Democracy in America - Partisanship, Polarization Pt 3

3.4 Resisting the Menu of Manipulation29 When examining our respondents’ willingness to punish candidates that violate democratic principles, we have so far pooled all democracy positions into two groups, neutral and undemocratic positions. We now examine the differences in Americans’ willingness to tolerate the distinct ways in which the individual undemocratic positions violate democratic principles and interpret them in light of several benchmarks. We estimate the following linear model: Pr(i votes for candidate 1) = α + X βk (Xi1k − Xi2k ) + ij . (5) k In (5), Xi1k and Xi2k are dummy variables for all possible values of experimentally manipulated attribute k in choice i between candidates 1 and 2, respectively.30 Figure 8 plots the estimates of βk . Bars represent the associated 95% confidence intervals;31 28 In the appendix, we supplement Figure 3 with a formal test of whether punishment rates decline with the distance between the candidates. We also show that this result is robust to specifications with combinations of partisanship and policies that more closely approximate real-world elections. 29 This subtitle paraphrases the title of Schedler’s (2002) seminal article. 30 Our estimates are substantively identical when we estimate the effect of each candidate’s attributes separately rather than taking the difference as we do above. This latter approach is more concise and closer in interpretation to the model in section 2. We discuss it in detail in the appendix. 31 These are almost identical when using block bootstrapped standard errors or normal approximations with clustered standard errors. The appendix presents complete numerical results with both sets of standard errors. ● Gender / Race Male (baseline) ● Female ● White (baseline) ● Asian ● Black ● Hispanic ● Occupation Business Executive (baseline) Farmer ● Lawyer ● Legislative Staffer ● Police Officer ● ● Served in the Army Served in the Navy ● Small Business Owner ● ● Party / Policy Teacher ● Same Party (baseline) ● Different Party ● Economic Policy Social Policy ● ● D+ Board of Elections (baseline) ● Neutral D+ Committee Structure ● D+ Legislative Staff ● D+ Legislative Procedure ● D+ Program Evaluation ● D+ Record−keeping Practices ● D+ Legislative Schedule ● D− Gerrymander by 2 ● Undemocratic D− Gerrymander by 10 ● D− Close Polling Stations ● D− Executive Order ● D− Ignore Courts ● D− Prosecute Journalists ● D− Ban Left Protests ● Valence D− Ban Right Protests ● V− Extramarital Affairs V− Underpaid Taxes −0.3 ● −0.2 −0.1 0.0 Coefficient Estimate Figure 8: Effects of candidate attributes and democracy positions on a candidate’s vote share. Dots represent coefficient estimates based on a linear regression model; bars represent 95% confidence intervals estimates without confidence intervals correspond to baseline categories. We interpret these coefficients as attribute k’s average effect on a candidate’s vote share, relative to the baseline attribute and averaging over all other attribute levels (Hainmueller et al., 2015). Consider first the coefficients associated with the democratically neutral, “generic” positions. All seven are individually (and jointly) statistically indistinguishable from 0, implying that they do not affect a candidate’s vote share. This validates our design and interpretation of these attributes as not only democratically neutral but also more generically inconsequential. Figure 8 also demonstrates a considerable variation in the effect of the individual undemocratic positions on a candidate’s vote share. While all undemocratic positions impact a candidate’s vote share negatively, the magnitude of that effect ranges from 9.5%to 16.0%. Respondents punish most severely candidates who want to prosecute journalists (14.3%) and ignore court rulings (16.0%). Respondents are least sensitive to candidates who endorse gerrymandering (by 2 seats) and suggest that the governor ban protests or rule by executive order. Figure 9 differentiates these estimates by respondents’ partisanship.32 Consistent with our earlier findings, we see few differences between supporters of the two major parties. To put the magnitude of these estimates in context, compare the effect of these undemocratic positions to that of other positional and valence candidate attributes. Consistent with our discussion in sections 3.1 and 3.2, the two main positional attributes –a candidate’s party and policy platforms – have an impact on a respondent’s candidate choice that is either comparable or greater in magnitude than individual undemocratic positions. Of the attributes assigned in the core 11 of the total 16 candidate choices that our 32 We computed marginal means to avoid allowing baseline differences to distort comparisons between subgroups (Leeper et al., 2018). ● D− Gerrymander by 2 ● D− Gerrymander by 10 ● Undemocratic D− Close Polling Stations ● D− Executive Order ● ● D− Ignore Courts Overall Democrat Republican ● D− Prosecute Journalists ● Valence D− Ban Protests ● V− Extramarital Affairs V− Underpaid Taxes ● 0.3 0.4 Coefficient Estimate 0.5 Figure 9: Effects of neutral, valence, and undemocratic positions on a candidate’s vote share by respondents’ partisanship. Dots represent the mean vote share among candidates adopting each undemocratic position; bars represent 95% confidence intervals. respondents made, the most naturally interpretable as valence characteristics are candidate age, years of experience, and profession. From among the nine professions, only military service and being a teacher are statistically significant but their effects are an order of magnitude smaller than that of undemocratic positions. Due to space constraints, Figure 8 omits candidates’ age and years of experience. With a few exceptions, effects of these attributes are also close to zero and not statistically significant.33 To help us further interpret the magnitude of undemocratic positions, we included in 33 We present a complete set of results in the appendix. two of the 16 choices that our respondents made two negative valence attributes intentionally unrelated to democracy. According to the first, the candidate “was convicted of underpaying federal income taxes;” according to the second, the candidate “was reported to have had multiple extramarital affairs.” Estimates associated with these two attributes appear at the bottom of Figure 8 and are labelled V − . We see that, on average, voters punish candidates for extramarital affairs and underpaying taxes more severely than they punish them for undermining democratic principles. 4 Structural Estimates of Support for Democracy Throughout our analysis, we have systematically found a significant, if modest decline in the vote share of candidates whose positions violate democratic principles. This finding is consistent with our theoretical framework that conceives of such positions as a “negative” valence and suggests that the civic virtue parameter δ in our model is positive. We now shift from the non-parametric approach employed so far to a structural approach that estimates our model’s fundamentals: civic virtue δ and the weights αk that voters place on other candidate attributes. This approach allows us to interpret the civic virtue parameter δ as the relative weight that voters place on democracy in their voting decisions, to express voters’ value for democracy in terms of other desirable candidate attributes that voters are willing to forgo to punish candidates who violate democratic principles, and to derive other theoretically informed quantities that yield insights about the robustness of support for democracy in the United States. The estimation of these quantities is aided by a close correspondence between our theoretical model, the design of our candidate-choice experiment, and the random utility model of discrete choice. Recall from section 2 that our assumption of standard Gumbel distributed error terms ij in (4) implies that the effect of candidate attributes on voter i’s probability of voting for candidate 1 can be estimated using the logistic regression.34 For instance, our discussion in sections 3.1 and 3.2 of voters’ willingness to trade off democratic principles for their preferred policies and party suggests the estimation equation Pr(i votes for candidate 1) = logit−1 [β0 + β1 M + β2 econ + β3 social + β4 party] , (6) where M , econ, social, and party are differences between candidate 1 and 2’s democracy 34 See the appendix for evidence of goodness of fit. Table 2: Structural estimates of our model of the public as a democratic check Standardized Natural (1) (2) (3) (4) 0.181∗∗∗ 0.144∗∗∗ 0.155∗∗∗ (0.168, 0.195) (0.122, 0.160) (0.144, 0.168) 0.126∗∗∗ (0.108, 0.140) α1 Economic policy 0.271∗∗∗ (0.253, 0.290) 0.215∗∗∗ (0.182, 0.238) 0.312∗∗∗ (0.292, 0.335) 0.254∗∗∗ (0.218, 0.278) α2 Social policy 0.327∗∗∗ (0.313, 0.346) 0.261∗∗∗ (0.222, 0.286) 0.342∗∗∗ (0.322, 0.363) 0.279∗∗ (0.240, 0.304) α3 Party 0.220∗∗∗ (0.202, 0.239) 0.176∗∗∗ (0.147, 0.195) 0.190∗∗∗ (0.173, 0.206) 0.155∗∗∗ (0.131, 0.171) 17,341 -10,331 17,341 -10,315 17,341 -10,685 17,341 -10,666 δ Democracy N Log-likelihood ∗ p<0.05; ∗∗ p<0.01; ∗∗∗ p<0.001 Note: 95% confidence intervals in parentheses positions, economic and social policy platforms, and partisan affiliation. Estimates from (6) allow us to express the weights δ and αk that voters place on democracy versus other candidate attributes k as δ̂ = |β1 | P |β1 | + K |βk | and αˆk = |βk | P . |β1 | + K |βk | Table 2 presents estimates of the weights δ and αk based on four alternative specifications. Column (1) estimates the logit model in (6) after the four input variables have been divided by two times their standard deviation. This standardization puts candidates’ democracy positions, economic and social policy platforms, and partisan affiliation on a common, dispersion-based scale, allowing us to compare their importance (Gelman, 2008). We see that the weight that voters place on democracy in their voting decisions is about 18% compared to 33% for social policy, 27% for economic policy, and 22% for partisanship. Column (2) in Table 2 extends the model in (6) to include all experimentally manipulated candidate attributes.35 The presented weights for candidates’ democracy positions, economic and social policy platforms, and party imply that these four factors jointly account for 79% of the systematic variation in voters’ candidate choices. These results are consistent with our earlier, non-parametric analysis: Americans value democracy – δ̂ is significantly greater than zero – but they value it less than major competing political considerations, especially economic and social policies and partisanship. Estimates of the weights δ and αk allow us to summarize the electorate’s support for democracy in terms of voters’ value for democracy. This concept expresses support for democracy in terms of other desirable candidate attributes that voters are willing to forgo in order to punish candidates who violate democratic principles. It is directly related to the marginal rate of substitution between democracy and policy that we approximated in section 3.1 by contrasting the control and treatment scenarios in Figure 2 and measuring at various points along the horizontal axis how much more attractive would candidate 1’s policies have to become to compensate for his switching from D+ to D− . Voters’ value for democracy summarizes the marginal rate of substitution between democracy and another candidate attribute parsimoniously – in terms of a single quantity. Specifically, the model in (6) implies that a voter’s payoff remains unchanged as long as a shift in a candidate’s democracy position M is accompanied by a shift of magnitude δ αk in some other attribute k. Put differently, the “price” voters are willing to pay to elect a more democratic candidate is δ αk of attribute k. Estimates of value for democracy are most easily interpreted when we re-estimate the model in (6) with candidate attributes entering in their most “natural” units. We do so by letting M and party enter as candidate-specific dummies and expressing economic and 35 We present a complete set of results in the appendix. social policies on a 0-1 proximity scale. Columns (3) and (4) display the resulting estimates of the weights δ and αk , parallelling the simple and extended models in (1) and (2). Estimates in (3) yield voters’ value for democracy of 0.497 (CI: 0.442, 0.556) in economic policy, 0.454 (CI: 0.407, 0.507) in social policy, and 0.819 (CI: 0.718, 0.921) in co-partisanship. These quantities imply that in order to punish a candidate who adopts an undemocratic position, the voter is willing to tolerate a .5 and .45 increase in the distance from her ideal economic and social policies, respectively – about half the scale. A value for democracy in terms of co-partisanship is smaller than one, implying that the typical voter is not willing to vote across party lines to punish a co-partisan for adopting an undemocratic position. This finding about the value for democracy in terms of co-partisanship is consistent with our earlier, non-parametric analysis of partisanship in section 3.1. There, we found that our respondents punish platforms that violate democratic principles less severely when adopted by co-partisan candidates. The present approach allows us to derive and estimate the degree of this tendency in the electorate in terms of a single, theoretically-informed quantity: the partisan bias π. Specifically, we extend the simple framework in section 2 by multiplying candidate j’s democracy platform Mj in (1) by the term δ[1 − πIj (party)] instead of δ alone, where Ij (party) is an indicator of co-partisanship between the respondent and candidate j. In turn, we may say that support for democracy is principled when π = 0 and voters punish candidates for violating democratic principles equally, regardless of their party; and support for democracy is instrumental when π = 1 and voters only punish undemocratic positions when adopted by candidates from the other party. This formulation implies that the degree of partisan bias π in the electorate can be estimated by adding to the logit model in (6) an interaction term between party and M . Denoting the coefficient on that interaction term by βint , we have π = 1 − βint . β1 The Figure 10: Isoelects depicting the combinations of economic and social policies that result in an equal probability of victory for both candidates. Candidate 1 is the respondent’s co-partisan, candidate 2 is not; gray bands reflect statistical uncertainty estimate of π is 0.496 (CI: 0.370, 0.613), implying that Americans are neither fully principled nor purely instrumental in their support for democracy. Rather, the average voter is about 50% more lenient toward violations of democratic principles by candidates from her own party. The relationship between the structural parameters estimated in this section is graphically summarized by Figure 10. It plots the combinations of economic and social policies that result in an equal probability of victory for both candidates, assuming candidate 1 is the respondent’s co-partisan but candidate 2 is not. We refer to these lines as isoelects. The solid black isoelect plots combinations of economic and social policies that result in an equal probability of victory for either candidate when both adopt a neutral democracy position (D+ vs. D+ ); the dashed blue and dotted red isoelects correspond to the D− vs. D+ and D+ vs. D− conditions, respectively. Combinations of economic and social policies to the right and above the isoelects correspond to scenarios when candidate 1 is more likely to win; policy combinations to the left and below the isoelects correspond to scenarios when candidate 2 is more likely to win. The slope of all isoelects is negative (− αα12 ) with an absolute value smaller than one (| αα12 | < 1). This implies that voters value a candidate’s greater proximity on both economic and social policies but place more weight on the latter.36 Meanwhile, voters’ value for democracy in terms of economic and social policies corresponds the distance between the D+ vs. D+ and the D+ vs. D− isoelect along the horizontal and vertical axis, respectively ( αδ1 and δ ). α2 The isoelects in Figure 10 summarize the impact of partisanship in two ways. First, note that the D+ vs. D+ isoelect is below the point (0, 0), reflecting the advantage ( αα32 ) conferred on candidate 1 by his co-partisanship with the respondent. Second, the 50% bias in the punishment of undemocratic positions that co-partisans benefit from is mirrored in the smaller distance between the D+ vs. D+ and the D− vs. D+ isoelects (compared to the distance between the D+ vs. D+ and the D+ vs. D− isoelects.)37 36 A value of 0 on the horizontal and vertical axes refers to scenarios when the respondent rates the two candidates’ economic or social platforms equally; positive values correspond to a higher rating of candidate 1’s platform. 37 This distance is αδ1 (1 − π) and αδ2 (1 − π) along the horizontal and vertical axis, respectively.