Public opinion on climate change: Is there an economy–environment tradeoff?

ABSTRACT

Does the state of the economy condition public concern for the environment? Scholars have long argued that environmental preferences decline during economic downturns as individuals prioritize short-term economic needs over longer-term environmental concerns. Yet, this assumption has rarely been subjected to rigorous empirical scrutiny at the individual level. The presumed link between economic and environmental preferences is revisited, using the first individual-level opinion panel (n = 1043) of US climate attitudes, incorporating both self-reported and objective economic data. In contrast with prior studies that emphasize the role of economic downturns in driving environmental preference shifts, using a stronger identification strategy, there is little evidence that changes in either individual economic fortunes or local economic conditions are associated with decreased belief that climate change is happening or reduced prioritization of climate policy action. Instead, the evidence suggests that climate belief declines are associated with shifting political cues. These findings have important implications for understanding the dynamics of political conflict over environmental policy globally.

KEYWORDS:

• Climate change
• public opinion
• economy–environment tradeoff

Acknowledgments

Special thanks to Gabe Lenz for sharing QCEW county-level employment data. Thanks also to Paul Quirk, Danny Hidalgo, Alexander Hertel-Fernandez, Lyle Scruggs, Salil Benegal, Michael Aklin, Maxwell Boykoff, Peter Howe, and workshop participants at George Mason University, Universitetet i Bergen, Columbia University, and the University of Connecticut for their comments on earlier drafts of this analysis. Funding support for this research came from the Surdna Foundation, the 11th Hour Project, and the Grantham Foundation for the Protection of the Environment.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental data for this article can be accessed here.

Notes

1. Additionally, Krosnick and MacInnis (2012) do not find a difference in climate opinion trends between 2010 and 2012 in states that were strongly or weakly affected by the recent recession.

2. Supplemental appendices are available on the Harvard Dataverse at doi:10.7910/DVN/CCYZ3M

3. In separate analysis, data from the recontact survey have also been used to study over-time patterns of media consumption and certainty about climate beliefs (Feldman et al., 2014).

4. The questions analyzed here come from the first part of a 2-part survey conducted in October and November 2008. Because of that larger survey’s length, the survey instrument was split in two. Each half was completed by the same group of respondents two weeks apart. All questions used in this current study were asked in the first half of the survey, completed between 7 October and 20 October 2008. Of 3997 individuals invited to participate, 2497 completed the first half of the survey (^∼62% response rate). Respondents were then invited to complete the second survey half between 24 October and 12 November 2008. While no questions from this second half are included in the analysis, only subjects who answered both survey waves were recontacted for the 2011 follow-up survey. Ultimately, 2164 respondents completed both halves of the initial survey (^∼87%), generating some mild initial selection in our sample. We discuss the more serious attrition in our sample between 2008 and 2011 below, and confirm our results are robust to this attrition in Appendix 2.

5. We use the terms global warming and climate change interchangeably here, although our panel survey uses the phrase ‘global warming’ consistently across both survey waves. While researchers have shown that survey responses to questions about climate change varies over these terms, our analysis seeks to explain changes over time using a common question, not baseline agreement. Further, there is no reason to believe that the observed decline in belief in climate change during the period of the study was conditional on question wording. For a detailed comparison of the affective and cognitive effects of using one term over the other, see (Leiserowitz et al. 2014).

6. A larger bank of policy prioritization questions also included similar questions asking about prioritization of education, health care, social security, the federal budget deficit, terrorism, tax cuts, illegal immigration, the wars in Iraq and Afghanistan, and abortion.

7. For wave 1 respondents, locations are derived from the ZIP+4 codes. For wave 2, respondents’ latitude and longitude coordinates were provided by Knowledge Networks, randomly jittered within a 0.15 km radius pre-analysis to protect confidentiality.

8. An alternative measure of strong recession impact was also coded that discretizes only respondents who answered ‘A lot’.

9. PRISM provides spatially disaggregated monthly climate data at the 2.5 arc-minute scale, which corresponds to about 4 km between grid units (Daly et al., 2002, PRISM Climate Group 2004). PRISM data exclude Hawaii and Alaska; correspondingly, respondents from both states drop out of panel analyses below that include weather covariates. PRISM data capture very local variations in weather conditions, including the presence of rain shadows, temperature inversions and coastal effects. For each respondent’s geolocation, the difference between mean temperature in the month of survey response and the average temperature for that month between 1971 and 2000 was calculated. This difference is a high-resolution estimate of how anomalous recent experienced weather was for each respondent at the time of survey response. Similarly, the difference between mean precipitation in the month of survey response at the respondent’s geolocation and annual precipitation patterns for that month between 1971 and 2000 was calculated, converted into an average of historic precipitation levels for that month.

10. Using a list published by CNN in early July 2011 as a record of tea party membership as of the wave 2 survey date. See http://politicalticker.blogs.cnn.com/2011/07/29/who-is-the-tea-party-caucus-in-the-house/ (Accessed 15 July 2013).

11. LCV scores range from 0 to 100% and are not only a function of the number of pro-environmental votes (the numerator), but are the number of votes included in each year’s scorecard (the denominator). Consequently, there is natural variability in many Congressional representatives’ LCV score. Further volatility is introduced because the types of environmental legislation available to be scored vary across Congresses. To make Congressional LCV scores more comparable over time, all LCV scores were converted into a 5-point scale where 0.2 represented a score in the lowest quintile, while 1 represented a score in the top quintile.

12. This dataset provides monthly counts for articles written about climate change in major US newspapers.

13. One common approach to increasing power is to use a random effects specification, which also exploits between-subjects variation. Random effects specifications are based upon the premise that individual-specific effects are uncorrelated with the model’s independent variables, a strong assumption in this context. Comparing fixed and random effects models (not shown) for belief in climate change, a Hausman specification test, χ² distributed with df = 9 has a test-statistic of 53.74 (p-value <0.000). A Hausman specification test for the priority model is also significant, with a test-statistic of 18.94 (p-value = 0.026). Assuming that the fixed effects estimator is consistent, this suggests that the random effects specifications are inconsistent and thus inappropriate for this setting.

14. This finding is incidental to Margalit (2013)’s work, which uses a climate-related question as a placebo.

15. The use of survey weights can also entail an efficiency loss, but only conditional on the model being correctly specified.

16. Our dependent variables are ordered factors; ordered likelihood models are known to be inconsistent in panel settings and cannot be used here.

17. QCEW data have minimal measurement error and are derived from employer-level Unemployment Insurance (UI) filings.

18. The null effect is also robust to changes in average 6-month employment growth rate (not reported), rather than shifts in employment levels.

19. One potential explanation we can exclude is that net shifts in house price value are smaller in areas that saw higher overall price volatility; then, smaller net shifts in house price values could still proxy stronger recession impacts. Instead, volatility in the zip-level home price index is increasing in net value shifts between the two waves.

20. Generally, data availability skews toward more urban areas of the country, with consequent imbalance between respondents for whom zip-level data are available on ethnicity and marital status, but notably not on partisan identification. This coverage issues originate with Zillow.com which do not provide complete historical zip-level home value indices.

21. Question details are provided in Appendix 3.

Additional information

Funding

Funding support for this research came from the Surdna Foundation, the 11th Hour Project, and the Grantham Foundation for the Protection of the Environment.