Religion and environmental politics in the US House of Representatives

ABSTRACT

Does religion affect legislators’ behavior on environmental policy in the US? Studies of environmental policy making have not examined this question, although the literature suggests that religion might affect legislative behavior on environmental policy. This study examines the relationship between US House members’ religion and roll-call voting on environmental legislation from 1973 to 2009. It finds significant differences across religious traditions. Legislators’ party and characteristics of constituencies relevant to environmental politics increasingly, but not entirely, mediate these differences.

KEYWORDS:

• Environment and religion
• religion and congress
• religion and roll-call voting
• league of conservation voters score
• religion and league of conservation voters

Acknowledgments

Originally presented at the 2012 Meeting of the Midwest Political Science Association, April 15–18, Chicago, IL. We thank Lee Kats, Rick Marrs, and Katy Carr for their support; Taylor Clayton, Paul Henderson, and Marc Vinyard for their research assistance; Sarah Anderson for sharing her data; and Amy Black, along with anonymous reviewers, for helpful suggestions.

Notes

1. As just a few examples, see the following: Southern Baptist Convention: http://www.baptistcreationcare.org/node/1; Anglican Church; http://www.anglicancommunion.org/ministry/mission/fivemarks.cfm; Catholic Church before Francis: http://www.vatican.va/holy_father/benedict_xvi/encyclicals/documents/hf_ben-xvi_enc_20090629_caritas-in-veritate_en.html; Judaism: http://www.coejl.org/resources/global-warming-a-jewish-response-2000/; Buddhism (Gross 1997).

2. In the PartiRep survey of European MPs, 45% of members reported being in contact with churches or religious organizations at least once every three months (see Steven 2014).

3. According to the LCV (see http://scorecard.lcv.org/methodology, accessed 11 June 2014), 20 experts from environmental organizations identify roll-call votes from 12 different environmental issue areas. Typically, votes count equally, but the LCV occasionally deems a vote important enough to count double.

4. For example, imagine two hypothetical years, one with a Democratic majority (with power to set the roll-call agenda) and the other with a Republican majority. Imagine that in year one, the House voted on legislation to: (1) cap carbon emissions on existing and new plants, (2) raise fuel efficiency standards on new cars, (3) eliminate the use of hydraulic fracking, and (4) increase regulations for construction near wetlands. In year two, the House voted on legislation to: (1) open the Arctic National Wildlife Refuge for oil drilling, (2) open protected land for increased logging, (3) require the construction of the Keystone XL pipeline, and (4) provide subsidies for companies using hydraulic fracking. A score of 50 in year one (supporting two strong pro-environment policies) differs from a score of 50 in year two (supporting two environmentally disruptive policies). Of course, these hypothetical years include agendas more extreme than normal, but they demonstrate the need for a shift of the second score downward (or the first upward) for scores to be comparable.

5. The method identifies a baseline year and transforms scores from other years to match this baseline year. Like converting temperature from Celsius to Fahrenheit, this transformation involves shifting the scores up or down, then multiplying them to ‘stretch’ them so that the ends of the scale hold the same meaning.

6. Coding these generic Christian responses as a separate category does not alter our results. We coded all representatives listing ‘Baptist’ as Evangelical Protestant and all representatives listed as ‘Lutheran’ or ‘Presbyterian’ as Mainline Protestant.

7. These cases include 19 African American members, the vast majority of whom were Catholic, and 12 Hispanic representatives, virtually all affiliated with a Protestant denomination.

8. The average is highest in 1990, when Congress passed a major extension of the Clean Air Act.

9. Robust standard errors adjust for heteroskedasticity, which can generate incorrect standard errors and therefore lead to misleading hypothesis tests. Breusch-Pagan tests found heteroskedasticity in about 40% of our models.

10. For example, we interacted the Evangelical indicator variable with southern residence to see whether southern Evangelicals differed from southern non-evangelicals. The difference between Evangelical Republicans and non-Evangelical Republicans in the South (five points) is smaller than the corresponding difference outside the south (13 points), but Evangelicals differed from non-Evangelicals at the 0.05 level in both locations. We present the simpler results for ease of presentation, though additional results are available upon request.

11. We thank Sarah Anderson for sharing the district-level data. The 1990 Census reports district-level variables based on the 1990s round of redistricting, which was not in place until 1992. Therefore, we cannot use these data to estimate a model for 1990.

12. Anderson aggregated the groups’ members and used this aggregation in models for the 105th Congress (1997–1998). Since Congressional district boundaries remained basically stable between 1992 and 2002 (redistricting years), we use these data in our models for 1995, 2000, and 2001. Anderson also analyzed data from the 109th Congress (2005–2006), though she could only collect district-level data from the National Resource Defense Council for these years. We use this measure in models for 2005 and 2009.

13. District-level income and district-level environmental group membership correlate at between 0.50 and 0.61. Percent of the district employed in the agricultural sector correlates with percent urban at 0.50–0.60.

14. From 2000 onward, district-level presidential vote and representation by a Democrat correlate at 0.61–0.70. District-level presidential vote and percent urban correlate at between 0.55 and 0.59. The Variance Inflation Factor for district presidential vote share ranges from 54 to 69, where scores >10 typically indicate significant multicollinearity.

15. Path models require that causality only runs from religious tradition to party affiliation. We think it makes sense that members’ religious traditions would shape their party preference (Guth et al. 2009). It seems unlikely that many people first affiliate with a political party and then as a result of that party choice determine which religious tradition to join. However, we recognize that path models can include considerable error in estimates, so we interpret the results with caution. Our purpose is simply to demonstrate that significant indirect effects are present, not to make precise claims about the exact size of those effects.