johnston (2009) used the social and economic dimensions of ideology as the jumping-off point to factor-analyze the structure of individuals’ issue positions into more precise categories. They found that a two-factor model of ideology was a better fit for the data than a one-factor model, and the two factors were identified as economic and social dimensions of ideology. Treier and Hillygus (2009) use Bayesian Item Response Theory to analyze policy preferences in terms of ideology, arriving at the same conclusion— there are distinct cultural and economic dimensions of ideology. The implication of measuring ideology using only one dimension if two dimensions have more explanatory power is that the 5 research could be ignoring powerful relationships between ideology and policy preferences that are more complex than previously thought (Feldman and Johnston 2009; Haidt et al 2009; Treier and Hillygus 2009). Most importantly for applying the results to symbolic ideology, these findings have demonstrated that some individuals are cross-pressured between different views on different types of political issues (Treier and Hillygus 2009). As a result of this cross-pressuring, individuals may consider themselves conservative on one set of issues and liberal on another set of issues. This pattern is easily demonstrated using policy preferences to measure liberal and conservative views, as Treier and Hillygus (2009) and Johnston and Feldman (2009) did, but crosspressures cannot be identified using the unidimensional survey measure of liberalconservative self-placement. Anyone who is cross-pressured will likely respond in the middle of the scale (“moderate”) or use the don’t know category since “liberal” and “conservative” fail to accurately describe all of their views (Treier and Hillygus 2009). This response pattern results in a “muddy middle” group whose policy preferences are unexplained by their liberal-conservative self-placement. By accounting for the two different types of operational ideology, economic and social, the cross-pressured individual’s policy preferences can be more clearly explained. Given that research demonstrates the existence of multiple underlying dimensions of operational ideology driving responses to policy preference questions, a logical question to ask is whether individuals conceptualize their liberal-conservative self-placement as different on social and economic dimensions. In other words, is symbolic ideology multi-dimensional? If the answer to this question is affirmative, one would expect these cross-pressured individuals to give different self-placements on items asking about different dimensions of policy views. For example, a cross-pressured individual might report their symbolic ideology as “liberal” for social 6 issues, but “slightly conservative” for economic issues. These measures would then provide better explanation for the individuals’ policy preferences since their symbolic ideology is allowed to vary by policy type. The dimension of the policy can be matched to the symbolic ideology dimension—e.g., a model explaining preferences on a policy generally perceived to be economic in nature would use the economic liberal-conservative self-placement measure as the indicator of ideology in the model. The result of this should be an improvement in the explanatory power of the liberal-conservative measure in the model since the two dimensional measures are more accurate indicators of symbolic ideology than the unidimensional measure. Data and Methods The dimensional structure of liberal-conservative self-placements is tested using data collected by Time-Sharing Experiments for the Social Sciences.1 The survey was a nation-wide experiment conducted by Knowledge Networks using their KnowledgePanel® internet survey panel, which is designed to generate nationally representative web samples.2 The experimental portion of the survey incorporated a simple random-split design that sent respondents to one of two ideology question tracks. The first track, the control, simply asked respondents the traditionally-used ideological self-placement question, worded as: “On a scale of political ideology, individuals can be arranged from strongly liberal to strongly conservative. Which of the following categories best describes your views?” Respondents were provided the traditional seven-point scale with values ranging from strongly liberal to strongly conservative and an alternative response of “I don’t think of my views in these terms” for those respondents who did not wish to place themselves on the continuum. Approximately half of the sample (N=558) was 1 2 NSF grant 0818839, Jeremy Freese and Penny Visser Principal Investigators Total N of 1108, Cooperation rate of 68.7% by AAPOR COOP1 rate. 7 randomly assigned to this control track. The experimental track, to which the remaining 550 respondents were assigned, consisted of three questions about the respondents’ ideological views. The first two, which appeared in random order, were designed to measure the social and economic dimensions of ideology separately. Definitions of social and economic issues were provided in order to reduce the potential for confusion among respondents regarding what was meant by these different types of issues and to explain the terms to respondents who do not often think about the nature of issues. The economic ideology question read: “On a scale of political ideology, individuals can be arranged from strongly liberal to strongly conservative. When thinking about your views on economic issues, which of the following categories best describes your views? ‘Economic issues’ are questions of how to distribute resources among people within a society.” The social ideology question simply changed “economic” to “social” in the text of the question, and defined social issues as “problems that affect many or all members of society, and often involve cultural or moral values.” The third question in the experimental track simply asked respondents to place themselves on the same ideological scale generally: “Now, considering your responses to the previous two questions, which of the following categories best describes your views overall?” All three questions had the same response options as the control track question: the standard seven-point scale that ranges from strongly liberal to strongly conservative, with the additional option of “I don’t think of my views in these terms.” Bivariate Analysis The first step in analyzing the effects of asking two different liberal-conservative selfidentification questions is to look at response patterns—did respondents in the experimental track 8 select different placements for social and economic issues? Table I indicates that at least some respondents did answer the experimental questions differently. Although the majority of responses in the crosstab do fall into the diagonal boxes indicating that the respondents chose the same placement for both questions, a nontrivial number of respondents fall into the off-diagonal boxes that indicate different responses, as indicated by the bottom row and rightmost column of the table. Overall nearly one-third (31%) of all respondents provided different self-placements on the two questions, and the highest rates of providing different responses are generally found among those who placed themselves in the “slightly” liberal or conservative categories (with the exception of those in the social strongly liberal category, but the low N in this category likely contributes to the higher volatility). [Table I here] Table II provides more information on the distance between placements for each respondent. The frequency shown is the absolute value of the number of categories between each respondent’s self-placement on the economic views question and the social views question. For example, a response of “slightly liberal” on the economic question and “liberal” on the social question would be a difference of 1; answering “conservative” on both questions would be a difference of 0. The respondents who indicated 0 difference between their economic and social self-placements constitute just under 70 percent of the sample. The remaining 31 percent indicated that there is some difference between their economic and social liberal-conservative self-placements, as shown in Table I. By far the most common difference in positions is only one category, but there is a handful of respondents that place their views as far away as four categories—that would be the difference between answering “strongly conservative” on one dimension and “slightly liberal” on the other. 9 [Table II here] Now that we know there are differences in placement for some individuals, the question becomes whether those differences matter when looking at aggregate statistics. Again, the answer to this question seems to be affirmative—the differences in individual-level placement do cause significant changes in aggregate statistics. The mean responses for all four questions, the control plus all three experimental items, are clustered around 4 (moderate), but the differences are enough to be significant in almost every pairwise comparison. The mean placement for the control group is 4.21, and all of the experimental question means are slightly more bent toward the conservative end of the scale, with the social views mean at 4.38, the economic views mean at 4.54, and the overall mean predictably in between the social and economic means at 4.43. Table III shows a matrix of the differences between the means. The differences between the control mean and all three of the experimental measures is statistically significant (p<0.05), as the first column shows. The experimental economic and social measures are significantly different from one another despite a high correlation (0.83), and the difference between the experimental economic and experimental overall measure is significant despite a slightly higher correlation (0.89). The only difference that fails to reach statistical significance is between the experimental social measure and the experimental overall measure (correlation 0.90). [Table III here] It is interesting that all three of the experimental measures are significantly different from the control in the conservative direction. It would seem that by prompting respondents to think about the scale in terms of policy positions when placing themselves on the liberal-conservative continuum, the question pushes respondents toward the conservative end of the scale. If the question wording does bias respondents, then we should see that the experimental measures are 10 less effective in explaining policy preferences than the control. Conversely, if the experimental measures provide a closer approximation of respondents’ true placement on the liberalconservative scale than the control, models using the experimental measures should explain policy preferences better. With this empirical question in mind, in addition to the theoretical question of how respondents conceptualize liberal-conservative self-placement, we turn to models of preferences. Multivariable Analysis Three commonly used policy preference measures are used as the dependent variables for constructing models to test the control and experimental measures. Two of the items asked respondents to place themselves on seven-point scales, with the simple text “Where would you place yourself on this scale?” The first dependent variable, regarding government services, labels the scale where one means “government should provide many fewer services,” four means “government provides the right amount of services,” and seven means “government should provide many more services.” The second dependent variable, government-guaranteed jobs, labels only the anchor points: one means that “government should make sure that people have jobs and a minimum standard of living,” and seven means that “government should let each person get ahead on their own.” The third dependent variable is respondents’ opinion about abortion. The question asks “Which of these statements best describes your opinion regarding abortion?” with the options of “By law, abortion should never be permitted; The law should permit abortion only in case of rape, incest, or when the woman’s life is in danger; The law should permit abortion for reasons other than rape, incest, or danger to the woman’s life, but only after the need for abortion has been clearly established; By law a woman should always be able 11 to obtain an abortion as a matter of choice.” This variable was recoded to a dichotomy in which 0 means that the respondent believes abortion should never be allowed, and 1 means that the respondent believes abortion should be allowed under at least some circumstances. The text and answer options for all three questions was taken from the 2004 American National Election Study. Table IV shows the relevant statistics from two models for each dependent variable: one that uses the control liberal-conservative self-placement measure and another using the experimental social and economic measures. (Full models are included as appendix tables.) There is not much difference between the control and experimental models, but the patterns are informative. The control liberal-conservative measure is consistently significant across all three dependent variables, as expected. Interesting patterns emerge in the models using the experimental measures, however. Both economic and social self-placement are significant predictors of the government services dependent variable, which is an intuitive finding since the services provided by government are a mix of social (e.g., public education) and economic (e.g., welfare programs). In the other models, only one of the two experimental measures is a significant predictor of the dependent variable, and the significant one is exactly what would be expected given the policy area of the variables. For the question of how respondents feel about government-ensured jobs, only the economic measure is significant—jobs are clearly viewed as an economic issue. The opposite is true when predicting whether respondents think abortion should be allowed: only the social measure is significant, indicating that abortion is viewed as a social issue. These patterns show that respondents not only place themselves differently on economic and social dimensions, but that they weight the two dimensions according to the policy issue in question. Self-placements on an economic liberal-conservative scale correspond with 12 opinion about predominantly economic issues, and likewise for the social scale and issues. [Table IV here] This finding is not entirely surprising; after all, the respondents were primed to think of economic and social issues in the experimental liberal-conservative placement questions. However, it is worth noting that the experimental measures predict the dependent variables at magnitudes equal to or higher than the control measure. The magnitude of the liberalconservative measure effects is roughly the same between experimental and control measures for the government services and jobs dependent variables, but social ideology is a much larger predictor of abortion preferences than the unidimensional measure. The magnitude comparisons indicate that the two-dimensional measures are at least equally valuable for policy preference prediction, and in some cases could be more influential than the traditional unidimensional question. Moreover, the link between the “correct” dimension of liberal-conservative selfplacement and the general policy area (abortion, a social issue, is predicted by social ideology) indicates that respondents might connect their liberal-conservative views to actual policy issues more than traditional political science literature often implies (e.g., Converse 1964; Jacoby 1991). A more interesting test of the measures investigates whether the experimental measures significantly contribute to explaining policy preferences where the control measure does not. As mentioned above, some literature argues that cross-pressured individuals will place themselves in the middle of a unidimensional liberal-conservative measure when they have conflicting views on different types of issues (Treier and Hillygus 2009). In addition to the cross-pressure hypothesis, moderates should show different patterns with respect to the explanatory power of liberal-conservative measures because it has been shown throughout time that indicators of 13 ideology do not explain these respondents’ policy preferences very well (e.g., Converse 1964; Feldman 1988; Kinder 1983). Table V shows the same models from Table IV, but this time only for respondents who identified as slightly conservative, middle of the road, slightly liberal, or did not think of their views in those terms. This group of “moderates” was selected using responses to the control liberal-conservative question and the third experimental question that asked respondents to place themselves on the scale overall. It is not surprising that the control liberalconservative measure is not a significant predictor of preferences among moderates for any of the three dependent variables, but the control measures are significant in models that include only non-moderate (or ideologue) respondents. This same pattern can be observed in similar models using ANES data. However, in the models with the experimental measures, at least one liberal-conservative item is a significant predictor of each dependent variable. Economic selfplacement is significant in the government services and government jobs models, and the magnitude of the effect is very similar to the magnitude of the effect in the full aggregate models in Table IV. [Table V here] It appears that using the experimental measures helps explain preferences better than the control measure at least among moderates (see Table V), but how do the different measures affect the overall fit of the models? Looking at the error associated with each model offers little insight into the overall fit. There are no significant differences in the mean or standard deviation of the prediction error from the control models compared to the experimental models for any of the three dependent variables. However, since there are nine or ten independent variables in each model, a relatively subtle change in measurement of one of those variables is not expected to substantially affect the total model error. The change in the measure should affect the error in 14 the model prediction as compared to the ideology variables—how well ideology works as a predictor of the dependent variable. Figure 1 shows a scatterplot matrix of the relationship between the predicted values of each dependent variable and the measure of ideology used in the model for each of the three dependent variables. The plots on the left show this relationship for the control models, and the plots on the right show the relationship for the experimental models. Since both social and economic liberal-conservative measures were used in the experimental models, the most appropriate measure for the dependent variable was used to create the scatterplot. For the first two scatterplots, government services and government-guaranteed jobs, the economic experimental measure was used as the x-axis variable. Social liberal-conservative placement was used for the experimental scatterplot of the abortion dependent variable. [Figure 1 here] A quick left-to-right glance at Figure 1 illustrates the key finding: the experimental measure greatly reduces the noise and error in the comparison between respondents’ predicted values on the policy variables and their ideological placements. Liberal-conservative selfplacement is assumed to have a directional influence on respondents’ policy preferences regardless of other predictors in the models, and this appears to be true for all three of the dependent variables. However, the control scatterplots on the left consistently show a lot of variance in predicted policy view for each category of liberal-conservative self-placement. This means that the control measure, although statistically significant in the models, is not a very strong predictor of the policy preference. The control measure seems to predict preferences best for the government services dependent variable, and worst for the abortion variable. Looking specifically at the three middle categories that represent moderates (3, 4, and 5), we see that the 15 predicted values on the dependent variables cover the majority of the possible variance— supporting the finding that there is no relationship between the control measure and the dependent variables among moderates. The plots on the right-hand side of Figure 1 show a notably closer relationship between the experimental liberal-conservative measures and the predicted policy positions. The directional relationship is very clear at first glance, and the overall variance in predicted policy position for each category of liberal-conservative self-placement is much smaller. The most striking comparison is between the control and experimental measure plots for the predicted probability of supporting abortion. The control plot shows almost no directional pattern, despite the statistical significance of the variable, but the experimental plot shows a very clear pattern of tightly clustered predicted probabilities for every one of the seven points on the social liberalconservative scale. Focusing on the three middle categories of liberal-conservative placement illustrates the differences between the control and experimental measures even further: there is a clear difference in the predicted direction of preferences among slight liberals, moderates, and slight conservatives for all three dependent variables. Discussion The preceding analyses show that individuals are able to conceive of their liberalconservative scale placements in the two separate social and economic dimensions, and that these multidimensional self-placements offer some advantages over the unidimensional measure with respect to the relationship between liberal-conservative ideology and policy preferences. These findings bring up new questions about how individuals cognitively perceive the dimensions—some of which are unanswerable using this data set and some of which we already 16 have clues to the answers. Any conversation about measuring multiple dimensions of ideology must attend to the issue of how individuals “weight” the dimensions—that is, whether the social and economic dimensions of ideology matter equally to individuals, or whether one is more important than another. Certainly this weighting process factors into specific issue considerations: the models above demonstrate that an economic liberal-conservative self-placement is ineffective in predicting preferences on a social issue such as abortion, and social placements are ineffective in predicting preferences on economic issues such as government-guaranteed jobs. The more difficult question is whether individuals consider one dimension as more important than the in their overall assessments of politics and their own political attitudes other. The third liberalconservative question in the experimental track offers some insight into this question by asking respondents to place themselves on the liberal-conservative scale again, this time “considering your responses to the previous two questions.” This wording effectively primed respondents to think about their placements on the social and economic scales in providing their response to this third question. The intention was to have respondents cognitively “average” their previous scale placements to get the answer to this item. There is no way to know whether that is the cognitive process that respondents used to answer the question, but the fact that the mean response to the overall question (4.43) is in between the mean response to the economic and social items (4.54 and 4.38) indicates that some averaging did take place. The relationship between these three means can be interpreted to show that, on average, respondents weighted the social dimension a bit more than the economic dimension in their overall assessments of their liberal-conservative placement. The difference between the social placement mean and the overall placement mean is only 0.05 and not statistically significant (see 17 Table III), whereas the difference between the economic placement mean and the overall placement mean is 0.11 and is statistically significant. This conclusion should be treated very cautiously, however, because it is based on aggregate means and because the survey did not measure the cognitive processes taking place as the respondents answered the overall question in the experimental track. It seems logical that the dimensions would be weighted differently for each person, depending on which issues are most important to them and on the context of that particular point in time. One would expect that economic issues would have been more important than social issues to most respondents in a survey taken in September of 2008; similarly, social issues would be more important than economic issues to a gay or lesbian couple who want to get married in California. Other respondents might simply pick a placement on the overall scale at random without regard for either type of issues. Future work on the multidimensional nature of ideology will need to tackle this problem of how people cognitively process the dimensions. This project supplied the first necessary part—evidence that individuals do perceive differences in the dimensions. Another question that comes up in dealing specifically with liberal-conservative selfplacements is whether differences in placement are artifacts of political knowledge or interest. The data collected do not include measures of political knowledge so it is difficult to test for knowledge effects, but the survey does contain items that serve as indicators of political interest and activity. These items ask respondents how interested they are in politics and public affairs (very interested, somewhat interested, slightly interested, not at all interested), and three yes/no questions that asked if, in the last 12 months, the respondent had contacted a government official, volunteered or worked for a political candidate, issue, or cause, or commented about politics on a political message board or Internet site. The logical hypothesis is that higher levels of interest 18 and activity would indicate higher political awareness in general, and a higher propensity to recognize and report a difference between liberal-conservative placements on social and economic issues. A comparison of the interest and activity measures to the variable indicating the distance between placements on the two experimental scales (distribution shown in Table II) shows that this hypothesis is not supported. No bivariate comparison shows a significant relationship between interest and distance between placements or between activity and distance between placements. In fact, none of the relationships even come close to significant. The correlation between interest and distance is 0.05, and the p-value of a chi-square statistic comparing the two variables is 0.38. Given the lack of relationship, there is no reason to believe that the differences in placements on social and economic dimensions are primarily attributable to political interest or activity. Conclusion This experiment has resulted in two critical findings: first, that at least some individuals will provide different answers when asked to place themselves on separate liberal-conservative scales for economic issues and social issues. Secondly, there is evidence that these separate measures improve the explanatory power of liberal-conservative self-placements for policy preference dependent variables, particularly among moderates. Plenty of evidence exists indicating that preferences are multidimensional by using the preferences themselves to discover the latent underlying dimensions (e.g., Altemeyer 1998; Feldman and Johnston 2009; Haidt et al 2009; Jost et al. 2009; 2003; Layman and Carsey 2002; Treier and Hillygus 2009; Zumbrunnen and Gangl 2008). What was unknown was whether individuals perceived those differences and could express that in liberal-conservative terms, or if the multidimensional structure is simply an 19 artifact of scholarly evaluation and analysis. The experimental data provides evidence for the former—people are capable of conceptualizing their ideological views in social and economic dimensions, and the relationship between policy preferences and liberal-conservative selfplacement is improved by using social and economic measures. Accounting for the differences between social and economic liberal-conservative self-placements allows researchers to account for more of the variance in policy preferences that is due to ideology, providing a more accurate picture of the factors that influence individual policy preferences. 20 References Abramowitz, A. I. and K. L. Saunders. (2008). Is Polarization a Myth? The Journal of Politics 70:542-555. Altemeyer, R. (1981). Right-Wing Authoritarianism. Winnipeg, Canada: University of Manitoba Press. Altemeyer, R. (1998). The Other ‘Authoritarian Personality’. In M. P. Zanna (Ed.) Advances in Experimental Social Psychology (vol. 30), (pp. 47-92). San Diego: Academic Press. Campbell, A., P. E. Converse, W. E. Miller, and D. E. Stokes. (1960). The American Voter. Chicago: The University of Chicago Press. Chong, D., H. McClosky, and J. Zaller. (1983). Patterns of Support for Democratic and Capitalist Values in the United States. British Journal of Political Science13:401-440. Conover, P. J., and S. Feldman. (1981). The Origins and Meaning of Liberal/Conservative SelfIdentifications. American Journal of Political Science 25:617-645. Converse, P. E. (1964). The Nature of Belief Systems in Mass Publics. In D. Apter (Ed.) Ideology and Discontent, (pp. 206-261). New York: Free Press. Ellis, C. and J. A. Stimson. (2009). Symbolic Ideology in the American Electorate, Electoral Studies. 28:388-402. Feldman, S. (1988). Structure and Consistency in Public Opinion: the Role of Core Beliefs and Values. American Journal of Political Science 32: 416-440. Feldman, S., and C. Johnston. (2009). Understanding Political Ideology. Paper presented at the annual meeting of the American Political Science Association, Toronto, Ontario. September 3-6, 2009. Haidt, J., J. Graham, and C. Joseph. (2009). Above and Below Left-Right: Ideological Narratives and Moral Foundations Psychological Inquiry 20:110-119. 21 Jacoby, W. G. (1991). Ideological Identification and Issue Attitudes. American Journal of Political Science 35:178-205. Jacoby, W. G. (2002). Core Values and Political Attitudes. In B. G. Norrander and C. Wilcox (Eds.) Understanding Public Opinion (2nd ed.) Washington, D.C.: Congressional Quarterly Press. Jost, J. T., C. M. Federico, and J. L. Napier. (2009). Political Ideology: Its Structure, Functions, and Elective Affinities. Annual Review of Psychology 60: 307-337. Jost, J. T., J. Glaser, A. W. Kruglanski, and F. J. Sulloway. (2003). Political Conservatism as Motivated Social Cognition. Psychological Bulletin 129:339-375. Kinder, D. R. (1983). Diversity and Complexity in American Public Opinion. In A.W. Finifter (ed.), Political Science: The State of the Discipline (389-425). Layman, G. C. and T. M. Carsey. (2002). Party Polarization and Party Structuring of Policy Attitudes: A Comparison of Three NES Panel Studies. Political Behavior 24:199-236. Lewis-Beck, M. S., W. G. Jacoby, H. Norpoth, and H. F. Weisberg. (2008). The American Voter Revisited. Ann Arbor: University of Michigan Press. Luskin, R. C. (1987). Measuring Political Sophistication. American Journal of Political Science 31:856-899. Maddox, W. S. and S. A. Lilie. 1984. Beyond Liberal and Conservative: Reassessing the Political Spectrum. Washington, D.C.: Cato Institute. Peffley, M. A., and J. Hurwitz. (1985). A Hierarchical Model of Attitude Constraint. American Journal of Political Science 29:871-890. Rosenberg, S. W. (1988). Reason, Ideology, and Politics. Princeton, NJ: Princeton University Press. Roucek, J. S. (1944). A History of the Concept of Ideology. Journal of the History of Ideas 5:479488. Sears, D. O., C. P. Hensler, and L. K. Speer. (1979). Whites’ Opposition to ‘Busing’: Self-Interest of 22 Symbolic Politics? American Political Science Review 73:369-384. Sears, D. O., R. R. Lau, T. R. Tyler, and H. M. Allen, Jr. (1980). Self-Interest vs. Symbolic Politics in Policy Attitudes and Presidential Voting. American Political Science Review 74:670-684. Swedlow, B. (2008). Beyond Liberal and Conservative: Two-dimensional Conceptions of Ideology and the Structure of Political Attitudes and Values. Journal of Political Ideologies 13:157180. Swedlow, B. and M. L. Wyckoff. (2009). Value Preferences and Ideological Structuring of Attitudes in American Public Opinion. American Politics Research 37:1048-1087. Treier, S. and D. S. Hillygus. (2009). The Nature of Political Ideology in the Contemporary Electorate. Public Opinion Quarterly 73:679-703. Zumbrunnen, J., and A. Gangl. (2008).Conflict, Fusion, or Coexistence? The Complexity of Contemporary American Conservatism. Political Behavior 30:199-221. 23 Table I: Economic Liberal-Conservative Self-Identification (rows) by Social LiberalConservative Self-Identification (columns) Strongly liberal Liberal Slightly liberal Moderate/ DK Slightly conservative Conservative Strongly conservative Percent offdiagonal Strongly liberal 10 1 -- 1 -- -- -- 17% Liberal 5 37 4 1 1 -- -- 23% Slightly liberal Moderate/ DK 3 9 23 9 3 -- -- 51% 1 9 14 142 20 7 2 27% Slightly conservative 1 1 4 19 42 6 -- 42% Conservative -- 3 3 5 12 75 8 29% Strongly conservative -- -- 2 1 2 15 42 32% Percent offdiagonal 50% 38% 54% 20% 48% 27% 19% 31% (Total) 24 Table II: Distribution of differences between Responses N % 0 371 68.32 1 122 22.47 2 34 6.26 3 10 1.84 4 6 1.10 543 100.00 Total 25 Table III: Difference of Means Tests Control Economic Social -- 0.33* 0.17* Economic: 4.54 0.33* -- 0.16* Social: 4.38 0.17* 0.16* -- Experimental overall: 4.43 0.22* 0.11* 0.05 Control: 4.21 *Significant at the p<0.05 level 26 Table IV: Liberal-Conservative Measure Coefficients for Three Policy Models Services DV Control Experimental Jobs DV Control Experimental Abortion DV (logit) Control Experimental -0.35 (0.05) -- -0.28 (0.05) -- 0.33 (0.08) -- Experimental Econ. Ideology -- -0.25 (0.07) -- -0.28 (0.08) -- -0.02 (0.12) Experimental Social Ideology -- -0.13 (0.07) -- -0.08 (0.07) -- 0.64 (0.12) 0.34 0.33 0.24 0.30 0.11 0.17 1604.29 1444.64 1660.98 1534.43 -334.37 -299.73 549 537 544 536 546 534 Control Ideology Adjusted/ Psuedo R² Sum of squares /log-likelihood N Bolded coefficients are significant at the p<0.05 level, one-tailed. 27 Table V: Liberal-Conservative Measure Coefficients for Three Policy Models, Moderates Only Services DV Control Experimental Control Ideology Jobs DV Control Experimental Abortion DV (logit) Control Experimental -0.11 (0.14) -- -0.05 (0.15) -- 0.06 (0.19) -- Experimental Econ. Ideology -- -0.25 (0.11) -- -0.32 (0.11) -- -0.06 (0.17) Experimental Social Ideology -- -0.08 (0.10) -- -0.06 (0.10) -- 0.51 (0.16) 0.14 0.15 0.08 0.10 0.04 0.05 766.92 653.78 786.74 689.65 -207.96 -198.43 315 307 311 305 314 304 Adjusted/ Psuedo R² Sum of squares /log-likelihood N Bolded coefficients are significant at the p<0.05 level, one-tailed. 28 Figure 1: Scatterplot Matrix of Predicted Policy Preferences by Liberal-Conservative SelfPlacement, Control and Experimental Models 29 Appendix Table 1: Full Models for All Respondents Services DV Control Experimental Jobs DV Control Experimental Abortion DV (logit) Control Experimental -0.35 (0.05) -- -0.28 (0.05) -- 0.33 (0.08) -- Experimental Econ. Ideology -- -0.25 (0.07) -- -0.28 (0.08) -- -0.02 (0.12) Experimental Social Ideology -- -0.13 (0.07) -- -0.08 (0.07) -- 0.64 (0.12) Black 0.79 (0.21) 0.86 (0.22) 0.03 (0.24) 0.54 (0.24) 0.03 (0.33) 0.57 (0.36) Hispanic 0.52 (0.22) 0.19 (0.20) 0.34 (0.24) 0.28 (0.22) 0.60 (0.35) -0.23 (0.34) Female -0.19 (0.12) -0.30 (0.12) -0.02 (0.13) -0.10 (0.12) -0.28 (0.19) -0.04 (0.20) Education -0.01 (0.07) -0.10 (0.06) 0.07 (0.07) 0.11 (0.06) -0.28 (0.10) 0.05 (0.11) Age -0.00 (0.00) -0.01 (0.00) -0.00 (0.00) -0.01 (0.00) -0.00 (0.01) 0.00 (0.01) Income -0.05 (0.02) -0.08 (0.01) -0.08 (0.02) -0.06 (0.02) -0.07 (0.02) -0.06 (0.03) Party ID 0.20 (0.04) 0.09 (0.04) 0.19 (0.04) 0.13 (0.04) -0.16 (0.06) -0.07 (0.06) Interest -0.16 (0.07) -0.09 (0.07) -0.15 (0.08) -0.13 (0.07) 0.02 (0.11) 0.03 (0.11) Constant 5.46 (0.44) 6.83 (0.46) 5.35 (0.48) 6.09 (0.48) 1.19 (0.68) -1.47 (0.77) Adjusted/ Psuedo R² 0.34 0.33 0.24 0.30 0.11 0.17 1604.29 1444.64 1660.98 1534.43 -334.37 -299.73 549 537 544 536 546 534 Control Ideology Sum of squares /log-likelihood N Bolded coefficients are significant at the p<0.05 level, one-tailed. 30 Appendix Table 2: Full Models, Moderates Only Services DV Control Experimental Control Ideology Jobs DV Control Experimental Abortion DV (logit) Control Experimental -0.11 (0.14) -- -0.05 (0.15) -- 0.06 (0.19) -- Experimental Econ. Ideology -- -0.25 (0.11) -- -0.32 (0.11) -- -0.06 (0.17) Experimental Social Ideology -- -0.08 (0.10) -- -0.06 (0.10) -- 0.51 (0.16) Black 0.65 (0.26) 0.78 (0.28) 0.10 (0.28) 0.31 (0.30) 0.16 (0.37) 0.18 (0.44) Hispanic 0.51 (0.29) 0.17 (0.26) 0.59 (0.31) -0.06 (0.28) 0.69 (0.43) -0.14 (0.40) Female -0.12 (0.17) -0.33 (0.16) 0.11 (0.18) 0.01 (0.17) -0.37 (0.24) -0.36 (0.24) Education 0.02 (0.09) -0.05 (0.08) 0.08 (0.10) 0.21 (0.09) -0.15 (0.13) -0.02 (0.13) Age -0.00 (0.01) -0.00 (0.01) -0.00 (0.01) -0.01 (0.01) 0.00 (0.01) -0.00 (0.01) Income -0.06 (0.02) -0.07 (0.02) -0.10 (0.02) -0.06 (0.02) -0.08 (0.03) -0.04 (0.03) Party ID 0.24 (0.05) 0.06 (0.05) 0.13 (0.05) 0.07 (0.05) -0.05 (0.07) -0.09 (0.07) Interest -0.09 (0.10) -0.12 (0.09) -0.08 (0.10) -0.07 (0.09) -0.06 (0.14) 0.01 (0.14) Constant 4.19 (0.78) 6.41 (0.64) 4.72 (0.83) 6.14 (0.72) 1.44 (1.12) -0.48 (1.00) Adjusted/ Psuedo R² 0.14 0.15 0.08 0.10 0.04 0.05 766.92 653.78 786.74 689.65 -207.96 -198.43 315 307 311 305 314 304 Sum of squares /log-likelihood N Bolded coefficients are significant at the p<0.05 level, one-tailed. 31