Political economy and the adoption of everyday environmental policies in the American states, 1997: an exploratory analysis

Abstract

The political economy model has been widely and effectively used to explain and predict adoption rates of highly salient and/or controversial policies in the American states. However, use of this model to predict policy adoption in noncontroversial domains has been limited. This article tests the extent to which the model is successful in explaining the adoption of less-salient, everyday policies intended to improve environmental quality among the American states. The addition of conditional terms related to the model's political and bureaucratic components resulted in explaining 57% of the variance in commitment to everyday environmentalism among the American states. In sum, the number of everyday environmental policies adopted by state governments is a function of economic considerations, legislative accountability and professionalism, bureaucratic commitment, political culture and previous levels of policy adoption. Alternately—and unlike more controversial environmental policy domains—partisanship, ideology, party control of government and interest group forces do not have an impact on adoption of everyday environmental policies.

Notes

1 In February 2001, G.M. announced a lawsuit against California, which had announced plans to require 2% of all cars sold in the state to have “zero” tailpipe emissions. The lawsuit elevates this policy item into the realm of high politics; yet, the relevant everyday environmentalism scale item deals with state purchase of vehicles and does not reflect the “two percent” regulation.

2 Scale construction, using 1997 dichotomous data from the Book of the States, 1998–1999, employed standard iterative Guttman scaling procedures. This procedure orders items and subjects with respect to some underlying cumulative dimension—and, from each state's score it is possible to predict the state's location on each item incorporated into the scale (e.g., a scale score of “three” indicates adoption of policies 1, 2 and 3 on the scale and failure to adopt the other items in the continuum. In this research, this means that a state adopted the first three items in the 1997 scale, but not the remaining policies within the dimension. See McIver and Carmines (1981:40–41) for a discussion. Traditional statistics were employed to assess scale reliability (Cr = .90 and Cs = .65). Scores assigned to each state are “scale scores” and offer improvements over simple indices: (a) Guttman scaling forces a position assignment to each item in the scale continuum; (b) acceptable Cr and Cs values indicate that all items should be included in a scale; and (c) Guttman scaling tests the unidimensional nature of the combined items, while additive indices simply assume unidimensionality. Use of a Guttman scale to measure the dependent concept limits the inquiry, i.e., while the scale represents the cumulative number of adopted policies, nothing can be said about the relatively simplicity, complexity or stringency of the included policies. The first attempt to create a Guttman scale of everyday environmental policies, circa 1997, included a tenth item: restricting state purchasing of products with CFCs. Inclusion of this item led to Cr and Cs coefficients that were below acceptable thresholds. Eliminating this term led to creation of the acceptable Guttman scale employed in this article. To increase variation on the everyday environmentalism scale, as well as to increase discrimination among cases, a .50 bonus was added to a state's score if its entry into the scalogram evidenced no scaling errors.

3 Attempts to either combine 1993 scale items into a single scale or divide the 1997 scale into comparable 1993 scales failed by substantial margins.

4 Another way to assess if 1997 scores are a function of the 1993 time period consists of generating a change score to measure the percent increase/decrease in policy levels across time. The change score strategy was rejected for two reasons: (a) there are two 1993 scales and only one 1997 scale (determining change scores under these conditions is untenable); and (b) change scores introduce bias into regression analysis (Markus, 1979). A sounder strategy is to use the 1997 policy levels as a dependent variable and treat the 1993 scales as lagged endogenous variables (Markus, 1979; 48).

5 Multiple regression, incorporating exploratory curved and interactive terms, was used to analyze the data. Techniques recommended by Aiken and West (1992) were used to assess the utility of included curved and interactive terms, while Achen's (1982) procedure was used to reduce the equation.

Traditionally, curved and interactive terms are not included in an OLS model unless a priori theoretical justification exists for their use. In this article these terms are included in a theoretical fashion and post hoc explanations developed to explain results. Some may view this as data mining. However, use of this exploratory strategy is justified. First, most social science models are often cast only in simple linear forms. In order to rigorously test these models, unarticulated curved and interactive relationships must be explored. Second, if knowledge is to be advanced then exploratory testing is required. One way to achieve these goals is by the inclusion of curved and interactive terms. The assumption guiding this analysis is that the world conforms to multivariate explanations—and within a multivariate context, important relationships may be conditional (i.e., interactive), non-linear (i.e., curved) or both. This assumption requires keeping one point in mind: results are exploratory—and the robustness of results can only be established through replication.

The process used to include such terms is complex and must be explained. First, all independent variables were standardized prior to constructing exploratory terms. This means that factor scores, which are already standardized, are not transformed. Under these conditions, the proper coefficient to report is the regression coefficient (b), not beta. Second, the dependent variable, as recommended by Aiken and West (1992), was not transformed but, instead, left in its raw value form. Third, a linear equation was constructed based on all 15 independent variables and used as a “base” equation. The curved and interactive terms were added, one at a time, to the base equation. If the exploratory term was both statistically significant and increased the explained variance in a statistically significant fashion (p < .05 in both instances) it was retained for later inclusion. Upon exhaustion of the pool of exploratory terms, those that passed the statistical threshold were included in a new equation. In the current analysis, no curved term met both requirements for retention. The equation is noted as the “interactive equation” on Table 4.

The interactive equation confronts the “small N—many variables” problem—i.e., when employing multiple regression a minimum variable to case ratio of 1:5 is recommended (Tabachnick and Farrell, 1989:128–129). To achieve the desired ratio, the interactive equation was reduced using Achen's (1982; 64) technique. This requires deletion of independent variables not included in exploratory terms, with t-scores less than 1.40. Deletion order is structured by removing the variable with the smallest t-score below 1.40 and then re-running the equation. This continues until all variables have t-scores ≥ 1.40. Deleted variables are then re-entered, one at a time, on a first-out/first-in basis, to determine if any excluded variable—upon re-entry—meets the t-score threshold. If one meets the threshold this time, it is retained. Hence, some variables included in the reduced equation are not statistically significant. Since the reduced equation has eliminated the “chaff” from the analysis results, it is the only equation that is meaningful in a statistical sense. The reduced equation's variable to case ratio (1:4.36) is not ideal, but does represent a vast improvement over the 1:2.83 ratio of the interactive equation.