Micro-foundations of Politics

Chapter 3 54 Micro-foundations of Politics Introduction In this chapter, I will be examining several aspects of individual-level political behavior that are theoretically related to the moral foundations. In Chapter 1, I outlined a general theoretical model that has many implications for the role of predisposing factors in political behavior. If these predispositions are the building blocks of political attitudes, there are certain correlational patterns we should expect to see in observational data. In this chapter, I will explore the relationship between the moral foundations and specific political identities, attitudes, and behaviors in more depth. Throughout the chapter, I will be paying special attention to the interaction between group identification the moral foundations. As I will describe in more detail below, we might expect the moral foundations to operate differently for individuals who identify with one of the major political parties. The chapter proceeds first by outlining the theoretical expectations. I develop expectations with respect to my measures of political predispositions relating to three domains: political identities (partisanship and ideology), political attitudes (interest

55 and issue positions), and political behavior (vote choice). These three domains constitute the dependent variables that I will examine in this chapter. First, I examine the role of pre-political dispositions in structuring and contributing to the stability (or instability) of partisan and ideological identities. The analysis moves from the fundamental concepts of identity to the perhaps more malleable realm of attitudes. Specifically, I examine general interest in politics and specific issue positions across an array of different domains. The final domain that I examine is the role of the pre-political dispositions in explaining vote choice. I look at both defections in general elections and evaluations of primary election candidates (all in the context of the 2008 presidential election). I then describe the data sets that I will be using in the chapter and move on to the results. The analysis in this chapter draws on panel data, bolstering some of the causal claims that I am making, and is replicated in several respects across two different datasets. I conclude the chapter by recapitulating the main findings and suggesting some areas for future research and further study. The Microfoundations of Politics For many structural and historical reasons, politics in the United States is divided into two major partisan factions. This basic fact of American politics may be the most important institutional factor influencing the political landscape within which individuals in the mass public find themselves. When this binary system is imposed upon a messy set of individual-level predispositions that exist in five dimensions, we can see where problems might begin to develop. If

56 the politicians are doing their job well, they will attempt to stake their claim to territories within this five-dimensional space that are the most populous, but at the end of the day, it will be impossible to draw any plane through that space that cleanly separates the public into two equal halves. Consequently, each of the major parties is composed of often uneasy moral coalitions. Political Identities: Partisanship and Ideology Political identities are central to almost all contemporary theories of political behavior, but we know surprisingly little about their causes. While some interesting work exists that attempts to identify the causal processes underlying individuals identification with one party or ideological label over another (Gerber, Huber, and Washington 2010b), political scientists have generally lacked the theoretical tools to talk about the roles of specific predisposing factors play in shifting individuals toward one identity over another. Using panel data, I will show how the moral foundations measures affect individual s identities toward one or partisan or ideological label or another. Partisanship is the single political identity that has received the most attention among political scientists. The degree to which individuals identify with a political party is a powerful predictor of the extent to which they will support candidates who also affiliate with the party, and an exhaustive literature exists documenting the consequences of party identification in almost every aspect of individual political behavior and attitudes (Green, Palmquist, and Schickler 2004; Johnston 2006; Groenendyk 2013). However, we know less about the origins of partisan attachments. The authors of The American

57 Voter and subsequent generations of scholars assumed parental transmission was the primary factor in explaining party identification in the mass public 1 Later research showed high correlations between the partisan attachments of parents and children (Niemi and Jennings 1991b). Most of the research in political science pertaining to change in partisan identity focuses on changes in the short or long term fluctuations in party attachment. 2 Unfortunately, the bulk of the political science research on the topic seems to have devolved into extended methodological debates about the proper analysis of time-series data (MacKuen, Erikson, and Stimson 1989; Schickler and Green 1993; Green and Palmquist 1994; Box-Steffensmeier and Smith 1996). 3 Most of the conversation surrounding partisan stability focuses on the role of retrospective evaluations of party performance on valence issues (e.g. economic performance, scandals, war). As I argued in Chapter 1, these kinds of evaluations are important, but they can only take us so far. In this chapter, I will present evidence that individual predispositions explain part of the variation in party identification at the individual level. Parties and partisan elites go to great lengths crafting arguments that appeal to basic values and predispositions. At times, these appeals may serve as cover for shifting policy positions that can be best explained by shifting interest coalitions within each party (Karol 2009), but the lengths to which partisans go to craft coherent arguments 1 Although, this argument runs into obvious problems if it is traced back too far. Parental transmission does not explain the origins of the parents political identities, and at some point we run into the question of the prime mover. 2 And, given the paucity of long-term panel studies in political science, scholars have relied (for the most part) on indirect evidence gleaned from studies of aggregate trends in cross-sectional data. 3 Groenendyk (2013) provides an example of work that does a great deal to move the theory of partisan identification forward while not getting distracted by methodological debates.

58 suggests that they are concerned with maintaining at least the appearance of consistency (Noel 2012). Individuals in the mass-public are largely ignorant of the powers and interests behind the scenes that are causing the changes within the parties, but they do respond to the arguments put forward by them (Carsey and Layman 2006). Comparatively less work has been done with ideological self-identity. Indeed, for many years political scientists held that ideological self-identity was so unimportant in the mass public that they did not ask individuals about it. It wasn t until 1972 that an ideological self-placement was regularly asked on the ANES, and there is an on-going debate as to what individuals mean when they describe themselves in ideological terms (Conover and Feldman 1981; Treier and Hillygus 2009; Jost, Federico, and Napier 2009; Weber and Federico 2013). To the extent that the words liberal and conservative are a part of the public debate in politics, individuals are capable of formulating evaluations of them and selfidentifying with one label or another. These self-descriptions need not conform with well-reasoned politico-philosophical definitions of the term to be meaningful. Symbolic ideology can tell us something even when it does not match up with one s operational positions. 4 In this chapter, I will treat ideological self-identification in much the same way as partisan attachment. The only difference being that ideologies do not benefit from the existence of formal institutions to ground their definitions with a stable group. Importantly for my argument, political identities should be affected in relatively uniform ways by predispositions. Individuals in the mass public should apply their predisposing values in the same way when evaluating the parties and ideological labels 4 For more on the current state of the meaning and impact of ideological identification, see Carmines and D Amico (2015).

59 available to them. For example, if one party appeals to individuals who place greater emphasis on the Sanctity domain, all individuals in the mass public who are paying attention and are themselves sensitive to Sanctity concerns should gravitate toward that party everything else held constant. Because these are social identities, inertia plays a large role, and we should not expect overly much change. Political Attitudes In the case of political attitudes, we should expect more variation and responsiveness to particular predispositions. A primary function of political parties and political elites is to describe specific issues in terms of general predispositions. Political actors frame specific issues in ways highlight specific elements of the problem that are most favorable toward their preferred position. Because of their important role in framing political issues, I expect to see different reactions to at least some issues for partisans. For example, I will show how the parties seem to activate moral concern differentially for some issues. In some cases, this takes the form of identifiers from one party activating moral concern to a greater extent than those from the other. In the case of raising taxes on the wealthy, for example, I will show that Republicans seem to connect the issue to a broad range of moral concern in ways that Democrats do not. In other issues, we see differential moral responses that are most likely a function of different arguments, and in other issues we see more commonality than difference.

60 Political Interest Political interest and attention are key components of individual political attitudes. Indeed, one s general posture toward the political world (whether he or she finds it interesting and relevant enough to engage with at all) is a precondition for most other political attitudes and behaviors. The political science literature is replete with examples of the importance of these fundamental attitudes, but (again) we know less about their origins. Several studies have shown demographic correlates of political interest and attention (Lane 1959; Bennett and Bennett 1989; Galston 2001). Education and socioeconomic status are consistently associated with political interest, and there are obviously some important preconditions of political interest that can be viewed more as resources. My interest lies in the motivations. Given two individuals of equal intellectual capacity, what explains the residual variation in political interest? One implication of the general argument I am developing is that political interest should be itself a function of political predispositions. To the extent that individuals find themselves in line with the predominant moral divides emphasized and reinforced by elite actors in the parties, they should be more likely to engage with politics in an intuitive way. On the other hand, people whose predispositions are not well represented in the political system will show less interest in political matters. This represents one way in which one s political predispositions might interact with one s political identity. Issue Positions With a few exceptions, 5 political scientists have had relatively little to say with regard to the precursors of individual issue positions. Indeed, there is an increasing trend to use 5 See, for example, Goren (2012).

61 issues as indicators of some underlying ideology without too much worrying into where they came from in the first place (Bafumi and Herron 2010; Jessee 2009b; Treier and Hillygus 2009). These ideology scales composed of issue placements are useful in certain applications, but they obviously cannot be used to explain positions on the very issues on which they are based. A focus on individual predispositions allows a way around this problem. In this view, determining one s position on a particular issue requires only a reaction to the arguments being made about it. We know that individuals are biased toward arguments made by their co-partisans (Rahn 1993), so I also expect these reactions to be filtered through partisan lenses. Intra-party Heterogeneity The two party system layered on top of multidimensional predispositions in the mass public ensures conflict between party positions and the intuitions of partisan supporters. While I expect partisans to be more susceptible to arguments from their own party, there is enough moral diversity within each major party to cause conflicts between official party stances and where individuals predispositions are leading them. The moral foundations measures should account for part of the diversity of opinion within the parties. Political Behavior Beyond affecting identities and attitudes, individuals predispositions are also expected to have consequences for their behavior. In this chapter, I will focus on vote choice, but I would expect predispositions to play a role in other behaviors as well. The prepolitical dispositions that are the centerpiece of this dissertation are linked closely to

62 emotions, and other research has shown the importance of emotion for political action (Marcus and MacKuen 1993; Valentino et al. 2011; Weber 2013). Vote Choice: Partisan Defections The moral foundations can also help us understand why individuals defect from their party s nominee. Even after accounting for differences in self-reported ideology, partisanship, and other demographic factors, measures of individual predispositions tell us something about vote choice (and abstention). Vote Choice in Primary Elections Of course, partisanship dominates general election decisions, but in primary elections partisan labels serve no heuristic function as all candidates share them. Primary elections are opportunities for partisans to explore the moral divides within their own parties, and candidates seeking to secure a nomination must walk a careful line between appealing to the moral concerns of their co-partisans while not positioning themselves too far out of the mainstream of the national constituency. I expect that the moral foundations will explain variation in ratings of primary candidates for their parties over and above what can be explained by strength of partisan affiliation, ideological identity, or demographic factors. A Note on Context Nearly all of the data that I rely on in this chapter was collected during the 2008 election campaign. This surely has consequences for the results that I demonstrate. It is difficult to identify the exact effects of these campaign-specific contextual factors without

63 a great deal more comparative work. However, it is important to remember that a great deal more than just the campaign occurred over the course of the year. Ultimately, the only way to resolve the issue will be parallel studies in different campaign contexts. Data For the purposes of this chapter, I will be drawing on two of the core datasets described in Chapter 2. The measures of the pre-political dispositions will also be the same as those derived in Chapter 2. Individual scores on the moral foundations items are derived from a response model applied to the moral foundations questionnaire shown (in Chapter 2) to produce measures that fit the data better and are less inter-correlated than traditional summation measures. 2008 ANES Panel Although the sample size is limited and the moral foundations questionnaire is missing several items, the ANES sample has several things to recommend it. First, it is designed to be a representative sample of the general population. As noted earlier, the other data that I rely upon comes from self-selected non-probability samples. Having a representative sample of the general public allows greater confidence in generalizing the findings. Secondly, the panel structure of the data allow me to look at how an individual s predispositions affect the stability of political identities and attitudes in ways that are not possible with cross-sectional data. For the purposes of the analysis in this chapter, I look at changes in political identities between the January and November waves.

64 YourMorals Database: MFQ30 A subset of individuals (around 20,000) who filled out the moral foundations questionnaire also completed a set of political issue items. These issue items included several follow-ups designed to measure the degree to which individuals feel their attitudes on particular issues are related to their core moral values and convictions. This dataset is missing several key demographics (most notably a good measure 6 of self-reported partisanship), and it is based on a non-probability sample. However the large sample size and the uniqueness of the moral follow-up items connected to individual issues makes this dataset valuable. Results Political Identities The ANES sample allows us to examine the effect of the moral foundations measures on the stability of political identities. 7 I expect to find that being within the moral majority of one s party increases the stability of affiliation while occupying a marginal position will decrease stability. For example, if I affiliate with the Democratic party in the first wave of the survey (January 2008), but I also place a great deal of emphasis 6 Individuals were asked an open-ended item to measure their partisan affiliation. The modal partisan responses are variations of Democrat and Republican, but by far the most common response is nonresponse or some version of independent. 7 Unfortunately, the moral foundations questionnaire was only administered at one time point. It would be ideal to have measures of both political identities and the MFQ at multiple time points to explicitly model the interrelationships between them. Lacking such measures, there is less I am able to claim about the causal relationship between an individual s moral predispositions and his or her political identities. Instead, I am forced to lean upon the theory and the empirical work with the MFQ that suggests that they tap into stable individual differences that are conceptually prior to identities like partisanship (Graham et al. 2011). This assumption has recently been challenged (Smith et al. 2016).

65 on the Sanctity domain, the probability that I continue to identify as a Democrat (or identify as strongly with the Democratic Party) in November should decrease. This is analogous to the concept of cross-pressures (Powell Jr 1976; Hillygus and Shields 2014). Partisanship Table 3.1 shows a simple tabulation of the partisan identity items at each time point. A glance at the table shows the stability of party identification. The vast majority of cases hew closely to the diagonal, and the bulk of the movement happens within one scale-point on the seven point scale. Despite the stability, there is change in partisan identification over time. 8 Table 3.1: Partisan Stability: January to November November Str. D Wk. D Ln. D Ind. Ln. R Wk. R Str. R Str. D 162 27 3 1 0 0 0 Wk. D 46 62 12 8 3 5 4 Ln. D 18 19 47 15 3 0 2 January Ind. 2 9 16 65 14 10 2 Ln. R 0 3 4 10 59 17 6 Wk. R 2 9 3 8 11 84 48 Str. R 0 0 0 0 8 25 139 Source: 2008-2009 ANES panel survey Table 3.2 shows the results of the stability analyses pertaining to partisanship. The table reports simple linear regressions with partisan identification in November as the response variable. 9 In the first column of the table, I have reported a simple model 8 It bears noting that most of this change happens within the different gradations of identification with one of the major parties. For example, the 18 individuals who went from leaning toward the Democratic Party in January to being strong identifiers in November. 9 Due to the limited nature of the response variable, OLS is not the correct model here, and the

66 that lumps all individuals in the sample together and does not include any demographic controls. This model shows the now familiar relationship between the moral foundations and political identity. Even after controlling for partisanship in the January wave, individuals who placed more emphasis on the Care and Fairness foundations were more likely to move toward greater affiliation with the Democratic Party. On the other hand, those who placed greater emphasis on the Respect and Sanctity domains moved toward greater affiliation with the Republican Party. As evidenced by the coefficient on partisan affiliation in the January wave, partisan identity is extremely stable. The second column of the table adds a set of control variables for several demographic factors (age, education, sex, political attention, and religious attendance). We can see that the moral foundations measures are robust to including these demographic controls. Still the Care and Fairness foundations pull individuals toward the Democratic Party and the Respect and Sanctity foundations exert an opposite pull toward the Republican Party. The demographic controls are very weak factors in partisan stability. They often have inconsistent effects and are generally statistically indistinguishable from zero. For my purposes, the third through fifth columns provide the most direct test of my claims. These subset the data to examine potentially differential effects on each side of the political spectrum. Column 3 looks at the effect of the moral foundations on partisan stability only among the set of respondents that identified with (excluding those who only leaned toward) the Democratic Party. Among Democrats, the effects of the moral foundations are almost universally stronger than in any other grouping results in the table are robust to more appropriate models that probably more accurately describe the data-generating process. However, for ease of explication, I have chosen to present the OLS results. The main advantage is the relative ease of interpretation. In the OLS setting, the effect sizes are readily interpretable in a way that is not always true of other models.

67 of the data. In 2008 at least, the moral foundations seem to be a better predictor of partisan change among those on the political left than the political right. 10 Column 4 shows the results among those who identified with the Republican Party in January. Although the coefficients are in the expected direction, none of the moral foundations measures are statistically significant. The last column in Table 3.2 shows the results among those who did not identify with one of the major parties in the January wave. Interestingly, this set of respondents was the most stable in their partisan identification as evidenced by the size of the coefficient on party identification in January. 11 The pattern of coefficients for political independents was very similar as seen for Democrats. 10 It bears reiterating that this provides a rather difficult test for my theory. Given the stability of political identity (especially among the subset of individuals who actually identify with a party) and the relatively short time frame under examination, there is not much variance left to explain. 11 This can most likely be explained by the fact that this subset was the only among those examined that straddled the divide between the two parties. Others have shown the partisan leaners are just as partisan in their behaviors as partisan identifiers (Keith et al. 1992), so it makes sense that they would actually be less likely to move in their partisanship. Unfortunately, the sample size is too small to focus exclusively on pure independents for whom we might expect the most movement.

68 Table 3.2: Moral Foundations and Partisan Stability Dependent variable: Party Identification in November All (no controls) All (controls) Dem. Only Rep. Only Ind. Only (1) (2) (3) (4) (5) PID (Jan.) 0.897 0.883 0.920 0.987 1.041 (0.017) (0.018) (0.117) (0.124) (0.084) Care 0.194 0.199 0.280 0.039 0.273 (0.057) (0.059) (0.100) (0.100) (0.110) Fairness 0.126 0.121 0.152 0.028 0.145 (0.044) (0.045) (0.077) (0.084) (0.080) Loyalty 0.024 0.037 0.177 0.042 0.182 (0.076) (0.079) (0.129) (0.140) (0.147) Respect 0.256 0.232 0.231 0.249 0.221 (0.078) (0.081) (0.133) (0.154) (0.143) Sanctity 0.146 0.159 0.214 0.080 0.133 (0.057) (0.062) (0.101) (0.124) (0.115) Constant 0.088 0.058 0.474 0.469 0.295 (0.041) (0.198) (0.419) (0.484) (0.333) Observations 991 954 320 325 309 R 2 0.797 0.798 0.287 0.225 0.442 Adjusted R 2 0.796 0.795 0.259 0.196 0.420 Note: p<0.1; p<0.05; p<0.01 Models (2)-(5) include unreported controls for education, age, sex, religious attendance, political attention, and race.

69 For individuals who identified with the Democratic Party in January, the effect of the Care foundation is greater than it is those who identified with the Republican Party in January. Substantively, this means that individuals who began as Democrats in January who were especially high (or low) on the Care foundation moved more toward (or away from) affiliation with the Democratic Party in December than was observed among the Republican Respondents. Put differently, Republican identifiers in January seemed insensitive to the pull exerted by the Care foundation on their partisan identification later in the year. The models in Table 3.2 are OLS, so the coefficients can be interpreted as the effect of a one standard deviation increase in a given foundation leading to a change on the 7-point party identification scale equivalent to the coefficient size. A onestandard deviation change in the Care foundation for Democrats was associated with a change of nearly a third of a scale point on the party identification scale. An equivalent change for Republicans was associated with less than one twentieth of a scale point change on the party identification scale. To test whether the observed differences between the effect sizes among the different subpopulations were themselves statistically significant, I conducted a simple bootstrapping test. 12 For the subset of Republican and Democratic respondents separately, I re-estimated the models in columns 3 and 4 by sampling with replacement from the set of Republicans and Democrats. I repeated this procedure 1,000 times retaining the estimated coefficients on the moral foundations measures. The bootstrapping procedure simulates independent samples from the sampling distributions of the coefficients, allowing a direct test of the difference between the coefficients for the Republicans and the 12 As the models were not nested (e.g. they were estimated based on different sets of individuals), it is not possible to test the equality of the effect sizes with more traditional approaches like the t-test.

Figure 3.1: Plots the distributions of the differences in the coefficients for each foundation estimated separately for the Republicans or Democrats. These plots allow us to test whether the differences in the estimated coefficients are significantly different across the different models estimated. In these plots, there is strong evidence that there are differences in the Care foundation between Democrats and Republicans and weaker evidence that there are differences in the Loyalty foundation. The other foundations are not significantly different in their effect sizes between the two different subpopulations. 70

71 Democrats. The results (densities of the differences between the two coefficients) are shown in Figure 3.1. When it comes to the influence of individual pre-political dispositions on partisan identities, the evidence seems to favor a uniform effect of the different dimensions across different subpopulations. With the possible exception of the Care domain, the other moral foundations did not seem to operate differently based on earlier partisan attachment. Ideology The ideology questions were similarly assessed at multiple points during the panel study allowing a parallel analysis with ideological identity as the dependent variable. Table 3.3 shows a tabulation of the January and November items. In a similar fashion to partisan identification, ideology was assessed in a branching question. The stability of ideological identity was not as high as partisanship, but it is still impressively stable (at least in the short term). Table 3.3: Ideological Stability: January to November November V. Lib Lib S. Lib Mod S. Con Con V. Con V. Lib 50 15 4 4 0 1 1 Lib 22 72 20 18 2 13 0 S. Lib 6 36 41 15 10 5 1 January Mod 3 13 21 107 24 30 6 S. Con 0 2 7 15 31 30 6 Con 3 8 8 15 27 139 33 V. Con 0 0 0 1 4 27 96 Table 3.4 is shows for ideological identification what Table 3.1 showed for partisanship. The models include the same covariates (swapping out January partisanship with

72 January ideology) with the same caveats as before, and we see a very similar pattern of results in the first and second columns of the table. The Care and Fairness domains are associated with movement in toward the liberal end of the spectrum, and the Respect and Sanctity foundations are associated with movements toward the conservative end. Moving into the analysis of ideological subgroups the story changes somewhat. Sanctity is robustly associated with conservativism for every grouping. The other foundations show less consistency. For self-identified conservatives in January, the Care domain seems to be associated with increased liberalism as the campaign progresses.

73 Table 3.4: Stability of Ideological Identity and the Moral Foundations Dependent variable: Ideological Identity in November All (no controls) All (controls) Lib. Only Cons. Only Mod. Only (1) (2) (3) (4) (5) Ideo (Jan.) 0.742 0.725 0.656 0.969 0.934 (0.023) (0.024) (0.198) (0.124) (0.089) Care 0.243 0.211 0.075 0.255 0.177 (0.064) (0.066) (0.146) (0.098) (0.112) Fairness 0.097 0.101 0.184 0.035 0.048 (0.050) (0.051) (0.150) (0.081) (0.081) Loyalty 0.063 0.053 0.011 0.155 0.008 (0.085) (0.088) (0.203) (0.135) (0.145) Respect 0.222 0.198 0.229 0.195 0.062 (0.088) (0.090) (0.212) (0.139) (0.150) Sanctity 0.382 0.380 0.567 0.219 0.219 (0.066) (0.071) (0.158) (0.112) (0.126) Constant 0.021 0.048 0.829 1.006 0.063 (0.046) (0.221) (0.639) (0.425) (0.342) Observations 992 955 214 348 393 R 2 0.639 0.644 0.228 0.266 0.301 Adjusted R 2 0.637 0.639 0.182 0.240 0.279 Note: p<0.1; p<0.05; p<0.01 Models (2)-(5) include unreported controls for education, age, sex, religious attendance, political attention, and race.

74 Figure 3.2 tests the extent to which the coefficients for liberals and conservatives are truly different from on another. Contrary to the asterisks in Table 3.4, the figure shows that there is no statistically distinguishable difference between the estimated coefficients for the Fairness domain. There is some evidence 13 that there are meaningful differences between the two subgroupings in the conditional relationship between ideological stability and the Sanctity domain (although the estimated effects run in the same direction, the coefficient for political liberals was consistently larger than that for conservatives). As with partisan identity, ideological identity seems to be affected in roughly similar ways for political conservatives and political liberals. Although the evidence is by no means conclusive, I am encouraged by these results. They suggest that for political identities the moral foundations operate similarly for all people. As we will see in the analyses below, this is not always the case. As detailed above, I expect there to be interactive effects between individual predispositions and political identities. Political Attitudes In the previous section, I presented some evidence that predisposing factors affect political identities. In addition to shaping identities, the moral foundations have been shown to correlate with political attitudes and reactions to political arguments (Clifford and Jerit 2013; Kertzer et al. 2014; Feinberg and Willer 2013). 13 In 949 out of 1,000 bootstrapped draws from the sampling distribution of the coefficients for the model estimated on data from liberals and conservatives, the Sanctity coefficient was smaller in the model estimated on conservatives than the estimate from the model based on liberals.

75 Figure 3.2: Plots the distributions of the differences in the coefficients for each foundation estimated separately for the liberals or conservatives. These plots allow us to test whether the differences in the estimated coefficients are significantly different across the different models estimated. In these plots, there is weak evidence that there are differences in the Sanctity foundation between conservatives and liberals. The other foundations are not significantly different in their effect sizes between the two different subpopulations.

76 Political Interest Again the 2008 ANES panel study provides the necessary data for one test of this proposition. In addition to measuring political identities at multiple time points, individuals were also asked to register their interest in the campaign as it developed. 14 We can look at how the moral foundations are associated with changes in political interest over the course of the campaign to try to understand their role in shaping this important dimension of political attitudes. Table 3.5 shows the results of a straightforward linear regression with political interest in November as the dependent variable and the moral foundations as the explanatory variables. All models control for political interest in January. In Column (1) we can see that political interest is a fairly stable attitude. Individuals who reported high interest in January were most likely among the respondents to report high interest in November as well. Table 3.5 differs from the results relating to political identities in the differential results within the sub-populations. Columns (3) and (4) show the results for Democrats and Republicans respectively. We can see that as the campaign developed, Democrats with high scores in the Sanctity domain shifted toward less political interest. We see a parallel shift away from political interest among Republicans who scored high on the Fairness domain. Tests of the equality of the coefficients from these different subsets allow us to confidently say that the coefficients are in fact different. 14 Individuals were asked, How interested are you in information about what s going on in government and politics? The available response options were, Not interested at all (0), Slightly interested (1), Moderately interested (2), Very interested (3), Extremely interested (4).

77 Table 3.5: Moral Foundations and the Stability of Political interest Dependent variable: Political Interest in November All All Dem. Only Rep. Only Ind. Only (no controls) (w/controls) (w/controls) (w/controls) (w/controls) (1) (2) (3) (4) (5) Political Interest (Jan.) 0.665 0.603 0.575 0.541 0.660 (0.027) (0.029) (0.041) (0.047) (0.053) Care 0.019 0.036 0.102 0.019 0.033 (0.045) (0.047) (0.064) (0.074) (0.086) Fairness 0.010 0.017 0.058 0.154 0.020 (0.035) (0.036) (0.051) (0.062) (0.062) Loyalty 0.055 0.042 0.095 0.075 0.011 (0.061) (0.062) (0.083) (0.098) (0.114) Respect 0.147 0.136 0.146 0.198 0.031 (0.061) (0.064) (0.086) (0.108) (0.112) Sanctity 0.018 0.086 0.131 0.062 0.099 (0.042) (0.050) (0.066) (0.086) (0.090) Constant 0.579 0.076 0.452 0.394 0.535 (0.079) (0.161) (0.209) (0.295) (0.265) Observations 922 867 383 382 280 R 2 0.424 0.452 0.512 0.436 0.499 Adjusted R 2 0.420 0.444 0.496 0.417 0.477 Note: p<0.1; p<0.05; p<0.01 Models (2)-(5) include unreported controls for ideology, education, age, sex, religious attendance, and race.

78 It is tempting to comment as well on the positive association between the Respect foundation and political interest for partisans compared with its near zero coefficient for political independents. The relatively small sample-size among the different subgroups prevents me from saying anything definitive beyond commenting on the suggestive evidence. It is potentially noteworthy that partisans of both stripes are more likely to give greater attention to their political leaders when they hold predispositions that incline them toward respect for authority figures. Political independents who hold no such affiliation do not seem to have the same kind of response. These findings with respect to political interest point toward potential mechanisms by which the moral foundations affect other political attitudes. In the next section, I turn to the effect of political predispositions on issue positions. It seems reasonable that individuals who find their own moral predispositions reinforced by leaders from their party, and thus exhibit greater interest in politics as shown in this section, might be more open to persuasion on particular issues. Contrarily, individuals who are turned off by their own political parties, are less likely to absorb the messaging associated with particular party positions and might more often find themselves even more out of step with their party on individual issue positions. Issue Positions ANES Panel Data The ANES panel included several issue items. I am primarily interested in understanding the extent to which individual predispositions explain diversity within the parties. I expect to see partisans of different sides applying their predispositions differentially to specific issues as they respond to the different substantive arguments that

79 are made on each side of the political debate. Because people are naturally inclined to give more credence to trusted sources, we might expect partisans to respond differently to messaging from both sides. Admittedly, this is an indirect test. It would be more satisfactory to have experimental data that exposes individuals to different arguments to directly gage reactions to specific arguments. A growing literature explores the interaction between issue framing and moral foundations at the individual level (Clifford 2013; Clifford and Jerit 2013; Feinberg and Willer 2013), but less has been done with the interaction between strategically crafted messages and individual political identities. Table 3.6 shows the results for three hotly contested issues. First, attitudes about the legality of abortion. Individuals were asked to register whether they favored or opposed legal abortions across a range of circumstances (rape and incest, health of the mother, financial reasons, health defects, sex of the fetus, etc.). There were seven items in total. Individuals were given a score based on the number of situations they approved of legal abortion. Scores ranged from -7 (legal in all cases) to +7 (illegal in all cases). Second, positions on gay marriage amendments. Gay marriage amendments continued to be an important political issue in 2008. Most visibly, the political fight over Proposition 8 in California was on-going when respondents were asked whether they favored or opposed constitutional amendments designed to limit marriage to heterosexual couples. Follow-up questions were asked to gage the strength of people s opinions, and the items were scaled from -3 (strongly oppose) to +3 (strongly favor). Third, individuals beliefs about the appropriateness of torture in interrogations of suspected terrorists.individuals were asked: Do you favor, oppose, or neither favor nor oppose the U.S. government torturing people, who are suspected of being terrorists, to try to get information? Follow-up questions were asked to gage the strength of people s opinions, and the items

80 were scaled from -3 (strongly oppose) to +3 (strongly favor). In the political issues models in addition to the demographic controls, I include ideological identification on the right-hand side of the models. In the previous sections, ideology was treated as a dependent variable, and it was seen that changes in ideology were related to individual predispositions. Including ideology presents a more difficult test for my theory as it is (both theoretically and empirically) related to pre-political dispositions. Any effects uncovered will most likely understate the true effect of predispositions on particular issue positions.

81 Table 3.6: Issue Positions (ANES Panel): Abortion, Gay Marriage, and Torture Dependent variable: Abortion Gay Marriage Torture Dem. Rep. Dem. Rep. Dem. Rep. Ideology 0.241 0.367 0.488 1.127 0.134 0.171 (0.072) (0.108) (0.106) (0.158) (0.056) (0.089) Care 0.141 0.026 0.358 0.192 0.268 0.186 (0.180) (0.201) (0.266) (0.296) (0.143) (0.170) Fair 0.006 0.113 0.122 0.661 0.188 0.174 (0.146) (0.176) (0.222) (0.252) (0.115) (0.146) Loyal 0.330 0.735 0.844 0.605 0.255 0.069 (0.253) (0.311) (0.383) (0.454) (0.204) (0.254) Respect 0.139 0.027 0.380 1.024 0.155 0.146 (0.249) (0.320) (0.370) (0.461) (0.193) (0.264) Sanctity 0.554 0.473 0.620 1.060 0.689 0.368 (0.183) (0.252) (0.268) (0.356) (0.142) (0.200) Constant 0.314 0.762 0.057 1.181 1.232 0.661 (0.571) (0.806) (0.839) (1.151) (0.449) (0.667) Observations 426 424 451 449 404 387 R 2 0.165 0.109 0.291 0.349 0.271 0.088 Adjusted R 2 0.140 0.083 0.271 0.331 0.248 0.058 Note: p<0.1; p<0.05; p<0.01 All models include unreported controls for education, political attention, age, sex, religious attendance, and race.

82 The first two columns in Table 3.6 show the results for abortion attitudes for Democrats and Republicans respectively. The results for this issue and many of the others show areas of commonality and difference for each party. Both Republicans and Democrats seem to be affected similarly by the Sanctity domain. Holding everything else constant, Republicans and Democrats alike thought abortion should be restricted in more circumstances the higher they scored on the Sanctity domain. There is an interesting difference with the Loyalty domain. Democrats show a positive (though statistically insignificant) trend while Republicans show a strong negative relationship. Increases in the Loyalty dimension are associated with favoring abortion in more circumstances. On its face, this is a somewhat strange relationship, and the Loyalty domain in general behaves a little strangely throughout these analyses. I will defer further discussion of it for later. The next two columns show the results for attitudes about amendments banning gay marriage. Again, we see commonality for both groups of partisans in the Sanctity domain. For this issue there is also an important difference between the two parties in the way in which the Loyalty domain is applied. This time, Democrats show the significant correlation with Republicans trending the other direction. On the issue of gay marriage, Republicans who place particular emphasis on the Fairness foundation were significantly less likely to support efforts to ban gay marriages constitutionally. The final two columns of the table show the effect of the moral foundations measures on attitudes about the torture of suspected terrorists. On this issue, we might have expected the Care foundation to stand out as the most important predictor. While the coefficients are in the expected direction, the statistical evidence is rather weak for a relationship between the Care foundation and attitudes about torture. The stronger association is with the Sanctity domain. For partisans of both major party, increased

83 emphasis on the Sanctity domain is associated with more support for the torture of suspected terrorists.

84 Table 3.7: Issue Positions (ANES Panel): Taxing the Rich, Affirmative Action, and Health Care Dependent variable: Tax Rich Affirmative Action Gov t Health Care Dem. Rep. Dem. Rep. Dem. Rep. Ideology 0.132 0.503 0.018 0.083 0.154 0.387 (0.059) (0.093) (0.059) (0.060) (0.067) (0.090) Care 0.248 0.370 0.039 0.214 0.509 0.364 (0.148) (0.173) (0.147) (0.112) (0.169) (0.167) Fair 0.018 0.373 0.493 0.051 0.112 0.272 (0.121) (0.151) (0.122) (0.095) (0.138) (0.146) Loyal 0.086 0.547 0.100 0.101 0.254 0.690 (0.209) (0.268) (0.211) (0.171) (0.238) (0.259) Respect 0.191 0.721 0.740 0.221 0.145 0.727 (0.206) (0.275) (0.204) (0.174) (0.234) (0.266) Sanctity 0.085 0.164 0.235 0.024 0.129 0.333 (0.151) (0.217) (0.148) (0.135) (0.171) (0.210) Constant 0.436 3.031 0.527 1.440 3.463 2.915 (0.470) (0.694) (0.462) (0.435) (0.535) (0.671) Observations 425 425 451 449 425 426 R 2 0.071 0.228 0.125 0.060 0.125 0.215 Adjusted R 2 0.044 0.205 0.101 0.034 0.099 0.192 Note: p<0.1; p<0.05; p<0.01 All models include unreported controls for education, political attention, age, sex, religious attendance, and race.

85 Table 3.7 shows the results for three more issues: raising taxes on the rich, 15 affirmative action policies, 16, and health care reform. 17 For this set of issues, there was more difference between partisans of both sides. The first two columns of Table 3.7 show factors that affect attitudes toward raising taxes on the wealthy. While there does not appear to be much going on on the Democratic side, this issue touches a lot of moral buttons for Republicans. Care, Fairness, and the Loyalty domains are all associated with more favorability toward increasing taxes on the wealthy. The Respect domain is a powerful mover in the other direction. Unlike the issues from Table 3.6, these issues do not seem to activate the Sanctity concerns to the same degree. Next in Table 3.7 is affirmative action. This issue seems to activate moral concerns for Democrats more than Republicans. The Fairness and Respect domains are the primary factors here, but they do not seem to important in the same way for Republicans. The last columns show the results for government health care. For both parties, the Care dimension is an important factor. Increasing scores on the Care dimension are associated with more support for government provided care. Like the issue of raising taxes on the wealthy, government health care activates a broader array of moral concerns for Republicans. The Fairness and Loyalty domains are associated with increased support, 15 Individuals were asked whether they favored or opposed raising taxes on individuals making more the $200,000 per year. The strength of attitude follow up items were collapsed into a single measure that ranges from -3 (strongly support raising taxes on wealthy) to +3 strongly oppose raising taxes on the wealthy. 16 Individuals were asked: Do you favor, oppose, or neither favor nor oppose allowing universities to increase the number of black students studying at their schools by considering race along with other factors when choosing students? Again follow-up items were collapsed into a seven point scale from -3 (strongly favor affirmative action) to +3 (strongly oppose). 17 The question wording was, Do you favor, oppose, or neither favor nor oppose the U.S. government paying for all necessary medical care for all Americans? The seven point scale ranges from -3 (strongly favor) to +3 (strongly oppose).

86 and the Respect domain is associated with opposition. This pattern of results mirrors that seen in taxing the rich exactly. Indeed even the magnitude of the coefficients are similar. But what is going on with the Loyalty domain? It was associated with liberal attitudes on taxing the rich, government health care, and abortion for Republicans. On the Democratic side, it was associated with more conservative positions on gay marriage. As noted in Chapter 2, this domain perhaps more than any other could use additional development, and it is certainly possible that this strange pattern of findings is an artifact of measurement. Despite the improvements that the model-based measurement of the pre-political dispositions yield, there are still important elements of this domain that seem to be unmeasured. That said, it could also be that the Loyalty domain is associated with more traditional notions of communitarianism. While this does not go all the way toward explaining the strange result on abortion attitudes, it might account for the other results. YourMorals Data The YourMorals issue supplement included the same policy areas as touched on in Table 3.6. Although asked in different ways 18 and to a different kind of sample, we can 18 The abortion item asked individuals to select from a number of different options: Abortion should be generally available to those who want it (coded as 1) Abortion should be available but under stricter limits than it is now (2), Abortion should be against the law except in cases of rape, incest and to save the woman s life (3), and Abortion should not be permitted at all (coded as 4). The gay marriage item similarly asked respondents to select the statement that best selected their own view from: Same-sex couples should be allowed to legally marry (1), Same-sex couples should be allowed to have a civil union, but not to marry (2), and Same-sex couples should NOT be allowed to marry nor have civil unions (coded 3). The torture item gave individuals the following options: It is OFTEN justified to use forceful interrogation techniques/torture to get information from a suspected terrorist (coded as 4), It is SOMETIMES justified to use forceful interrogation techniques/torture to get information from a suspected terrorist (3), The use of forceful interrogation techniques/torture is ALMOST NEVER justified (2), and The use of forceful interrogation techniques/torture is NEVER justified (1).

87 compare the pattern of findings shown in Table 3.8. On some points the two tables agree and on some points they disagree. For example, Democrats in both samples who placed greater emphasis on the Sanctity domain were more likely to support restrictions on abortion access. In the YourMorals sample, the Care domain was important for abortion attitudes in a way that it was not in the ANES panel. There are a host of potentially confounding factors that could be driving these differences. As mentioned, the samples are very different. The ANES panel is a probability sample where the YourMorals data is a convenience sample. The questions are different, and the set of available controls are different or at times were measured in different ways.