New Data on State Legislative Outcomes and Opinion in Energy and the Environment

• In Event: Representation and Direct Democracy in the US

Abstract

The increasingly polarized politics surrounding energy, environmental and especially climate change issues have constrained the adoption of these policies at the federal level. Yet these barriers are far weaker at the state level, and indeed there has been a recent surge of climate change mitigation policies being considered in state legislatures and on ballot initiatives (Rabe 2011; Bromley-Trujillo, et al. 2016). The context surrounding climate change policy has changed in many significant ways since 2010, with both parties polarizing. The number of proposals has exploded, especially in the more progressive states, which makes sense given diffusion dynamics (Berry, et. al 2015; Bromley-Trujillo, et. al 2016).

In this paper we construct a new “big data” set of legislative activity and public opinion on these related issues, giving special attention to climate change policy. Then we model legislative outcomes using multilevel data on legislators, their constituencies, and the legislative, institutional, and partisan context.

We begin by assembling the largest-ever data set of bills related to energy, the environment, and climate change at the state level (over 100,000 bills, 38,000 committee votes, 56,000 roll calls, and nearly 4 million individual votes). This is derived from Shor 2024, a dataset which includes all bills, roll calls, votes, and legislators in all 50 states from 2011-2023. That data is subject coded using the new keyword-assisted topic models just introduced in Eshima, Imai, and Sasaki 2023 to identify energy, environment, and climate legislation. We validate this machine learning approach by comparing our results with more traditional automated keyword searches (Garlick 2000) and two completely-manual subject codings (NCSL 2023, Center for the New Energy Economy 2023). Topic modeling allows much finer grain subject coding; we will have dozens of categories.

But conservative and liberal utility regulation bills are likely to be very different, so subject coding alone is not enough. We need a way to characterize the substantive policy *direction* of each bill. Thus, we join bill sponsorship data to existing party and ideology data (Shor and McCarty 2011, 2023). This allows us to use the ideal point of the median cosponsor as a proxy for a bill’s policy content.

We plan to model roll call votes at the individual and aggregate roll call levels. In addition to party and ideology, we add fine-grained opinion on environmental and climate change issues at the state legislative district level. This is derived from a series of surveys with geocoded respondents that are asked dozens of questions on energy and environmental issues. The total number of respondents is in the tens of thousands, which is necessary to estimate opinion at the state legislative district level via multilevel regression with poststratification (MRP) (Gelman and Little 1997, Park et al 2004, Lax and Phillips 2009). It leverages individual-level survey data with multilevel-model-based reweighting via high quality (typically Census) post-stratification data to adjust for sparse survey coverage across geographic units. Most widely associated with state level measures, it has been validated for smaller constituencies (Warshaw and Rodden 2012) but very seldom used. We will use opinion on specific issue items, as well as use a latent measurement model to estimate a continuous measure of district climate change opinion.

Finally, using our models we will attempt to identify the coalitions necessary to pass climate mitigation policy in a variety of contexts.

Authors

• Boris Shor, University of Houston

• Jennifer Hayes Clark, University of Houston