Complex constituencies: intense environmentalists and representation

Abstract

The role of environmental subconstituencies in determining the voting behaviour of members of the US Congress is assessed. Using ordinary least squares and symmetrically censored least squares (a method especially appropriate for bimodal data), the impact of environmentally-concerned constituents, as measured by original data on environmental group membership, on congressional voting is quantified. In general, members of the US Congress vote more pro-environmental when they have more environmental group members in their districts, which may be representation or factionalism at work. Surprisingly, even though members of environmental groups are a traditionally Democratic constituency in the United States, both Democrats and Republicans are responsive.

Keywords:

• environmental group
• Congress
• voting
• political parties
• constituency

Notes

 1. The CCES study (Ansolabehere 2006) includes a sample of 30,000 persons conducted through the collaborative efforts of a consortium of universities. The 2006 CCES survey was administered in August 2006 by Polimetrix, an Internet survey firm. More information can be found at http://projects.iq.harvard.edu/cces

 2. See, for example, the ABC News/Washington Post poll from September 2002 finding that 70% of independents trust the Democrats more to handle the environment.

 3. LCV does not have an official, all-encompassing definition of pro-environment, but generally scores votes that preserve open space, reform grazing law, limit mining and timber harvests, limit dependence on nuclear energy and fossil fuels, and strengthen regulations under the Endangered Species Act as pro-environmental.

 4. In addition to the four groups that provided data, I requested membership data from the Environmental Defense Fund, Friends of the Earth, Greenpeace, the National Audubon Society, and the World Wildlife Fund.

 5. The analysis was performed using the Zip Code Tabulation Areas data set from the ESRI Data & Maps 2006 Collection and the 105th and 109th Congressional Districts Cartographic Boundary Files from the US Census Bureau. Using GIS, the zip code layer was superimposed over each congressional district layer, and areas of the union of the zip code and congressional district were calculated and divided by the total area of its corresponding zip code. Due to inherent geometric inconsistencies among the data sets, any ratios below 0.05 were treated as artefacts of imperfect matching and thus removed. The data was then renormalised to ensure that the sums of all subsections for each zip code were 1. To calculate the number of members in a congressional district, the number of members in each zip code that composed part of the congressional district was multiplied by these ratios and then summed.

 6. The ratio of the maximum values of membership to the minimum values are quite high, ranging from 374 for the Sierra Club to 16 for the National Wildlife Federation but results using logged values are substantively the same.

 7. When computing the eigenvectors, the loadings from the factor analysis are NRDC 0.531, NWF 0.479, TNC 0.517, and SC 0.469. A second component accounts for an additional 12% of the variance.

 8. Results using Markov Chain Monte Carlo (MCMC) estimates of a normal theory factor analysis model are almost identical, suggesting that indeterminacy is not a significant problem here. The initial 25,000 MCMC scans were discarded as burn-in. The posterior summaries of the estimates used are based on a posterior sample size of 20,000 from the subsequent scans with the loading for TNC constrained to be positive. Results available from the author.

 9. This R-squared is only reduced to 0.73 when ideology is omitted. There are several districts with extremely high Sierra Club and National Wildlife Federation membership, mostly in California. When these six districts are omitted, the results remain substantively the same.

10. As Jackson and Kingdon (1992) argue, this may simply be a result of NOMINATE scores tapping the same or a related dimension of voting as LCV scores. After all, NOMINATE scores include the same roll call votes as LCV scores (and many others). Employing the most conservative interpretation of the coefficient on ideology, Model 1 shows that environmental voting taps generally the same dimension as all other voting in the House of Representatives. Moreover, Jackson and Kingdon note that the inclusion of a vote-based measure of ideology leads to an attenuation of the importance of the other inputs to voting. Even with this underestimation of the impact of constituent interest, this study finds an impact of constituent interest on voting, indicating that the impact of constituency is likely even greater than these results show.

11. The variance inflation factors (Davis et al. 1986) indicate that there may be a problem with multicolinearity when ideology is included, but results available from the author are substantively the same when it is excluded. A variance inflation factor greater than 10 is usually thought to be cause for concern. Variance inflation factors are as follows: Environmental group membership factor 1.83, DW-NOMINATE 13.04, Party 9.09, Median household income 2.59, Mining employment level 1.44, Agriculture employment level 1.82, Manufacturing employment level 1.51, Urbanisation 2.83, Clinton vote 3.36.

12. This can be estimated in Stata using ado files downloaded from http://elsa.berkeley.edu/∼kenchay/. Each SCLS estimation uses 100 iterations.

13. A model with each independent variable interacted with party reveals similar patterns. None of the interactions with party are statistically distinguishable from zero.

14. See http://www.grist.org/article/post_election/