The politics of environmental enforcement: the case of the Resource and Conservation Recovery Act

Aldrich (1983), Baron (1994), Wright and Schaffner (2002), and Moon (2004) argue that party polarization is due to party activists and core partisans who heavily influence primary votes and general elections. Adams and Merrill (2003), Owen and Grofman (2006), and Hummel (2010) model elections as a two-stage process with a primary and a general election, where the constituency in the primary is significantly more partisan. In Owen and Grofman (2006), primary voters are concerned both with candidates’ policy positions and their viability in the general elections; this leads to policy divergence. Adams et al. (2010) incorporate voter turnout (with a voter “alienation threshold”) to explain candidate policy divergence even in close elections. Alternatively, politicians may simply have policy convictions or “character” (Kartik and McAfee 2007) and set policies strictly in accordance with their ideology. In the citizen-candidate model by Osborne and Slivinksi (1996) and Besley and Coate (1997) politicians cannot credibly commit to implement policies that deviate from their own preferences; because voters foresee politicians reneging on all other policy choices, politicians must make campaign promises in accordance with their most preferred policies. The “two constituencies thesis” argue that same-state senators from opposing parties offer different policies because they appeal to different constituencies (Peltzman 1984; Schmidt et al. 1996). Partial convergence models include Lindbeck and Weibull (1993) where elections are probabilistically driven by policy and, e.g., competence. Policy-motivated parties face a trade-off between policy and election chances, converging incompletely toward the median voter.

Decentralized policymaking may lead to inefficiently weak policies because of inter-jurisdictional capital competition or neglected transboundary pollution spillovers (Oates 1972; Silva and Caplan 1997; Kunce and Shogren 2007). Alternatively, decentralization may result in efficient policies or inefficiently strong policies due to absentee capital ownership or cross-hauling of capital (Wellisch 1995; Kunce and Shogren 2005, 2008; McAusland 2002; Lee 2005). There are also informational and political advantages to allocating decision making to the local level where specific environmental benefits, costs, and preferences are more easily taken into account (Oates 1972, 1999; Sigman 2003). Oates and Schwab (1988) find that both centralized and decentralized policymaking yield the first-best policy as long as no political incentives are present, while environmental policy may be too weak/strict with a heterogeneous population of voters or a Leviathan ruler. Fredriksson and Gaston (2000) find that while lobbying incentives differ across decentralized and centralized regimes, the aggregate incentives are equal resulting in equivalent policies across the two institutional designs. Using a median voter model, Roelfsema (2007) argues that environmental regulations in a decentralized system may be either too weak or too strict due to strategic delegation by the median voter. Complete decentralization could help avoid a possible bias due to the majority party in Congress favoring its own home districts, according to Fredriksson et al. (2010).

Empirical evidence suggests that US states may engage in strategic interaction in their environmental policy making (Fredriksson and Millimet 2002a, b; Fredriksson et al. 2004; Levinson 2003; Konisky 2007). Several studies find evidence of free-riding behavior among countries and US states, including Sigman (2002, 2005, 2014), Helland and Whitford (2003), and Gray and Shadbegian (2004). Banzhaf and Chupp (2012) simulate the trade-offs between decentralized versus uniform centralized control of air pollution from the US electricity sector. The centralized policy outperforms the state policy due to interstate spillovers and the marginal cost functions. Fredriksson and Wollscheid (2014) suggest that political centralization raises the stringency of decentralized environmental policies.

In other policy areas, Reed (2006) and Leigh (2008) find no effect of governor party affiliation on state taxes, while Besley and Case (2003) establish that a higher share of Democratic state legislators is associated with higher state spending. Fredriksson et al. (2013) find the Democratic governors raise taxes more that Republicans in their first incumbency period, but not overall if seen over two terms in office. Ferreira and Gyourko (2009) find no political party difference spending across US mayors, suggesting policy convergence.

These are weaker assumptions than independence of \( \left( {\varepsilon ,v} \right) \). Moreover, heteroscedastic \( \varepsilon \) and \( v \) are allowed (Newey et al. 1990).

Heckman (1979) corrects for selection bias using the control function approach.

2SLS can still work in the case of nonlinear models in endogenous variables, but its estimation requires special care. For example, one may be tempted to implement it by running a “forbidden regression” (Wooldridge 2010) without treating the nonlinear terms properly as separate endogenous variables.

These three weighting schemes differ in terms of their respective definition of “neighbors.” Contiguous neighbors are those bordering a state. The BEA regional classification is based on the homogeneity in various economic and social factors, and the Crone classification is based on similarities in economic activity. Irrespective of weighting schemes, the weights are equal for all neighboring states, and zero for non-neighboring states. As noted in Fredriksson and Millimet (2002a), the BEA regional classification system was introduced by the US Bureau of Economic Analysis in the 1950s and has remained unchanged. Crone’s (1998/1999) regional breakdown for US states is based on cluster analysis.

Our treatment variable would not be well defined if we were to include candidates from other political parties.

Separate 2SLS models are run in this case.

We also repeat our analysis with standard errors clustered at the state level. The results remain unchanged.

Note that when party affiliation is treated as endogenous, six observations have missing information on vote margin and thus a small discrepancy arises between the sample sizes in Specification A and Specifications B and C, respectively.

The similarity in punitive actions may indicate that firms behave similarly (in terms of violations) no matter the party affiliation of the governor in office; this is a topic of further research.

We are grateful to a referee for pointing this out.

While we only have suggestive evidence of this behavior, further research may address this issue.

Southern states are Alabama, Arkansas, Delaware, Florida, Georgia, Kentucky, Louisiana, Maryland, Mississippi, North Carolina, Oklahoma, South Carolina, Tennessee, Texas, Virginia, and West Virginia.

We re-estimated the same models by bootstrapping the entire estimation process. The qualitative results remain unchanged (See Table 7 in “Appendix”).