ABSTRACT
Partisan polarization of public opinion is a major trend in American environmental politics. While the national pattern is widely recognized, scholars know much less about the polarization of public opinion over time at the state level. This lack of knowledge is unfortunate because geographic variation in the polarization of opinion is essential for explaining the origins of partisan polarization and evaluating its consequences for policy. To fill the gap, the multilevel regression and poststratification technique is applied to provide credible estimates of state-level environmental public opinion for both Democrats and Republicans, 1973–2012. It appears that the growing partisan gap reflects increased pro-environmental opinion among Democrats across many states, whereas Republican state-level public opinion is converging toward a much lower baseline. Cross-state variation among both parties has decreased over time, contributing to greater partisan polarization in the aggregate. Changes in the sorting of voters in and out of political parties cannot explain these patterns of polarization.
Supplemental data
Supporting information and replication data for this article can be accessed here.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1. The specific variable we use from the GSS data is natenvir.
2. We focus on these demographic variables because data are available from both the GSS and the US Census, which we need to use for the MRP approach. Also, these factors were found to be important predictors of individual attitudes toward spending on environmental protection over multiple years (McCright et al. Citation2014). The interaction between age and education was included as it is an important predictor for partisan affiliation. State-level presidential support was included to assess state-level ideology as this was found to be a better measure of state-level partisan identification (Kastellec et al. Citation2015). Note that including only these variables required estimating environmental attitudes for 14,400 different demographic-partisan-state types.
3. We again follow Kastellec et al.’s (Citation2015) approach in dealing with multinomial response in implementing MRP. In the survey, respondents were asked to choose ‘too much’, ‘too little’ and ‘about the right amount’. This creates another complication as it is difficult to implement MRP with non-dichotomous variables. We employ nested multinominal MRP: we estimate the percentage of pro-environmentalists by predicting the share of pro-environmentalists versus other in a binary logistic regression; and then drop all observations with pro-environment attitudes so that we can use the remaining data to predict neutral attitudes versus anti-environmental attitudes, conditional on not being pro-environmental attitudes. We repeat the process starting from the other side (starting with anti-environmental attitudes versus other) and average the results from both orderings.
4. As Gallup poll data are not available for every year, we used the estimated frequency of each subgroup population from the year 1980 for all the years before 1980. For other years, we weight the frequencies based on the two nearest decades to the years of survey. For instance, for the survey conducted in 1995, we use demographic–geographic–partisan frequencies equal to 0.5×1990 frequencies + 0.5 × 2000 frequencies. As for the census data, we used the Census Public Use Microdata Area (PUMA) data for 1980, 1990 and 2000 and the Census American Community Survey data for 2009 as the 2010 PUMA sample was not released.
5. The specific GSS variables we use are nataid, natrace, natarms, nateduc, natheal and natfare. All questions were asked with the same wording as described above. Each variable measures individual attitudes toward spending on ‘foreign aid’, ‘improving the conditions of Blacks’, ‘the military, armaments and defense’, ‘improving the nation’s education system’, ‘improving and protecting the nation’s health’.
6. For state-by-state estimates, see Figure A7 in supporting information. The Pearson rank correlation of state-by-state estimates for 2003–2012 with those for belief in global warming in Howe et al. (Citation2015, Figure 1, supporting information), is 0.79.