In Chapter 2 I examine the general effect of heterogeneity in information costs on measures of issue importance derived from spatial patterns of voting. I again find evidence of heterogeneity in the influence of issues on vote choice in the US electorate and the role that information costs play in determining issue salience. 4 Issue salience and information costs 61 4.1 Measuring issue salience in a ranked-rank spatial voting model.
Chapter 1 Introduction
Most studies of the impact of issues on vote choice ignore the possibility of heterogeneity in issue weights. Failure to understand heterogeneity in the impact of issues on vote choice also obscures our substantive understanding of election outcomes and voter behavior. I find evidence of heterogeneity in the impact of issues on vote choice in the American electorate.
- Uncertainty in the Measurement of Candidate Issue Positions
- An Empirical Test of the Effect of Uncertainty on Estimates of Salience
- Discussion
Here, as in most studies of the impact of issues on vote choice, I celebrate the difference between the candidate and the voter on the issue scale when calculating distance. Any estimates of the impact of issues on vote choice will therefore be biased upwards. Including undecided respondents in the calculation of the average positions of candidates biases our estimates of the impact of issues.
Chapter 3 Survey Measures of Issue Salience
- The Reliability of Survey Measures of Issue Salience
- Survey Measures of Issue Salience in a Spatial Model of Voting
- Empirical Application of the Survey Measures of Issue Salience
- Discussion
- Appendix
It is possible that some of the questions used were more effective in determining issue salience than others. 1See the Appendix for the exact question wording of each of the questions used in these studies. If A significantly improves the fit of the vote choice model, the survey measures of issue strikingly improve our understanding of voter behavior.
To test the effect of self-reported salience measures on the vote choice model, I create an issue salience variable in the same way as an issue distance variable, except that this variable includes only the issues identified as most important by the respondent. First, I examine the effect of open-ended survey measures of issue salience on the electoral choice model. However, including an additional variable to add weight to the issue identified as most important does not improve model fit.
In none of the models is the coefficient for the most important issue variable significant. Further, likelihood ratio tests show that adding this variable does not statistically improve the fit of the vote choice models. Once again, an additional variable measuring the additional explanatory power of the issue or issues identified by the respondent as most important adds nothing to the model of vote choice.
The sample size of the 1968 CSEP is extremely large, especially relative to the sample sizes available for the other datasets. Thus, the large sample size of the 1968 CSEP may be the reason the estimation issue is salient. The scale measures of issue salience elicit salience judgments for all the issues represented by the issue placement scales, but exhibit a strong upward bias, with respondents reporting nearly every issue as salient.
Issue Salience and the Costs of Information
- Measuring Issue Salience in a Spatial Voting Model with Rank Ordered Data
- Issue Salience and Heterogeneity
- Issue Salience and the Cost of Information
- Discussion
This is conditional logit, where the choice properties of the equation are the issue distances between the candidate's platform and the individual's ideal point. In terms of the model described above, this would involve estimating a separate /3 for each voter. This is another version of the problem of independence of irrelevant alternatives (IIA), which is generally not a desirable property in models of candidate choice (Alvarez and Nagler 1998).
In the empirical work in the following section, I only include respondents who have ranked at least five candidates (giving at least four observations per respondent). However, we can still learn something about problem heterogeneity by examining the sign of the estimated coefficients for each individual. This implies that the distance to this question had some weight in the construction of the ranking, as respondents tend to prefer candidates closer to them in the question over those who are further away.
Likewise, a non-negative sign on the estimated coefficient indicates that the issue has not had any weight in the construction of the ranking order, since candidates closer to the respondent are not necessarily preferred over those further away (this follows from the theoretical assumption that the emission weights should be non-positive). The effect of information costs on the emission significance estimates presented here is the subject of the next section. Another way to examine the effect of information cost on issue salience is to relate information costs to the total number of issues an individual found salient.
Although heterogeneity in issue weights is not surprising in theory, few studies of the American electorate address this possibility.
Chapter 5 Heterogeneity in Issue Salience in the American Electorate
- Homogeneity, Distributions, and Random Pa- rameters Logit
- An Application to the 1996 Presidential Elec- ti on
To ignore the possibility of heterogeneity in issue weights, restrict the analysis of the impact of issues on vote choice to the top two cells in figure 5.1. Allowing for heterogeneity in issue weights enables a much more complete picture of the impact of issues on vote choice. I show that random parameters logit contains all the information that models that assume homogeneity do, plus I uncover evidence of heterogeneity in the weights voters placed on issues.
Thus, the random logit parameter provides a clear statistical test for heterogeneity in the influence of issues on vote choice. A conditional logit model is also estimated to show that the random logit parameters reveal all the information that the conditional logit does, plus information on the heterogeneity in the emission weights. To test the effectiveness of random logit parameters in analyzing the impact of issues on vote choice, I applied it to the 1996 presidential election.
This shows that the means of the distributions in the logit of the random parameters give us the same information as the coefficients in the conditional logit. Note also that the coefficients of the logit random parameters are always larger. In the random parameter logit, some utility that is not observed in the conditional logit (and thus captured in t:) is captured by the standard deviations.
Thus, a more complete picture emerges of the impact of issues on voter choice in the 1996 presidential election, as shown in Figure 5.3.
5. 3 Discussion
These issues were, on average, important to voters, and there were no major differences in the weight that individuals attributed to these issues. Because welfare reform and the fate of affirmative action were important topics in the media at the time, it seems likely that these issues were important to many voters in the 1996 election. One way to determine whether heterogeneity in the use of ideology is determined by electoral context or information costs, is to compare the 1996 estimates with results from a more "typical" year.
It was a race between two candidates, and the ideology of the candidates was much more exposed in the media and more clearly distinguishable between the candidates. However, we do not expect dramatic differences in the relative costs of information for different subgroups in the electorate between 1988 and 1996. Thus, comparing the results between 1988 and 1996 might shed some light on the source of heterogeneity in weighted ideology.
The data is from the National Election Study from 1988, coded in the same way as the data from 1996. However, the standard deviation of the distribution of ideological weights was not statistically significant. To show the difference in the influence of ideology in 1988 compared to 1996, I calculated the first differences between the two years for the average voter in the sample.
However, in Figure 5.5 there is little difference in the influence of Dukakis' ideological position on the vote choice of the average voter between the low weight case, the "medium" weight case, and the "high" weight case.
Heterogeneity and Salience
Heterogeneity and Salience, Conditional Logit
Salient Overall?
Heterogeneity and Salience, Random Parameters Logit
Chapter 6 Conclusions
Second, the heterogeneity in issue salience we observe is related to the information costs each voter faces. The heterogeneity in the impact of issues on vote choice revealed here should come as no surprise. Estimating standard deviations for issue weights using random parameter logit gives us a clear statistical test for heterogeneity in the impact of issues on vote choice.
If heterogeneity in issue is significantly correlated with voter issue positions, then any model that does not take into account these biased coefficient estimates, and therefore will have a misleading picture of the impact of issues on vote choice. Furthermore, I am able to relate heterogeneity in issue weights to the information made available in political campaigns. Heterogeneity in issue salience increases when voters are presented with confusing or contradictory information.
This is presented in Chapter 5, which examined the different levels of heterogeneity in the emphasis placed on ideology in 1988 and 1996. Much work remains before the heterogeneity in the impact of issues on vote choice is fully understood. Although problem heterogeneity clearly exists, I cannot precisely determine the source of the heterogeneity without the kind of dataset used in Chapter 4.
The serious nature of this implication underlines the importance of understanding the heterogeneity in the impact of issues on vote choice and its source.
Bibliography
34; An Expository Development of a Mathematical Model of the Electoral Process." American Political Science Review, 64:426-49. 34; On the Efficiency of Methods of Simulated Moments and Maximum Simulated Likelihood Estimation of Discrete Response Models." Econometric Theory, 8:518-52. 34;Mixed Logit with Repeated Choices: Households' Choices of Appliance Efficiency Level." Review of Economics and Statistics, forthcoming.
34;Validity Tests of Complete Unit Analysis in Surveys Subject to Item Nonresponse or Attrition.” Unpublished manuscript.
