Public Opinion, Organized Interests, and Policy Congruence in Initiative and Noninitiative U.S. States

Transcription

1 Public Opinion, Organized Interests, and Policy Congruence in Initiative and Noninitiative U.S. States James Monogan, University of North Carolina at Chapel Hill Virginia Gray, University of North Carolina at Chapel Hill David Lowery, Universiteit Leiden abstract We consider the impact of direct democracy on state public policy by examining whether initiatives alter the effects of predictors in models of general state policy liberalism. These analyses are couched in Erikson, Wright, and McIver s (1993) opinion liberalism model of state public policy, as augmented by Gray et al. s (2004) inclusion of measures of organized interests. After presenting our theoretical expectations, we test the different ways that public opinion, organized interests, and initiatives can interact in the determination of public policy with a variety of models. We find little evidence that initiatives alter the congruence between opinion and policy. the instruments of direct democracy have become increasingly important in U.S. state politics, attracting the interest of the public and journalists, and garnering renewed attention from political scientists. In the last decade or so, direct democracy scholars have focused not only on the issues of the day, such as gay rights, abortion, the death penalty, and affirmative action (e.g., Gerber 1999; Arceneaux 2002), but have also carved out a niche in fiscal policy (e.g., Matsusaka 1995; Camobreco 1998), and have increasingly incorporated game theoretic models in their analyses (for a review see Lupia and Matsusaka 2004). While most research remains focused on voter behavior on ballot initiatives, a small but growing band of researchers has directed attention to the impact of initiatives on public policy. While this prior research teaches us a good deal, we still lack a sufficient understanding of the initiative s impact on public policy at the state level. The current research base has three shortcomings. First, most of the policies selected for study are topical ones, such as abortion regulations or same-sex marriage, and studies about these policies have unknown generalizability. State Politics and Policy Quarterly, Vol. 9, No. 3 (Fall 2009): pp by the Board of Trustees of the University of Illinois

2 fall 2009 / state politics and policy quarterly 305 The Progressives expected their reform to have a much broader impact on public policy than just shaping one realm. The availability of the initiative option was supposed to be a powerful deterrent to legislative activities of a more routine sort. Thus, we need to look at the initiative s effect on state public policy more broadly and also across a broader range of time. 1 Second, most of the initiative literature is focused narrowly on the impact of the initiative as compared to the state legislature and is not set within a general model of state policymaking. The initiative is not the only, or even the most important, instrument of democratic control in state politics. Therefore, to assess its impact, we need to evaluate the initiative within the context of a general model of state public policymaking that includes other important channels of influence, such as public opinion and organized interests. Third, aside from Elisabeth Gerber s book The Populist Paradox (1999), studies of the initiative s impact on public policy rarely include measures of organized interests, although most scholars acknowledge their influence in the initiative process. The Progressives intended to employ the initiative to sidestep the influence of entrenched economic interests on overly pliant state legislatures in the early twentieth century, although David Broder (2000) argues that powerful economic interests today have subverted the initiative for their own purposes. Thus, their role remains a critical part of the ongoing debate about the initiative. Our analysis addresses these three issues by looking at the impact of initiatives on 1980 and 2000 measures of general state policy liberalism developed by Erikson, Wright, and McIver in Statehouse Democracy (1993) and Gray et al. (2004). Importantly, the analyses are also couched in Erikson, Wright, and McIver s (EWM) opinion liberalism model of state public policy as augmented by Gray, Lowery, Fellowes, and McAtee s (GLFM) inclusion of measures of the structure of organized interests in the states. 2 In the first section of the article, we present our theoretical expectations. The second section presents the findings excluding the influence of organized interests. The third section of the article then adds the interest system measures to the models. We conclude by assessing the overall impact of the presence and use of initiatives on general state policy liberalism. the initiative and opinion-policy congruence Our basic approach entails couching initiative presence in the general model of congruence between public opinion and public policy developed by EWM. Thus, our first set of hypotheses does not yet address the role of organized interests except by implication. Further, the initiative might have several

3 306 monogan, gray, and lowery different kinds of effects. The first, following the expectations of the Progressives noted earlier, is that the simple presence of the initiative option could pressure the legislature to enact laws more in line with citizen preferences. If powerful conservative interests are omnipresent, as is often assumed, then the relationship between opinion liberalism and policy liberalism will be suppressed in noninitiative states. Indeed, Gerber (1999) found that policies on abortion and the death penalty in initiative states more closely matched the state s median voter than such policies did in states without the initiative, indicating that the initiative might have an indirect impact upon legislation. That is, the initiative can pressure state legislatures to move their legislative outputs closer to the median voter, an indirect effect above and beyond the direct impact of legislation enacted through the initiative process. To test this first hypothesis, we include a dummy variable for initiative presence in the EWM and GLFM models, along with an interaction between initiative presence and opinion liberalism. 3 Our strongest expectations, given the Progressives arguments, are for the interaction terms in the two models to be positive, indicating that the effect of opinion liberalism on policy liberalism differs between initiative and noninitiative states. This hypothesis gains particular strength from the conventional wisdom of legislative under- or non-responsiveness to the public given omnipresent conservative interests. If true, then, the emergence of responsiveness as a consequence of the initiative will produce a positive coefficient for the interaction term. We use a two-tailed test for the dummy variable estimate, however. That is, some authors (i.e., Schrag 1998; Smith 1998; Broder 2000) argue that the initiative is merely a new tool employed by entrenched interests or wealthy entrepreneurs to pursue rents. In other words, the initiative should have little effect on the opinion-policy linkage and merely add to the overall level of conservatism in state public policy. Consistent with the mixed set of expectations from the interest group literature, Matsusaka (2004) generates a mixed set of expectations from public opinion, asserting that the presence of the initiative makes a state s fiscal policy more conservative in the modern era and more liberal in the progressive era, although he does not include mass opinion as a predictor in his model of policy. 4 These first models are similar to Lascher, Hagen, and Rochlin s (1996) analysis, which also employed EWM s basic policy congruence model. More to the point, they found little evidence of such a simple effect of initiative presence. They then reasoned that since usage of the initiative tool varied widely, they should also include in the model a term for the interaction of initiative usage and ideology. This interaction also produced weak results, leading them to speculate that this failure was due to the role organized

4 fall 2009 / state politics and policy quarterly 307 interests play in the initiative process, a variable omitted in their analysis. We will replicate their basic analysis for 1980 using both a dummy measure for presence of the initiative and a measure of initiative use. 5 We will then replicate their analysis with more recent data on opinion and policy from But in our second set of models, we will also consider more directly the influence of organized interests. We refer to this first set of models as the naïve models, in that they assume either that organized interests do not matter or that they are omnipresent. By contrast, the second set will be called bias models to refer to the biases that interests might introduce to opinionpolicy congruence through the initiative. A Methodological Debate It is worth noting that there is an objection to the sort of interaction analysis we perform in this article. Matsusaka (2001) argues that studying opinion-policy congruence with general measures of opinion liberalism is problematic, since the public s exact policy preferences are unknown. He contends that states can overreact to public opinion, creating more dissonance between opinion and policy. Thus, greater congruence might mean a steeper slope or a shallower slope but the two cannot be sorted out without an exact measure. Hagen, Lascher, and Camobreco (2001, 1260) challenge this idea by analytically showing that if the opinion index is a linear transformation of an unknown policy preference, then the coefficient on the interaction term will be zero only if the relationship between opinion and policy is not conditional on the presence of the initiative process. Since only two of their nine models produced significant (and negative) t-tests on the interaction between opinion and the presence of the initiative, they concluded that the initiative did not influence opinion-policy congruence (Lascher, Hagen, and Rochlin 1996). Our reading of this exchange is that we can proceed to use an interaction term to assess relative congruence. While the interpretation of a statistically significant negative or positive interaction effect might be contested, an interaction that is statistically indistinguishable from zero indicates that the presence of the initiative process does not condition opinion-policy congruence in the U.S. states. Expectations for Heteroskedasticity Before considering more directly the role of organized interests, however, we also want to consider one other way in which the initiative might matter. Rather than having an impact on the level of policy liberalism produced by a given level of opinion liberalism, the residual variance of policy among initiative states might be smaller than its variance among noninitiative states. That

5 308 monogan, gray, and lowery is, initiative states have less institutional friction, to use Baumgartner and Jones term (Baumgartner and Jones 1993; Jones and Baumgartner 2005), or fewer defects in the language of the Progressive reformers, so we might expect more regular adjustments of policy to conform to public opinion. Initiative states also offer access to larger agendas, governed by different rules, than state legislatures; thus, they decrease the cognitive friction of gaining attention to the public s business as well as overcoming institutional friction. On the other hand, in the set of noninitiative states, some states could have a convergence of opinion and policy after a recent policy punctuation has upset the policy equilibrium. But there will be a countervailing divergence of the opinion-policy relationship in other states if institutional stickiness has not been overwhelmed by a recent policy punctuation. So with more friction, some states will have made a major shift toward policy congruence, while others might not have. This could produce a wide range of policy liberalism given the varying discrepancies between opinion and policy. Initiatives, however, could induce major punctuations toward opinion-policy congruence, meaning that fewer states are out of equilibrium, which implies that policy liberalism will be less widely varied. Hence, the variance in policy liberalism that is unexplained by public opinion should be smaller in initiative states than in noninitiative states. The Role of Organized Interests So far, we have not directly considered the role of organized interests. The relationship between organized interests and the initiative remains at the heart of our analysis because of both the role they played in the Progressives diagnosis of the ills meriting adoption of the instruments of direct democracy and because of the more recent debates about whether the influence of organized interests is diminished or enhanced by the initiative. We wait to consider the role of organized interests, in our aforementioned bias models, until after examining the baseline or naïve models; given the many interactions in our models, there is a strong possibility that collinearity will influence our results. Therefore, we run the simplest specification first, to ensure the results are not sensitive to the inclusion of additional predictors. Importantly, GLFM found in their earlier analysis mixed support for the expectation that organized interests bias the translation of citizen opinion into public policy. 6 Yet, their findings might well have been biased by failure to consider the impact of initiatives. Following GLFM, we will respecify the models discussed above to consider two ways in which the structure of the community of organized interests in the states might influence general opinion-policy congruence. Their first

6 fall 2009 / state politics and policy quarterly 309 hypothesized source of bias results from interest system density as measured by the total number of organized interests active in a state. More specifically, they examined three competing hypotheses. First, the rent-seeking model suggests that density directly reduces the degree of policy liberalism provided in a state. Under this idea, organized interests primarily seek private goods from the government. If elites are overrepresented among organized interests, then successful rent-seeking behavior draws scarce government resources away from programs that might be redistributive or directed toward greater public goods. Ultimately, then, policy will become more conservative. Second, the pluralist model suggests that density interacts with public opinion to provide either more or less policy liberalism, but probably more. This idea reflects the view that organized interests will emerge to represent a broad array of policy needs in society and transmit these preferences to the government. On the one hand, it could be that a conservative public, when represented by more numerous interest groups, will press for less policy liberalism. On the other hand, given that several conservative interests, such as for-profit businesses, can more easily solve problems of organization and are thereby present even in small interest communities, a larger number of groups probably indicates that liberal interests are gaining interest representation relative to conservative interests, such that they can promote liberal policy. Finally, the policy lurches model suggests that density weakens the opinion-policy relationship in a nondirectional manner so that EWM s results should be heteroskedastic with respect to density. GLFM further expected that interest system diversity would influence the translation of opinion into policy. Diversity was operationalized as the dominance of for-profit interest groups, versus nonprofit groups, in a state as measured by the proportion of the former. High for-profit dominance should bias public policy in a conservative direction. The last step is to consider how the expectations of the bias models discussed above might work in the initiative setting. To date, the relation between the initiative process and organized interests has been studied mostly from the reverse causal standpoint (Boehmke 2005, 2008; Smith and Tolbert 2004; Banducci 1998). Here, we are interested in how the structure of the interest community will modify opinion-policy congruence differently in initiative and noninitiative states. Basically, the answer depends on whether organized interests work to impede opinion-policy congruence, as suggested by transactions theorists, or whether they are instead vital instruments through which congruence is achieved, as suggested by the pluralists (Lowery and Gray 2004). Using two-tailed tests, we generally expect that the hypotheses outlined earlier for the naïve model will be either attenuated or enhanced in the models controlling for the density and diversity of interest communities. We test these

7 310 monogan, gray, and lowery competing expectations with additional interactions between the variables already discussed for the naïve model and the organized interest variables. testing the models To test these hypotheses, first we replicate the EWM model. Their composite policy liberalism index, the dependent variable, was based on data from the early 1980s on eight issues on which liberals and conservatives disagreed in predictable ways (EWM 1993). Our second analysis entails testing the EWM model in the contemporary period. A more recent policy liberalism measure is an index developed by Gray et al. (2004). It was based on five state-level issues that currently divide liberals and conservatives: gun control, abortion, welfare eligibility and activity requirements under TANF (the successor to AFDC), right-to-work laws, and the progressivity of the tax structure. For both measures, the component issue measures were standardized and summed. By using these measures for 1980 and 2000, we can study general policy including both social and economic measures. When modeling the 1980 and 2000 policy liberalism indices, an exogenous variable could have an effect in two basic ways. First, the input variable could explain a large amount of variance in a few elements of the index. By extensively explaining variance in liberalism of TANF and tax progressivity, for example, a predictor like public opinion could show a significant effect on general policy liberalism. A small effect on such programs alone, however, would not suffice to conclude that the predictor affected general policy liberalism. Alternatively, the predictor could moderately explain variance in nearly every component of the index, which would become manifest in an explanation of overall policy liberalism. Thus, it is possible for predictors to affect individual issues without affecting general policy liberalism or for policy responses across a range of issues to aggregate into an overall effect. In modeling policy in both 1980 and 2000, we will work with a modified version of EWM s final path model (1993, 130), in which policy liberalism was determined by two independent variables in 1980: state opinion liberalism and legislative liberalism. For 1980, opinion liberalism was measured by the standard ideology question posed in CBS/NYT telephone polls between 1976 and For 2000, we use public opinion liberalism scores for each state averaged over the period (Wright website: For 1980, EWM measured legislative liberalism indirectly: parties legislative strength was used to weight previously derived party elite ideology scores to produce a measure of each state legislature s liberalism, with nonpartisan

8 fall 2009 / state politics and policy quarterly 311 Nebraska omitted. This measure was based on surveys of Congressional candidates (taken by CBS/NYT), state legislators (Uslaner and Weber 1977), local party chairpersons (Cotter et al. 1984), and national convention attendees (Miller and Jennings 1986). Many of these studies do not have a more contemporary analogue, so the most similar measure we can use for 2000 is Berry et al. s (1998) measure of government ideology, which is based on the governor s party, vote shares in congressional elections, roll-call voting of the congressional delegation, and party division in the state legislatures. Table 1 shows the descriptive statistics for each year s policy liberalism indices, as well as the predictors. For each variable, Table 1 shows the mean, standard deviation, minimum, and maximum. 7 For both time periods, interest density is the count of registered organized interests in the state, that economic dominance is the proportion of these interests that have a for-profit orientation. Finally, as explained earlier, the moderator variables of our models are the presence and use of initiatives. For both 1980 and 2000, the presence variable is a simple dummy for whether a state has provisions to place an initiative on the ballot. 8 To operationalize initiative usage for our 1980 models, we count the number of times an initiative was on the ballot from 1972 to For models of 2000, we similarly count the number of initiatives on a state s ballots between 1992 and We study a span of years for each time period to determine whether initiatives are regularly used in a state. Examining only one year might give, by chance, an inaccurate perception of a state s usage tendencies. When setting a range of years, we chose a range preceding the year of study because the use of direct democracy should precede its consequences. Also, many of the measures of opinion and policy liberalism use data from a number of years, so this range encompasses the time span of measurement. Table 1. Descriptive Statistics for 1980 and 2000 Measures Variable Mean Standard Deviation Minimum Maximum EWM policy ideology Interest density Economic dominance Opinion liberalism Legislative liberalism GLFM policy ideology Interest density Economic dominance Opinion liberalism Legislative liberalism

9 312 monogan, gray, and lowery Naïve Model Results The first three columns of Table 2 report results for 1980 for our simplest expectations about the impact of initiatives. 10 The results in the first column replicate a model from Statehouse Democracy in which policy liberalism is a function of public opinion ideology and legislative liberalism. These results match those reported by EWM, with increases in public opinion liberalism and legislative liberalism each leading to an increase in liberal policy. The results in the second column test the hypothesis that opinion will more strongly predict policy in initiative states by using a dummy variable for the presence of an initiative law and interacting this dummy with public opinion. While the main effects for public opinion and legislative liberalism remain significant, the hypothesis is not supported. The dummy estimate for the presence of the initiative is not significant, although we did not have a strong prior for this coefficient. More importantly, the interaction is not positive, as we expected. Indeed, under a two-tailed test, the negative coefficient would have been significant. In other words, the positive association between opinion and policy liberalism is potentially smaller in initiative states than in noninitiative states. Similarly, the third column tests the hypothesis that initiative usage increases the congruence between opinion and policy using the count of initiative use from and interacting this with public opinion. Using a two-tailed test to reflect our lack of a strong prior, the coefficient for the count of initiatives is significantly greater than zero. Thus, increased use of initiatives is associated with policy that is more liberal. The interaction between the count variable and public opinion, however, fails to support the hypothesis; the coefficient is neither positive nor significant, but negative and indiscernible from zero. The last three columns of Table 2 report the comparable results for This model uses the GLFM policy liberalism index as the dependent variable. For comparability, it includes Berry et al. s (1998) measure of government ideology in lieu of EWM s measure of legislative liberalism. 11 For the baseline model of the fourth column, the estimate of public opinion liberalism is again positive and significant. Moving to substantive considerations, the fifth column reports insignificant estimates for the initiative dummy and its interaction with public opinion, which does not support the hypothesis that policy liberalism should be more congruent with opinion in initiative states. The sixth column includes a count of ballot initiatives for each state from and the interaction between this count and opinion liberalism. Again, the data do not support our hypotheses for the year 2000; the interaction term is not significant, as policy is not noticeably more responsive to public opinion.

10 Table 2. Naïve Model of Opinion-Policy Congruence for 1980 and Models 2000 Models Independent Variable Baseline Initiative Presence Initiative Usage Baseline Initiative Presence Initiative Usage Opinion liberalism 0.084* (0.019) 0.104* (0.020) 0.077* (0.019) 0.273* (0.036) 0.269* (0.047) 0.278* (0.040) Legislative liberalism 0.157* (0.083) 0.232* (0.082) 0.170* (0.081) (0.014) (0.015) (0.015) Initiative presence (0.154) (0.613) Initiative presence opinion 0.061# (0.022) (0.069) liberalism Initiative use 0.036# (0.016) (0.032) Initiative use opinion (0.002) (0.004) liberalism Constant (0.081) (0.108) (0.081) (0.295) (0.449) (0.302) R *p<0.05, one tailed test; #p<0.05, two tailed test Note: Figures under coefficients are standard errors N=42; 2000 N=48

11 314 monogan, gray, and lowery Our final hypothesis in the naïve model is that the variance of policy after considering the relationship between opinion and policy ought to be lower in initiative than in noninitiative states. Our test examined the mean squared error after fitting the baseline models of Table 2 to the sub-samples of initiative and noninitiative states. Larger mean squared error indicates greater unexplained variance; the results do not support this expectation. In both 1980 and 2000, the mean squared error is higher for initiative states (0.278 and 5.205, respectively) than noninitiative states (0.251 and 2.160, respectively), the reverse of expected. Thus, the hypothesis that initiatives can help to overcome institutional friction and thereby better link opinion and policy is not supported for the naïve model. Like Robinson et al. (2007), we have been unable to confirm factors that promote punctuated behavior in public policymaking. Expanding the Specification for Interest Bias Given the weakness of the naïve initiative hypotheses, we now turn to a model that controls for the influence of organized interests on opinion-policy congruence, which we call bias models. Starting with 1980, Table 3 reports the results for a model that includes a dummy for initiative presence and its interaction with all variables from the baseline model derived from Gray et al Table 4 reports the same results, but with initiative usage substituted for the initiative dummy. Since the results for these two tables are remarkably similar, we discuss them together. The only variables that are significant in the two tables in any of the specifications are public opinion, which always is significant, and legislative liberalism, which loses significance in some of the interaction models. Importantly, none of the interest organization variables produced estimates discernable from zero. 13 Our most important concern, however, is whether the inclusion of the interest organization variables altered the results for the initiative variables. They did not. None of the initiative variables or interactions in either Table 3 or Table 4 produced a statistically significant estimate. Turning to the 2000 period when initiative usage was higher, Table 5 reports models of the GLFM policy index with the initiative dummy, while Table 6 reports results for the same model, but with initiative usage substituted for the initiative dummy variable. Again, the results are quite similar across the two tables, allowing us to discuss them together. Public opinion liberalism has the expected positive, significant relationship in all specifications. In one instance, the interaction between interest density and public opinion is positive and significant, indicating that opinion liberalism has a more positive impact on policy liberalism in states with dense interest

12 Table 3. Bias Model of Opinion-Policy Congruence with Initiative Presence Dummy, 1980 Model Specification With Presence With Presence With Dummy Independent Variable Baseline Dummy Interactions & Interactions Interest density (0.046) (0.047) (0.066) (0.067) Economic dominance (1.550) (1.593) (1.904) (1.984) Opinion liberalism 0.087* (0.020) 0.088* (0.021) 0.123* (0.027) 0.124* (0.028) Legislative liberalism 0.163* (0.083) 0.159* (0.086) (0.138) (0.146) Initiative presence (0.168) (0.164) Initiative presence interest density (0.091) (0.093) Initiative presence economic dominance (3.293) (3.355) Initiative presence opinion liberalism (0.040) (0.041) Initiative presence legislative liberalism (0.179) (0.185) Constant (0.080) (0.121) (0.080) (0.123) R *p<0.05, one tailed test; #p<0.05, two tailed test Note: Figures under coefficients are standard errors. N=42

13 Table 4. Bias Model of Opinion-Policy Congruence with Initiative Use Variable, 1980 Model Specification Independent Variable Baseline With Initiative Usage With Usage Interactions With Usage & Interactions Interest density (0.046) (0.047) (0.059) (0.049) Economic dominance (1.550) (1.567) (1.778) (1.752) Opinion liberalism 0.087* (0.020) 0.085* (0.021) 0.096* (0.022) 0.089* (0.022) Legislative liberalism 0.163* (0.083) 0.166* (0.084) (0.087) 0.149* (0.086) Initiative usage (0.013) (0.017) Initiative usage interest density (0.008) (0.008) Initiative usage economic dominance (0.404) (0.406) Initiative usage opinion liberalism (0.004) (0.004) Initiative usage legislative liberalism (0.012) (0.012) Constant (0.080) (0.080) (0.084) (0.084) R *p<0.05, one tailed test; #p<0.05, two tailed test Note: Figures under coefficients are standard errors. N=42

14 Table 5. Bias Model of Opinion-Policy Congruence with Initiative Presence Dummy, 2000 Model Specification With Presence With Presence With Dummy Independent Variable Baseline Dummy Interactions & Interactions Interest density (0.072) (0.072) (0.115) (0.115) Interest density opinion liberalism (0.008) 0.015* (0.009) (0.013) (0.013) Economic dominance (6.116) (6.312) (11.028) (11.302) Opinion liberalism 0.269* (0.038) 0.267* (0.038) 0.293* (0.050) 0.291* (0.051) Legislative liberalism (0.015) (0.015) (0.026) (0.026) Initiative presence (0.636) (0.699) Initiative presence interest density (0.186) (0.189) Initiative presence interest density (0.020) (0.020) x opinion liberalism Initiative presence economic dominance (14.254) (14.333) Initiative presence opinion liberalism (0.093) (0.093) Initiative presence legislative liberalism (0.033) (0.033) Constant (0.296) (0.439) (0.347) (0.477) R *p<0.05, one tailed test; #p<0.05, two tailed test Note: Figures under coefficients are standard errors. N=48

15 Table 6. Bias Model of Opinion-Policy Congruence with Initiative Use Variable, 2000 Model Specification Independent Variable Baseline With Initiative Usage With Usage Interactions With Usage & Interactions Interest density (0.072) (0.073) (0.099) (0.105) Interest density opinion liberalism (0.008) (0.009) (0.012) (0.012) Economic dominance (6.116) (6.626) (6.634) (8.175) Opinion liberalism 0.269* (0.038) 0.266* (0.038) 0.258* (0.051) 0.266* (0.054) Legislative liberalism (0.015) (0.015) (0.017) (0.017) Initiative usage (0.025) (0.057) Initiative usage interest density (0.013) (0.014) Initiative usage interest density (0.001) (0.001) x opinion liberalism Initiative usage economic dominance (0.525) (0.807) Initiative usage opinion liberalism (0.007) (0.008) Initiative usage legislative liberalism (0.002) (0.002) Constant (0.296) (0.298) (0.324) (0.391) R *p<0.05, one tailed test; #p<0.05, two tailed test Note: Figures under coefficients are standard errors. N=48

16 fall 2009 / state politics and policy quarterly 319 systems. But the comparable estimate in the full model was not significant. More importantly, however, none of the interaction effects is significant. It is also worth noting that in all of the models from Tables 3 through 6, the coefficients for the interaction between economic dominance and initiative presence or usage were uniformly insignificant. Not only was this the case, but all post hoc t-ratios from substitution fail to produce a significant effect for economic dominance. 14 This result fits well with Gerber s finding that economic interests are generally ineffective at defeating proposed initiatives, instead using the initiative process as a sometimes credible threat in the legislature (1999). Thus, the model does not support the idea that ballot initiatives unequivocally empower interest groups. Finally, regarding the variance hypothesis for the bias model, we examined the mean squared error of the baseline bias models when they are applied to the sub-samples of initiative and noninitiative states in 1980 and Again, our expectation was that variance would be lower for the sub-sample of initiative states because ballot initiatives provide an opportunity to overcome institutional and cognitive friction. This hypothesis is again not supported because the mean squared error is higher for initiative states (0.270 and 5.207, respectively) than for noninitiative states (0.217 and 3.268, respectively). Thus, the variance hypothesis results for both the naïve and bias specifications fail to support punctuated equilibrium theory, which fits well with the result presented in Robinson et al. (2007). conclusion Throughout our analysis, the importance of public opinion as a predictor of policy liberalism is consistently clear. We have seen that public opinion significantly influences policy in both initiative and noninitiative states, and in much the same way. More to the point, we have seen that neither the presence of the initiative nor its level of use seems to alter this basic relationship between opinion and policy liberalism. Just as important, this basic finding survived inclusion of controls for measures of the structure of state interest communities, which lies as the heart of both the Progressives advocacy of direct democracy and contemporary debates over whether the promise of the Progressives reforms have strengthened or weakened opinion-policy congruence. Clearly, the initiative has had remarkably little effect on the overall direction of public policy, despite the great attention given to it in recent years. Lupia and Matsusaka (2004) argue that extant literature on the initiative process paints a picture of the initiative as a majoritarian device that increases

17 320 monogan, gray, and lowery policy congruence. Yet our study agrees with the findings of Lascher, Hagen, and Rochlin (1996) and Camobreco (1998), which together demonstrate that the initiative does not always increase congruence. Our study also fits with the findings of Burden (2005), who shows that the initiative does not enhance responsiveness to general policy liberalism. By placing the initiative process within a generalized model of state policymaking and including measures of organized interests, we provide a more accurate assessment of its impact. Controlling for the impact of public opinion on the public policy index, we do not detect much impact by the initiative in either time period, nor do we find much impact by organized interests. In addition, we use the policy punctuation literature to propose an innovative test not previously employed: that the initiative process should reduce institutional friction and stickiness and that the variance of the opinion-policy linkage among initiative states should be smaller than its variance among noninitiative states. But this test also fails to produce a positive result for the efficacy of the initiative on policymaking. We cannot find evidence in our results that the initiative must be present in order for the states to have majoritarian government. Nevertheless, the initiative can have a significant impact on the adoption of specific policies. Indeed, we accept prior conclusions that initiatives matter on specific policies including abortion (Gerber 1999; Arceneaux 2002; Burden 2005), the death penalty (Gerber 1999), and governance reform (Pippen, Bowler and Donovan 2002). 15 Additionally, others have demonstrated that the initiative has educative effects on voters (Smith and Tolbert 2007). But these policy results could well be the exception to the more general rules governing policymaking in the states to the extent that the 1980 EWM and 2000 GLFM policy liberalism indices actually measure what they purport to measure generalized policy liberalism. We believe that these indices are such general measures, as each is composed of a variety of items that cluster together to form an index of policy liberalism. The results are robust across time periods and versions of the policy liberalism index; in all of our models the initiative does not alter the opinion-policy linkage. The relationship between the public s desires and state government policies remains strong and close, whether achieved via direct or representative democracy. endnotes An earlier version of this article was presented at the Annual State Politics and Policy Conference, Austin, Texas, February 22 24, The authors appreciate the helpful suggestions of Todd Donovan, Liz Gerber, Thad Kousser, Andrea McAtee, Eric Plutzer,

18 fall 2009 / state politics and policy quarterly 321 Gerald Wright, and the anonymous reviewers, and they thank Erik Godwin for his research assistance. Replication data are available at the State Politics and Policy Quarterly archive at the University of Dayton. 1. Research on state tax revolts of the 1970s found some issues that are the focus of ballot measures have a marked periodicity in popularity (Sigelman, Lowery, and Smith 1983). 2. Burden (2005) runs similar models to test for institutional effects in representation. He models EWM s 1980 general policy liberalism measure as well as 1990 measures of abortion parental consent and notification policy and death penalty policy. Our models keep the indirect effects of initiatives more explicitly in mind and study general policy liberalism, not just in 1980 but in 2000 also. This goal also leads us to include measures of the interest group population in the models, another difference from Burden s research. 3. Marschall and Ruhil (2005) argue that estimates of initiatives effects on fiscal policy can suffer from endogeneity bias if the model does not consider that several factors can simultaneously affect the establishment of the initiative and policy decisions. That argument is directed primarily at models suggesting that the institution of the initiative has a direct effect on policy. Since our models ask whether policy is relatively congruent with opinion in initiative states, this endogeneity argument is less of a concern. 4. The empirical evidence Matsusaka presents is not so firm as to offer any settled conclusions about state fiscal policy in the presence of the initiative. In the time period the initiative dummy is always insignificant (Matsusaka 2004, 75), and in the period, when 99 percent of variance is explained, the initiative dummy is significant only at the.10 level for revenue and at the.05 level for expenditures. With the addition of the signature requirement to the model, the dummy variable for initiative is nonsignificant for revenue and only significant at the.10 level for expenditures (p. 32). He also reports the estimated annual effect of the initiative on expenditures for ; there was a significant effect at the.05 level in only 14 of these 43 years. 5. Bowler and Donovan (2004) show that of four measures of the initiative, measurement by frequency of usage shows the initiative having the greatest impact on policy and politics. 6. GLFM found that a model including opinion liberalism (positive), legislative liberalism (positive), and economic dominance (negative) fit best in Interest density and its interaction term were not significant. For 2000, EWM s opinion liberalism variable again produced a positive and significant estimate, but legislative liberalism did not. Of the organized interest variables, density was positive and significant in both direct and interactive terms; economic dominance was weakly significant (at the 0.15 level) and negative, but the interactive term was not significant. So the best fit model for 2000 was opinion liberalism, density, a density-opinion interaction, and economic dominance. 7. See GLFM (2004) for a discussion of these measures construction and validity. 8. Mississippi is the only state that changes from 1980 to 2000, adopting the initiative in Illinois is included as an initiative state in each period of study, although it should be noted that only matters related to the structure of the legislature can be considered. 9. The Initiative and Referendum Institute at USC, The sample sizes differ for the two years due to data limitations. Replicating the 2000 models for the same sample as 1980 does not change the results presented in any table.

19 322 monogan, gray, and lowery 11. Omitting legislative liberalism yields similar results, which is important since research suggests legislative ideology is unimportant in that time frame (Gray et al. 2004, 419). 12. We centered all continuous variables at zero, which many believe diminishes the correlation between predictors, to counteract possible multicollinearity from interactions. The only consequence for our estimates is a change in the value of the constant. In the final column of Table 2, two predictors had variance inflation factors (VIFs) exceeding 10: legislative liberalism and the interaction between legislative liberalism and initiative presence. An F-test for these two legislative liberalism variables together was insignificant: F (2, 32) =1.82. In Tables 4 and 5, none of the VIFs exceeds 10 in the final models. In Table 6 s final model, the VIFs for the interaction between initiative use and interest group density and the interaction between use, density, and opinion exceeded 10. The F-test on these interactions and the main effects was insignificant: F (4, 36) = The models in Tables 3 6 also were run with economic policy as the dependent variable in case this was the sole area in which interests had an effect. Economic policy in 1980 was measured with laws on gambling, consumer protection, and tax progressivity, and in 2000 it consisted of right-to-work states and tax progressivity. All results had the same sign and significance. We also ran models that replaced the initiative presence dummy with a dummy for permitting statutory initiatives and again with a dummy for constitutional initiatives. Some insignificant coefficients changed signs, but the results were similar. 14. This includes inserting quartile levels of initiative use into the count variable specification. Post hoc t-ratios are described in Brambor, Clark, and Golder (2006). 15. If initiatives increase good government legislation, and such legislation increases congruence, then initiatives might indirectly affect long-term opinion-policy congruence. references Arceneaux, Kevin Direct Democracy and the Link between Public Opinion and State Abortion Policy. State Politics and Policy Quarterly 2: Banducci, Susan A Searching for Ideological Consistency in Direct Legislation Voting. In Citizens as Legislators: Direct Democracy in the United States, eds. Shaun Bowler, Todd Donovan, and Caroline J. Tolbert. Columbus, OH: Ohio State University Press. Baumgartner, Frank R., and Bryan D. Jones Agendas and Instability in American Politics. Chicago, IL: University of Chicago Press. Berry, William D., Evan J. Ringuist, Richard C. Fording, and Russell L. Hanson Measuring Citizen and Government Ideology in the American States, American Journal of Political Science 42: Boehmke, Frederick J The Indirect Effect of Direct Legislation: How Institutions Shape Interest Group Systems. Columbus, OH: Ohio State University. Boehmke, Frederick J The Initiative Process and the Dynamics of State Interest Group Populations. State Politics and Policy Quarterly 8: Bowler, Shaun, and Todd Donovan Measuring the Effect of Direct Democracy on State Policy: Not All Initiatives Are Created Equal. State Politics and Policy Quarterly 4:

20 fall 2009 / state politics and policy quarterly 323 Brambor, Thomas, William Roberts Clark, and Matt Golder Understanding Interaction Models: Improving Empirical Analyses. Political Analysis 14: Broder, David Democracy Derailed. New York, NY: Harcourt. Burden, Barry C Institutions and Policy Representation in the States. State Politics and Policy Quarterly 5: Camobreco, John F Preferences, Fiscal Policies, and the Initiative Process. The Journal of Politics 60: Cotter, Cornelius, James Gibson, John Bibby, and Robert Huckshorn Party Organizations in American Politics. New York, NY: Praeger. Erikson, Robert S., Gerald C. Wright, and John P. McIver Statehouse Democracy. New York, NY: Cambridge University Press. Gerber, Elisabeth R The Populist Paradox: Interest Group Influence and the Promise of Direct Legislation. Princeton, NJ: Princeton University Press. Gray, Virginia, David Lowery, Matthew Fellowes, and Andrea McAtee Public Opinion, Public Policy, and Organized Interests in the United States. Political Research Quarterly 57: Hagen, Michael G., Edward L. Lascher Jr., and John F. Camobreco Response to Matsusaka: Estimating the Effect of Ballot Initiatives on Policy Responsiveness. The Journal of Politics 63: Jones, Bryan D., and Frank R. Baumgartner The Politics of Attention: How Government Prioritizes Problems. Chicago, IL: University of Chicago Press. Lascher, Edward L., Jr., Michael G. Hagen, and Steven A. Rochlin Gun behind the Door? Ballot Initiatives, State Policies, and Public Opinion. The Journal of Politics 58: Lowery, David, and Virginia Gray A Neopluralist Perspective on Research on Organized Interests. Political Research Quarterly 57: Lupia, Arthur, and John G. Matsusaka Direct Democracy: New Approaches to Old Questions. Annual Review of Political Science 7: Marschall, Melissa J., and Anirudh V. S. Ruhil Fiscal Effects of the Voter Initiative Reconsidered: Addressing Endogeneity. State Politics and Policy Quarterly 5: Matsusaka, John G Fiscal Effects of the Voter Initiative: Evidence from the Last 30 Years. The Journal of Political Economy 103: Matsusaka, John G Problems with a Methodology Used to Evaluate the Voter Initiative. The Journal of Politics 63: Matsusaka, John G For the Many or the Few. Chicago, IL: University of Chicago Press. Miller, Warren E., and M. Kent Jennings Parties in Transition: A Longitudinal Study of Party Elites and Supporters. New York, NY: Russell Sage Foundation. Pippen, John, Shaun Bowler, and Todd Donovan Election Reform and Direct Democracy: Campaign Finance Regulations in the American States. American Politics Research 30: Robinson, Scott E., Floun say Caver, Kenneth J. Meier, and Laurence J. O Toole, Jr Explaining Policy Punctuations: Bureaucratization and Budget Change. American Journal of Political Science 51: Schrag, Peter Paradise Lost: California s Experience, America s Future. New York, NY: Free Press.

21 324 monogan, gray, and lowery Sigelman, Lee, David Lowery, and Roland Smith The Tax Revolt: A Comparative State Analysis. Western Political Quarterly 6: Smith, Daniel A., and Caroline Tolbert Educated by Initiative: The Effects of Direct Democracy on Citizens and Political Organizations in the American States. Ann Arbor, MI: University of Michigan Press. Smith, Daniel A., and Caroline Tolbert The Instrumental and Educative Effects of Ballot Measures: Research on Direct Democracy in the American States. State Politics and Policy Quarterly 7: Smith, Daniel A Tax Crusaders and the Politics of Direct Democracy. New York, NY: Routledge. Uslaner, Eric M., and Ronald E. Weber Patterns of Decision Making in State Legislatures. New York, NY: Praeger.