CHAPTER 11: Political Polarization: Parties, Candidates, and the Mass Public – Perceived Polarization

I assess these questions in this chapter using perceptual data from the ANES on the positions of the parties and candidates relative to the reported positions of the mass public. How does the American voter view the parties? Are elites increasingly polarized and extreme relative to the public’s own relatively moderate positions? If so, has this occurred independent of polarization at the mass public level or in spite of attitude stability or even depolarization on the issues in public opinion? How the respondents themselves perceive the parties and candidates can provide a window into the mass-elite relationship on polarization.

An important part of the political polarization story is the relationship between the mass public and elites on the issue dimensions. Fiorina argues that elites have polarized on social issues while the mass public attitudes on social issues have remained stable. Thus elites have moved away from the voters and citizens they are supposed to be responsive to. Thus an empirical question is born: have elites polarized relative to the mass public? In the early part of the Twentieth century, tips on horse racing were known to circulate among the ‘authorities’ such as owners, jockeys, and punters. The most trustworthy sources on the likely winner were those “closest” to the horse (stable boys, trainers, etc.). One step ‘closer’ to the horse than the inner circle of horse aficionados was, of course, the horse itself. Thus was born the colloquialism: if you want the truth, go straight to the horse’s mouth. We can do exactly that, in examining political polarization in the ideological and issue dimensions. Respondents in the ANES time series were asked not only to place themselves on issue and ideological scales, but also to place the parties and candidates on these dimensions as well. We can thus assess directly the perception of the citizenry on the distance between themselves and the parties and candidates respectively on social, defense, and economic issue dimensions. If elites have moved away from their constituents, then citizens should increasingly locate both parties and the candidates of both parties further distant from their own positions. Of course elites can move away from a part of the electorate while staying proximate or even moving closer to other parts of the electorate. Indeed, if we accept the polarization thesis then we should expect Republicans to view the Democratic Party and its candidates further from them, and likewise for Democratic identifiers and elite Republicans. Thus the increasing dispersion between the parties and candidates would be intelligible as responsiveness to the public rather than neglect of it.

Data

The data for this analysis is culled from the American National Election Study (ANES) cumulative file.⁷⁴ I use the ANES studies from 1970-2004.⁷⁵ Distance measures are calculated on the seven-point issue scales from the ANES where respondents were asked to place themselves and the parties and/or the candidates.

The substantive variables included in the analysis and the years for which data was collected on those variables are listed in Appendix G. The ANES time-series includes a number of issue-oriented ordinals scales with which respondents can place themselves, the parties, and candidates in an issue space. While a number of interesting issues have come and gone in the ANES time-series, there are a number of issues that span the breadth of the time series that the ANES has collected data on consistently. Every issue variable rated on an ordinal scale in this analysis has at minimum a respondent self-placement on the issue. However, in order to calculate distance measures, the study must also include the placement of the candidates or the parties in the issue dimension.⁷⁶ The candidate placements always include the placement of the presidential candidates in a presidential election year. In some of the presidential and off-year elections, the House and Senate candidates for the respective parties are included. In a maximum of sixteen (liberal/conservative scale) of the study years the party placements are also collected, though the party placements were not collected in all study years in which the respondent self-placements on the issue were included.

The issues include respondent attitudes on government aid to Blacks, women’s role in society, government spending, defense spending, and jobs. The issue placement that is collected in the most study years for the respondent, candidates, and parties is the ideological placement variable (asking respondents to place themselves and the parties and candidates on a 7-point ideological scale ranging from strongly conservative to strongly liberal). Also included are the candidate and party placements on party. While the abortion self-placement variables are asked for all but one of the study years, the candidates and parties were placed by the respondent only in the study years 1980, 1992, 1996, 1998, 2000 and 2004. The ANES Cumulative Data File does not include the respondent placements of parties and candidates on the abortion scale. Given the importance of the abortion issue to understanding the culture wars, I merged the abortion party and candidate placement variables from the individual studies into the cumulative file so I could include them in this analysis. Given that there is only one year in which they appear prior to 1992, there is insufficient data to assess statistically significant trends. This does not mean the abortion data is meaningless. It is an important indicator of perceived distance on the social issue dimension. As such I examine differences in the abortion distances outside of the framework of a linear trend model.

Expectations

Below are precise statements of expectations given either polarization or depolarization trends. In the models that assess the perceptions of all respondents, we can assess whether the public perceives a particular party and/or its candidates has moved closer to their position or further away over time. If political polarization has occurred at the elite level independent of the masses, we would expect to see that the distances between the mass positions on the issues and the elite positions have increased. Some scholars term this ‘alienation’ in the literature: elites have become more alienated from the public on the issues.

However, as I noted above, merely identifying alienation between the average positions of the mass public and the perceived positions of the elites is insufficient to conclude that the elites have polarized while the public has moderated. Elites may be responding to an electorally significant portion of the public which has polarized as well (or they may have polarized together). I consider the most likely model for such a phenomenon: partisan polarization. If the electorate has experienced partisan polarization, then we would find alienation at the aggregate mass public level that is fundamentally misleading on the polarization question. If mass Republican identifiers have polarized from Democratic identifiers, and the Republican elites have moved in response to (or coincident to) this mass polarization of their base, then we would expect two specific conditions: 1) Republicans at the mass level would not perceive an increasing distance between themselves and elite Republicans. 2) Republicans at the mass level would perceive a substantial increase in the distance between themselves and the Democratic elites. And the same would go for Democratic identifiers and Republican elites. The precise expectations of partisan polarization are laid out in the Base and Opposition models.

Perceived Partisan Proximity on Issues Expectations (Distance)

Elites (Candidates) / Mass Divergence (Alienation)

E_o: No trend in the mean absolute distance from both Republican and Democratic Candidates.

E_a: A decrease in the mean absolute distance from both Republican and Democratic Candidates (non-alienation).

E_a: An increase in the mean absolute distance from both Republican and Democratic Candidates (alienation).

E_o: No trend in the mean absolute distance between Republican identifiers and the Republican Party or in the distance between Democratic identifiers and the Democratic Party (constituent status quo).

E_a: An increase in the mean absolute distance between Republican identifiers and the Republican Party or an increase in the distance between Democratic identifiers and the Democratic Party (constituent alienation).

E_a: A decrease in the mean absolute distance between Republican identifiers and the Republican Party and/or an increase in the distance between Democratic identifiers and the Democratic Party (constituent responsiveness).

E_o: No trend in the mean absolute distance between Republican identifiers and the Democratic Party or in the distance between Democratic identifiers and the Republican Party (partisan polarization status quo).

E_a: A decrease in the mean absolute distance between Republican identifiers and the Democratic Party or a decrease in the distance between Democratic identifiers and the Republican Party (partisan depolarization).

E_a: An increase in the mean absolute distance between Republican identifiers and the Democratic Party or an increase in the distance between Democratic identifiers and the Republican Party (partisan polarization).

One test of the elite / mass divergence hypothesis is to examine the perceived issue distance of the candidates to the mass public. Using the ANES, we can place the candidates and respondents in the issue space. We can then calculate the distance between the candidates and the respondents using Euclidian geometry. Thus the issue distance measures are the relative, perceived difference between a respondent and each of the candidates, parties, or elected officials on an issue dimensions (e.g. aid to Blacks). . Let S_ij denote the self-placement of voter j on an issue dimension i, and let the Democratic and Republican Party or those parties candidates’ perceived locations on issue i by respondent j be D_ij and R_ij.

Equations 11.1 & 11.2: Perceived Distance from Respondent to Parties or Candidates

We can assess the proximity of the responded to one candidate or party relative to the other by calculating the absolute value of the distances between the candidates or parties and taking the difference of these absolute distances. Again, let S_ij denote the self-placement of voter j on an issue dimension i, and let the Democratic and Republican Party or those parties candidates’ perceived locations on issue i by respondent j be D_ij and R_ij. Then, distance on issue i is the absolute difference between self-placement and the perceived candidate [party, official] location. The relative difference on issue i, for respondent j, is:

Equation 11.3: Perceived Partisan Proximity Measure

where a positive value indicates closer issue proximity to the Democratic presidential candidate or Democratic Party, a negative value indicates the respondent is closer to the Republican candidate or party, and a value of zero indicates indifference between the two. A larger magnitude for issue distance indicates greater perceived difference between the presidential candidates or political parties (Bough et al. 2004).

Continuing with the logic set out above, we can also measure the total perceived distance from the respondent position to each of the candidates, or alternatively, both parties. Rather than assess which candidate the respondent perceives as closer to him or her on a particular political issue, this measure assesses the total distance from the respondent’s declared position on the issue and the position they believe the candidate has on that issue. The total distance on issue i, for respondent j, is:

Equation 11.4: Total Perceived Distance Measure

For those respondents who gave non-responses, if the respondent failed to replace him or herself on an issue, the distance value was set to zero. If a respondent failed to place one of the candidates, the perceived candidate position for that respondent was set to the mean candidate position for that issue. If a respondent refused to answer, they were set to missing.

As I have argued in previous chapters, the ideological dimension is a convenient proxy for the sum of the issue dimensions at play in the public policy debate. Tables 11.1 reports distance measures for ideology (a proxy for the aggregate issue space). Abortion distance measures are assessed in Table 11.2 and the distance measure trend models for public opinion on government’s role in providing jobs can be found in Table 11.3. Distance measures for defense spending and government spending are located in Appendix I.⁷⁷ I do not include those models in the main text because there are no significant linear trends for defense spending and the only significant model for government spending is a positive trend in the perceived distance between respondents and the Democratic presidential candidates and the Republican House candidates. This is not to say that there is nothing of significance to be found in these models. The fact that there are no significant increases in the total perceived distance between the candidates or parties on defense spending suggests the public does not believe the elites are out of touch with them on foreign policy and national defense issues. On government spending it is essentially the same story, except that there was one significant trend in total distance: a significant decline in the total distance between the public and the Republican and Democratic House candidates. The issue distance tables contain four different classes of measures. The first section includes respondent self-

** significant at .05 level

***significant at .01 level

placements and the placements of parties and candidates on the specific issue scale. The second section reports the distances between the respondent and the parties or candidates. So a distance measure here would be the perceived distance between the respondent and the Republican Party. The third section reports the relative absolute differences between the perceived distance between the respondent and the Republican Party and the respondent and the Democrat Party (proximity scores). The fourth section reports total distance measures on the issue for the parties and candidates. So a total distance measure adds the perceived distance between the Republican Party or candidate and the Democratic Party or candidate for a measure of the sum of the absolute distances from the respondents and both political parties and candidates.

On ideology there is a significant trend in the average respondent self-placement in a conservative direction. The American public has become more conservative over the past three decades by, on average, .003 points every survey year (R² = .164). It is not a strong trend, nor is it a substantively large change in the average ideology of the American public, but it is statistically significant. Furthermore, the American public sees a conservative trend in the Republican Party and its presidential candidates. The perceived conservative tilt in the Republican Party is particularly strong (0.010) relative to the coefficients of the other models, with the linear trend model explaining 45% of the variance in the Republican Party’s ideological position (R² = .456). The trend for Republican presidential candidates is similar but not as strong (0.007), explaining just under 40% of the variation (R² = .393).

On the relative distances, the models show significant positive trends in the distance between the respondents and the Democratic Party (0.011) and the Democratic House Candidates, suggesting that Democrats are increasingly to the ideological Left of the average respondent. The significant negative coefficients for the Republican Party and Republican presidential candidates show that Republican elites have increasingly moved to the Right relative to the average respondent. The Republican relative distance models illustrate strong predictive power, with the linear trends accounting for over 60% of the variance over the mean distance for the Republican Party (R² = .628) and the Republican presidential candidates (R² = .668).

In the absolute distance measures, the relative partisan proximity model for House candidates demonstrates that, from the viewpoint of the ‘average’ member of the public, the Republicans are further distant from the Democrats. Interestingly, this is not the case for the Republican Party or its presidential candidates. While both the Republican and Democratic parties are perceived to have polarized relative to the average citizen, there isn’t strong evidence that either has polarized more than the other. There is strong evidence, however, that the Democratic and Republican elites have increasingly polarized on the ideological dimension. The total relative partisan distance measures, combining the absolute perceived distance for both Democrats and Republicans, show strong positive linear trends. The average citizen perceives both parties (0.025), both parties’ presidential candidates (0.022), and both parties’ House candidates (0.023) as increasingly distant from their own position. The presidential candidate model explains over 85% of the variance in the distances, and exhibits a strong, positive linear coefficient (R² = .858). The other two models explain over 50% of the variance in distances. The total distance measures indicate that the average respondent finds the parties further distant from him or herself today than in the past.

At first blush, the total distance measures would appear to be evidence in favor of Fiorina’s conjecture that elites have polarized away from the masses. However, jumping to that conclusion is unwarranted. Recall that political polarization implies greater dispersion. The total distance measures are relative to the average (mean) respondent. Hence these measures indicate that the parties have polarized relative to the average member of the mass public. The culture wars thesis (political polarization) predicts that the Republican and Democratic Parties will move towards the poles of the distribution relative to the center point. So the question isn’t whether the parties have moved away from the center on any particular issues. The question is whether the Parties have diverged independent of the voters.

Abortion Distance Measures

As noted earlier, the abortion distance measures are assessed in Table 11.2. However, the statistical test for the abortion distances is different than those used for the other issue dimensions. The abortion issue placements for the parties and candidates were only collected in five survey years.