Direct Democracy and Political Efficacy Reconsidered

Transcription

1 DOI /s x ORIGINAL PAPER Direct Democracy and Political Efficacy Reconsidered Joshua J. Dyck Æ Edward L. Lascher Jr. Ó Springer Science+Business Media, LLC 2008 Abstract Some studies have contended that direct democracy has secondary benefits unrelated to its impact on policy. In particular, recent scholarship claims that the American ballot initiative process enhances political efficacy. We began with concerns about the logic and empirical methods underlying this conclusion. We connect this research to the broader political psychology literature and in doing so find little reason to expect a positive relationship between direct democracy and efficacy. Our other contribution is to subject the empirical claim to more extensive testing. In contrast to prior research, we draw from multiple data sources and consider sampling methods. The results consistently fail to indicate that direct democracy generally enhances political efficacy. We find cause for skepticism about the secondary benefits of the ballot initiative process. Keywords Direct legislation Ballot initiatives Political efficacy Introduction On balance, is the presence of direct democracy beneficial? This is a question that has long interested political scientists and some other scholars, though systematic empirical research relevant to answering this question is comparatively recent. However, in the past two decades there has been a boom in such studies. Much of the extant research has focused on the ballot initiative process in the United States, J. J. Dyck (&) Political Science Department, University at Buffalo, SUNY, Buffalo, NY 14260, USA jdyck@buffalo.edu E. L. Lascher Jr. Department of Public Policy and Administration, California State University Sacramento, Sacramento, CA , USA tedl@csus.edu

2 taking advantage of the sharp variation in usage across the states to draw inferences about the effects of direct democracy; some work also compares access to direct democracy across local governments in the U.S. or sub-national jurisdictions in other countries, such as Switzerland. Where once empirically-oriented academic journals ignored the ballot initiative, it is now common to find multiple articles about this topic in a single volume. Much of the recent empirical work has examined topics directly related to the types of public policy choices encouraged by direct democracy. Consistent conclusions are not the norm. Some scholars have offered findings suggesting that direct democracy leads to policies that are closer to what the public wants (see for example Matsusaka 2004), that the ballot initiative process avoids disproportionate harm to minority group interests (e.g., Hajnal et al. 2002), and that information cues allow people to make well reasoned choices in the absence of encyclopedic knowledge about ballot measures (e.g., Bowler and Donovan 1998; Lupia 1994; for an article summarizing much of the positive evidence in all three of the above areas see Matsusaka 2005). Quite to the contrary, others have presented arguments and evidence suggesting that the presence of direct democracy does not lead to outcomes more in accord with public opinion (e.g., Lascher et al. 1996), that minority groups such as gays and lesbians are likely to be disproportionate losers in ballot contests, 1 and that voters often use very limited information in making choices on ballot measures (e.g., Magleby 1984). In the last several years, a new line of debate has commenced about the merits of direct democracy unrelated to effects on specific public policies. Scholars have begun examining whether the ballot initiative process has beneficial secondary consequences in areas such as civic engagement. In their book length study of such consequences, Smith and Tolbert (2004) emphasize that one of the major original rationales for establishing the ballot initiative process in the American states was the belief that it would have educative effects on the citizenry, aside from any instrumental benefits for public policy. Accordingly, Smith, Tolbert, and others have turned their attention to assessing such claims through new sets of comparison across the states. Evidence has been presented indicating that greater use of direct democracy leads to higher voter turnout (Smith and Tolbert 2004, chap. 2), greater knowledge about political matters (Smith 2002; Smith and Tolbert 2004, chap. 3), enhanced sense of political efficacy (Bowler and Donovan 2002; Hero and Tolbert 2004; Mendelsohn and Cutler 2000; Smith and Tolbert 2004, chap. 4), and even greater happiness or life satisfaction (Frey and Stutzer 2000a, b; but see Dorn et al. 2007). Given continued controversy about policy impact, such arguments provide an empirical basis for normative arguments in favor of direct democracy. Scholars have offered especially strong claims about the impact of direct democracy on political efficacy. For example, the last sentence of the abstract for the 2002 Bowler and Donovan article (p. 371) reads as follows: We present 1 See especially Haider-Markel et al. (2007) for a useful review of the entire literature in the area of direct democracy and minority interests as well as further evidence of a negative impact. On the logic underlying concerns about protections of rights under direct democracy, see also Ellis (2002, chap. 5) and Gamble (1997).

3 evidence that the effects of exposure to direct democracy on internal and external political efficacy rival the effects of formal education. Additionally, Smith and Tolbert explicitly argue that simply residing in a state where direct democracy is used more commonly enhances people s political efficacy. They write (2004, pp ): We find strong evidence that citizens living in states with frequent use of direct democracy are more likely to claim that government is responsive to their needs, controlling for other factors. However, we argue that there is reason for doubt about the positive impact of direct democracy on political efficacy. The extant literature in this area is not well connected to broader research on political psychology, and one of the contributions of this article is to make that connection. In so doing we find little reason to expect a positive relationship between ballot initiative usage and efficacy. On the contrary, we find that such a relationship is inconsistent with findings about what information people tap in answering survey questions, and especially the relatively small amount of attention people give to state politics and institutions. Furthermore, earlier studies gloss over research casting doubt on whether simply making more decisions, or merely participating in politics, enhances efficacy. Additionally, we find that if direct democracy is to have any impact on efficacy it is likely to be among particular subsets of the general population, such as more informed citizens or voters as opposed to non-voters and the less informed. Yet the existing literature largely does not consider such distinctions. Our other contribution is empirical. We subject the claim about the connection between direct democracy and political efficacy to a far more extensive series of tests than has been conducted in the past. Prior studies have been based on analysis of a single data set, but we use multiple sources of information including the American National Election Studies (ANES), the General Social Survey, and the 1997 PEW Trust in Government study. Based on analysis of different data sets over many different years we find no support for the general claim that direct democracy has an effect on internal or external efficacy. To the extent that effects are observable, the presence and usage of the initiative tends to widen the participatory resource gap of eligible voters, exerting a positive effect on the external efficacy of moderately informed voters, a positive effect on the internal efficacy of highly informed voters and a negative effect on the internal efficacy of low resource citizens, notably those with low levels of political awareness and non-voters. Theory and Critique Before providing a critique of arguments that have connected direct democracy and political efficacy, it is helpful to revisit the origins of the concept, which has its earliest roots in political science in the landmark work of Angus Campbell and his colleagues (Campbell et al. 1954; Campbell et al. 1960). Efficacy, as understood in these studies, referred to long term and relatively stable predispositions towards politics. Rather than being captured by measures of political interest and concern, efficacy referred instead to a latent expression of passivity in general responses to politics (Campbell et al. 1960, pp ). Gamson defines the concept more

4 clearly when he says, The efficacy dimension of political alienation refers to people s perception of their ability to influence. (1968, p. 42). It is now generally accepted that there are two dimensions to political efficacy: internal and external (Lane 1959). The internal dimension of efficacy refers to a person s view of his or her own capabilities in democratic politics whether or not the individual is sufficiently informed to participate, can make good decisions, etc. In contrast, the external dimension refers to a person s view of the responsiveness of government: whether or not government is actually listening. A number of studies have demonstrated empirically, using methodologically sophisticated dimension reduction techniques, that there are in fact multiple dimensions to a person s political alienation including political trust, internal efficacy and external efficacy (Balch 1974; Craig 1979; Craig and Maggiotto 1982; Craig et al. 1990; Niemi et al. 1991). In addition, the implications of these indicators can be quite different. 2 Scholarly interest in efficacy has most frequently been connected to an interest in political participation. The earliest studies showed that individuals who were the most efficacious were significantly more likely to vote (Campbell et al. 1960, p. 105). In a more fully specified model, Rosenstone and Hansen (1993) estimate that internal efficacy is marginally but positively related to turnout by a factor of about two percentage points, while external efficacy is more strongly related to turnout by a factor of about ten percentage points (pp ). Additionally, they find that changes in efficacy account for only about 9% of the change in voter turnout between 1960 and 1980, with mobilization effects exerting a much larger substantive effect (pp ). Research using panel data has also suggested that about a third of the relationship between efficacy and participation is endogenous, with those who participate in politics reporting higher levels of efficacy (Finkel 1985; Leighley 1991). 3 While it is commonly included in the literature on the secondary benefits of direct democracy, a link between processes like ballot initiatives and referendums and the external dimension of efficacy is also a direct test of the perceptive responsiveness of government. Several studies have advanced the concern about opinion-policy responsiveness as the most obvious way to test whether or not direct democracy fulfills its promise; suffice it to say, there is a considerable debate (Hagen et al. 2001; Lascher et al. 1996; Matsusaka 1995, 2001, 2004). As emphasized previously, some existing literature finds a positive connection between exposure to direct democratic institutions and both the internal (Bowler and Donovan 2002) and external (Bowler and Donovan 2002; Smith and Tolbert 2004) dimensions of political efficacy. In terms of democratic theory the authors of these studies have, if anything, undersold the apparent finding. Should exposure to ballot initiatives lead to increases in efficacy, this not only signifies that the health of participatory democracy is improved, but also that direct democracy helps create a greater perception of opinion-policy congruence. While such a finding would be of potential importance to evaluating democratic systems, we contend that the 2 For instance, efficacy but not trust is related to political participation (Rosenstone and Hansen 1993). 3 Clarke and Acock (1989) find that efficacy is much more sensitive to winning than to participating.

5 theoretical connections between direct democracy and efficacy are less clear than have been presented in the literature. Political Awareness, State Politics, and Top-of-the-Head Responses Existing empirical literature on the impact of state-based direct democracy on political efficacy is generally based upon cross-sectional survey data combined with information about the number of initiatives appearing on statewide ballots. Questions addressing internal efficacy include queries about whether government is too complicated to understand and whether respondents feel qualified to participate in politics. Questions addressing external efficacy include queries about whether respondents have any say in what government does and whether government cares about what average people think. While the recent literature on the positive secondary consequences of direct democracy makes explicit and helpful links to democratic theory, it provides minimal attention to relevant political psychology findings generally, or findings regarding political efficacy specifically. Given what is known about people s inattention to political constructs such as ideology, it is doubtful that many survey respondents have given much prior thought to political efficacy per se. Rather as Zaller (1992, p. 62) indicates in his influential work on mass public opinion, people answer survey questions on the basis of ideas that are most salient to them at the moment of response. Accordingly, when answering efficacy questions in the ANES and elsewhere people could be expected to act like cognitive misers and tap a few important considerations that are accessible in memory (see for example Krosnick 1988; see also Hetherington 2005, chap. 2). The recent direct democracy literature offers neither logic nor direct or indirect evidence to support the notion that the extent of direct democracy in a respondent s state would be included in such considerations. Perhaps scholars believe that survey respondents will recall the experience of voting directly on major policy issues. Yet it is at least as likely that respondents will draw upon such factors as perceived governmental performance or the extent to which politics is dominated by powerful lobbying groups (on the critical importance of perceptions of special interest dominance, see Hibbing and Theiss-Morse 1995, 2002). 4 Furthermore, the ANES political efficacy questions do not distinguish between state and national governments, asking simply about government. Yet it is reasonable to think that while direct democracy occurs only at the state and local levels in the United States, personal opinions about national government may dominate the formulation of answers to these political efficacy questions. In support of this contention, existing empirical work commonly finds that while state government is in 4 Boehmke (2002) finds that direct democracy tends to increase the size and diversity of interest group populations in states. While this might be a positive development from a pluralist perspective, surveys of citizens in states with direct democracy show that one of the biggest complaints about the system is that special/moneyed interests are too powerful (Dyck and Baldassare n.d.). Even though empirical research suggests the influence of money in these contests is not all that powerful (Banducci 1998), perception would matter more to political efficacy than reality.

6 theory closer to the people, knowledge of national government is higher because of much greater media attention (Delli Carpini and Keeter 1996; Songer 1984). State election outcomes are routinely influenced by national factors like presidential coattails (Campbell 1986). Furthermore, despite substantial evidence to suggest that state party systems adapt to the ideology of the electorate (Erikson et al. 1993), identities and opinions about national politics pervade (Green et al. 2002). In short, while people have low levels of information about politics generally, what knowledge they have is more likely to relate to national politics. This would suggest that in order to connect state outcomes (such as voting on ballot initiatives) to a measure of governmental performance such as external efficacy, voters would need relatively high levels of political awareness. The link between direct democracy and internal efficacy is perhaps more straightforward, and less hampered by concerns of political psychology. However, the directionality of the relationship is also more ambiguous. Perhaps the presence of policy issues on a ballot gives a boost to voters confidence in their own abilities. However, given that a common complaint regarding ballot initiatives is that they can be confusing and difficult to decipher, the low-information voter may feel less internally efficacious due to exposure to ballot measures. Particularly with respect to internal efficacy, we would expect differential effects of direct democracy to be contingent on a respondent s level of political awareness and status as a voter/nonvoter. Bowler and Donovan (2002) also consider the contingent impact of race (White vs. non-white respondents) on efficacy, which given the frequently racially charged initiatives which appeared on ballots after 1992, might find support in more recent data. 5 Our analyses consider this possibility. The Tenuous Link between Participation and Efficacy Even if we accept the questionable assumption that survey respondents would tap knowledge about their ability to vote on state level ballot measures in answering political efficacy queries, the question remains whether it is sensible to think that simply making more decisions at the ballot box gives people a stronger sense of political efficacy. The empirical research bearing most directly on political efficacy in the United States measures the impact of direct democracy by the average number of ballot measures on a statewide ballot over a particular time period (Bowler and Donovan 2002; Hero and Tolbert 2004; Smith and Tolbert 2004). The hypothesis examined in this line of research is that, holding all other factors constant, each increase in the average number leads to a greater sense of efficacy. Yet longitudinal studies of political efficacy suggest that, if anything, there is only a modest and contingent causal impact of political activity on efficacy, with political activity more likely to affect the external form than the internal (Finkel 1985). Based on an ANES panel study of pre-election and post-election efficacy, Clarke and Acock (1989, p. 561, emphasis in original) conclude: One of the best known 5 A recent study by Bowler et al. (2006) suggests that response to racially and ethnically charged ballot initiatives have moved Latino voters to the Democratic party. It stands to reason, then, that the potential for majority tyranny might affect efficacy levels of non-white voters.

7 hypotheses in democratic theory asserts that participation per se enhances political efficacy. Analyses indicate that neither voting nor campaign activity in the most recent American national elections had such an effect. It is notable that Clarke and Acock found that voting for victorious candidates had a positive effect on efficacy. While scholars have included voting for winning candidates as a control variable in previous models, none of the studies we have referenced have measured whether respondents feel more efficacious if the results of ballot initiatives turn out the way such respondents want. Instead, the extant empirical work on efficacy and direct democracy simply examines the connection between the raw number of ballot measures in a state and individual political efficacy. Moreover, some recent studies of public opinion and American political institutions suggest that people are more satisfied if they make fewer political decisions, provided they perceive that those making the decisions are not acting in their own self interest. Based on an extensive analysis of both national survey and focus group data, Hibbing and Theiss-Morse (2002 and 1995) conclude that Americans prefer a type of stealth democracy whereby democratic institutions and decision making are visible only when people perceive there to be a problem to which they need to attend. In support of this conclusion, Hibbing and Theiss-Morse provide evidence that, for example, people are more satisfied with institutions such as the U.S. Supreme Court and Federal Reserve Board that offer minimal opportunities for public input than with institutions such as Congress that offer more public accountability. Interestingly, Smith and Tolbert (2004, chap. 4) draw from Hibbing and Theiss-Morse in support of the notion that voters believe that the current political process is too dominated by elected officials. This implies that people are more satisfied with the political system if direct democracy is used more frequently. By contrast, we read the essence of the stealth democracy argument being that voters are more satisfied if decisions are made more often by objective, non-elected officials such as judges, provided such decision makers are not seen as furthering their own narrow self-interest. Indeed, the citizens more likely to favor direct institutions are also more likely to espouse anti-democratic views about compromise and support for increasingly anti-democratic decision-making by businesses and un-elected elites (Hibbing and Theiss-Morse 2002, p. 140). Support for Direct Institutions is Not Monolithic Even the relatively small literature pertaining to opinions about direct democracy as an institution suggests caution in believing that greater use of ballot initiatives lead to an enhanced sense of political efficacy among individuals. Through a variety of empirical studies it has been demonstrated that the existence of the ballot initiative is popular with voters (for a summary of several such studies see Bowler and Donovan 2002). Nevertheless, some of the same studies that indicate high overall satisfaction with direct democracy also find many concerned about the number of decisions they must make. Consider the following research regarding public opinion about direct democracy. One study finds that while Californians generally favor direct democracy, a significant minority (41%) also support limiting the number of initiatives that could

8 be voted on during a particular election, while a majority supported requiring a two thirds vote (rather than the current simple majority) for approval of a ballot initiative (Hagen and Lascher 1998). Another study finds that majorities think that the process is dominated by special interests and support curtailing usage, while principally favoring the concept of voting on citizen initiated laws (Dyck and Baldassare n.d.). Florida, in fact, recently passed a popular initiative that raised the minimum threshold necessary for passage of measures. 6 The current literature is perhaps best summarized by Craig et al. (2001) who indicate: in all likelihood, many Americans in general do not have well-developed attitudes about this particular topic. (p. 41) Aggregate Inconsistencies and Endogeneity Concerns Our final theoretic concern turns to aggregate trends. In Fig. 1, we present aggregated measures of external efficacy and frequency of ballot initiative usage from 1952 to While ballot initiative usage has been more frequent in the last two decades than in prior decades, there has been a slow downward trend in external political efficacy, with the exception of 2002 when political efficacy was temporarily higher after the September 11 terrorist attacks. The trend is generally inconsistent with the arguments about the positive effects of direct democracy on efficacy. Direct democracy usage peaked in the 1990s when external efficacy reached its lowest levels. This raises an immediate statistical red flag that if there is a link between usage of ballot initiatives and efficacy, the causality is potentially reversed. To be sure, the literature is quite clear on the fact that initiative entrepreneurs are atypical voters (Ellis 2002; Smith 1998). Therefore, we have strong reasons to doubt that the public starts to agitate more for initiatives when people view the government as unresponsive. However, agenda-setters are constrained by majority rule voting which requires approval of proposed policies by the median voter (Romer and Rosenthal 1978; Lupia 1992). To the extent that ballot measure agenda-setters are strategic actors and are able to perceive an environment ripe for policy change, we might expect direct democracy usage to be endogenous to levels of efficacy. In summary, the existing literature on efficacy and ballot initiatives, which is based on the theory of participatory democracy, gives short shrift to a series of other relevant concerns that cast doubt on the premise that a link exists between direct democratic institutions and political efficacy. Based on the concerns we have raised, a reasonable working hypothesis is to expect no relationship between the propensity to use direct democracy in a state (or number of ballot measures voted upon in a state) and the level of political efficacy among citizens residing in the state. In addition, if we observe effects anywhere, they should occur in the form of interactions among those who have actually participated in elections and have higher levels of political awareness. To be sure, the empirical evidence presented in 6 Florida s Amendment 3 passed in 2006 with 58% of the vote; Florida only has the constitutional initiative process. 7 The measure is a combination of two dichotomous questions about external efficacy from the ANES cumulative data file: public officials don t care much about what people like me think and people like me don t have any say about what government does.

9 Mean of 3-Point External Efficacy Mean Level of 3-Point External Political Efficacy Scale American National Election Studies, Year 100 Aggregate Yearly Initiative Usage for All States Number of Initiatives Year Fig. 1 Aggregate efficacy and initiative usage over time the studies by Smith and Tolbert (2004) and Bowler and Donovan (2002) purport to demonstrate an empirical link between direct democracy and efficacy. Our own analysis considers a critique, replication and extension of these models and finds that the general claims of the literature do not hold up to more careful inspection. Research Design, Data, and Methods Current studies connecting direct democracy to higher levels of internal and external efficacy have exclusively used the American National Election Study data archive, specifically the 1992 ANES (Bowler and Donovan 2002) and a cumulative study from 1988 to 1998 (Smith and Tolbert 2004). Some skepticism is warranted with regard to using these data, however, given that ANES respondents are not chosen randomly from a population, but rather are chosen using a multi-stage stratified cluster sample. The ANES randomly samples metropolitan areas, and then

10 completes random face to face interviews in those areas. The resultant sample is representative at the national level, but is not necessarily representative when moving to sub-samples of the population, given that many states are excluded and within state samples are limited to sometimes a single metropolitan area (Hill and Hurley 1984). Since the explanatory variable of interest in these studies is a measure of state level direct democratic context, it is possible that there are biases within state samples. It is also important to note that in any given survey, not all states are even represented in the sample (for the problems associated with this see Brace et al. 2002). In the 1992 ANES used in the Bowler and Donovan (2002) study, for instance, respondents are selected from 63 different primary sampling units across 33 states. One way to alleviate the single sample problem is to pool the surveys over time. While doing so generates greater state and PSU-level variation (Smith and Tolbert 2004), the potential for bias still remains. The operative word, however, is potential. A recent study by Brace et al. (2002) comparing pooled estimates of state populations from the General Social Survey to a series of aggregate population measures showed the pooled GSS surveys to be generally free of bias. Resolving the questions about direct democracy and efficacy within the literature, then, becomes somewhat problematic. On the one hand, the most thorough measures of internal and external efficacy exist in PSU-sampled data, but interpreting the results from these data requires a certain degree of skepticism, given that these data were not designed for sub-sampling state-level populations. In general, we contend that there are three operating principles from which to proceed: (1) it is important to examine multiple datasets and look for consistency of results, (2) when using PSU-sampled data, pooled surveys will eliminate more error than single surveys, and (3) reliability can be checked by corroborating findings with surveys that use alternative sampling techniques. We follow these principles by considering data from the American National Election Study, the General Social Survey, and the 1997 PEW Trust in Government Survey. Our modeling strategy uses the available data from each survey measuring political efficacy (internal and/or external). We regress the efficacy variables on indicators of direct democracy usage, measured by the frequency of initiative use as well as a dummy variable for the existence of the initiative. Some studies have measured direct democracy context based on whether or not a state has direct democratic institutions (Citrin 1996; Gerber 1999; Lascher et al. 1996; Matsusaka 1995, 2004). The usual reason for doing so is that many of the effects of direct democracy are thought to be indirect, resulting from the threat of initiatives. Recent research has argued, however, that initiative use should be treated as a continuous variable given the vast differences across states in initiative use and sharp divergence in state laws with respect to placing restrictions on the initiative process. Notably, studies inferring a positive relationship between direct democracy and political efficacy measured direct democracy as a continuous variable (Bowler and Donovan 2002; Hero and Tolbert 2004; Smith and Tolbert 2004, chap. 4). Rather than stipulating a priori that one approach is preferable, we use both continuous and dichotomous measures. The measure of initiative frequency is computed by taking the average number initiatives in the state from 1972 until the

11 survey year. For example, in pooled data, the 1988 data points are assigned an average measure from 1972 to 1988, while the 2004 data points are assigned an average measure from 1972 to The dichotomous measure is simply a dummy variable for whether or not the respondent s state had any initiatives during that period. It is noteworthy that some states (e.g., Washington, Maine, and Ohio) allow initiatives during odd-numbered years, so it is appropriate to consider every year in computing an average. In addition to measuring direct democracy usage, a series of relevant individual level control variables were included in previous models: age, gender, race, education, partisanship, retrospective evaluations of the economy and measures for a loser s consent in candidate elections. Previous studies also posited a number of other contextual explanations including state diversity, the presence of divided government, and political culture. While there are theoretic reasons for including the contextual variables, we note previous researchers have rarely found them to be statistically significant. All of our models include as many of these explanatory variables as are available in the data. We also posit three additional individual level explanations which were omitted in previous research. First, income is strangely absent from previous work as a socioeconomic control. In addition, the close and partially endogenous connection between voting and efficacy suggests that a measure of participation should be included in the model. Finally, given some interest in interactions (which we take up in a subsequent section), we also include political knowledge as a variable, hypothesizing that those with greater levels of political knowledge will be both more internally and externally efficacious. In the data analysis that follows, we offer 24 different model specifications spanning 10 datasets from three different survey organizations. In the process of preparing this manuscript, we have estimated no less than 100 iterations of models connecting direct democracy to political efficacy. Regardless of the specifications, and with very few caveats, the story told by the data is one of consistency. Across the board, in different surveys, in varied years, with different measures of ballot measure usage, the prevalence of direct democracy appears to be unrelated to both internal and external efficacy. ANES Data We begin with the ANES data used in previous studies. Since 1988, the American National Election Study has consistently asked two questions thought to tap external efficacy. Respondents are asked to agree, disagree, or neither agree nor disagree with the following statements: People like me don t have any say about what government does and I don t think public officials care much what people like me think. 8 External efficacy is the scale created by the additional of these variables. 8 Before 1988, the ANES only asked the question as a dichotomy, and so we use only the answers. The 1992 ANES uses a five point scale: strongly agree, somewhat agree, neither agree nor disagree, somewhat disagree, strongly disagree.

12 Using the 1992 ANES, Bowler and Donovan (2002) also made use of a rich set of questions about internal efficacy. 9 Unfortunately, internal efficacy questions have not appeared in any consistent form over time as a battery of questions, with the exception of the question: sometimes politics and government seem so complicated that a person like me can t really understand what s going on. This question is available biennially from 1988 to We proceed by replicating the 1992 study of Bowler and Donovan for internal and external efficacy, as well as the Smith and Tolbert (2004) model from 1988 to 1998 and then extending the analyses with available data. Coding specifics are presented in the appendix. For all multiyear models, we include year dummy variables, but suppress them for space and clarity of presentation. Table 1 includes our replication of previous works connecting direct democracy to external and internal efficacy, as well as extensions and alternative specifications of the model. With the exception of the three replication models, standard errors presented are robust, clustered at the state level. The first two columns in Table 1 present results for replications of the Bowler and Donovan (2002) and Smith and Tolbert (2004) external efficacy models. As near as possible the models are faithful replications of those presented in the aforementioned studies. 10 The replicated models do not support the earlier conclusions about the positive linkage between direct democracy and external efficacy. Coefficients for the initiative usage variable do not approach statistical significance. By extending the external efficacy analysis through 2004 and including income, participation and political awareness as additional control variables, the model fit improves significantly (see the third and fourth columns of Table 1). Yet in each case the standard errors are larger than the coefficients suggesting that, at any reasonable level of confidence, we cannot reject the null hypothesis. Simply put, a review of Table 1 fails to support the claim that direct democracy is related to external efficacy. With regard to internal efficacy, the prior finding that direct democracy is related to internal efficacy holds up using the 1992 data, but only for Whites. The coefficient for non-white voters is negative and significant, suggesting that direct democracy may in fact lower the amount of internal efficacy expressed by minority voters. Indeed, as we extend the analysis, the negative relationship between direct democracy and internal efficacy among non-white voters is far more robust than the positive relationship between direct democracy and internal efficacy among White voters. 11 The 6th, 7th and 8th columns of Table 1 suggest that 1992 may have been 9 See Appendix for details. 10 For the 1992 ANES data, we were unable to find the variable State Economy Worse used in the model after a very thorough search of the available data. The variable is not significantly related to either composite measures of internal or external efficacy used in Bowler and Donovan (2002). We specified numerous models in an attempt to replicate the finding from Smith and Tolbert (2004) using the cumulative ANES. We were unable to do so and are willing to share our data and logs of attempts at this replication upon request. Notably, our sample size is larger by over 1000 cases. 11 Given that the outcome is measured as a three category variable, we considered alternative specifications to OLS for Models 6 8 in Table 1. Using an ordered logit did not alter the substance of the findings in any meaningful way; these results are available from the authors upon request.

13 Table 1 Models of direct democracy s impact on political efficacy, American national election studies, Independent variables External efficacy models Internal efficacy models Bowler and Donovan Replication 92 Ordered logit Smith and Tolbert Replication OLS full model OLS full model 92, 96, 00, & 04 Bowler and Donovan Replication 92 OLS full model OLS full model 88, 92, 96 & 00 OLS full model 88, 96 &00 b SE b SE b SE b SE b SE b SE b SE b SE Direct democracy use Age Female African American Hispanic Asian Non-White DD use 9 Non-White Education Income Strong democrat Strong republican Pure independent State racial diversity California resident (0,1) Divided state government State political culture National economy worse Presidential loser Senate loser

14 Table 1 continued Independent variables External efficacy models Internal efficacy models Bowler and Donovan Replication 92 Ordered logit Smith and Tolbert Replication OLS full model OLS full model 92, 96, 00, & 04 Bowler and Donovan Replication 92 OLS full model OLS full model 88, 92, 96 & 00 OLS full model 88, 96 &00 b SE b SE b SE b SE b SE b SE b SE b SE Voted in election Political knowledge N R Bolded coefficients are significant at p \.05, 1 tailed test. Cell entries are unstandardized coefficients; standard errors are robust, clustered by state (exception is two replication models). Year dummies are included in all multi-year models, as is a constant, but these have been suppressed in results presentation

15 a somewhat exceptional year for the relationship between direct democracy and efficacy, further casting doubt on the findings presented in the literature. 12 Additional Data: GSS and PEW Following the principles set forth in the research design, we considered alternative data sources in an attempt to validate the results garnered from the ANES data. More specifically, we examined the 1996 General Social Survey (GSS) and the 1997 PEW Trust in Government Study. 13 The GSS data, while suffering from both the single year and pooled-cluster sample problems common to the ANES data, have an extensive set of questions about both internal and external efficacy that allow us to build comprehensive indices. This is an important comparison case as the 1992 ANES data had multiple five point Likert scale measures, while the over-time data set has more limited variation in the dependent variable. The 1996 GSS included six questions about political efficacy, well divided between the areas of internal and external efficacy. All these questions used a five point Likert scale and the questions are detailed in the appendix. We then summed responses across the different questions. 14 Results of our analysis from the GSS are presented in Table 2. We find no empirical evidence of a connection between use of direct democracy and the respondent s sense of political efficacy. Regardless of whether we measure initiative use as a continuous or dichotomous variable, none of the coefficients for the direct democracy variable approach statistical significance. Meantime, the control variables, especially education level, tend to have the expected significant relationship with the dependent variable. An important question to consider is how insignificant the findings are. When we use the preferred measure of direct democracy use proposed in the literature, none of the t-values are greater than 1. For internal efficacy, the coefficient is in fact negative, although highly insignificant ( t =.375). The results using GSS should be viewed with caution, as they suffer from the problems associated with using PSU-sampled data, particularly in the context of a single survey. However, they do provide a nice juxtaposition to the analysis of Bowler and Donovan (2002) who find relationships in a single-year PSU-sampled study between direct democracy and both internal and external efficacy. Any 12 Appendix Table 5 demonstrates that the findings from Table 1 do not change if we alter the specification of the independent variable of interest, Direct Democracy Use, to a dummy variable. 13 We also examined data from the 1985 GSS that included efficacy questions. We do not include this analysis in the present article because some contextual control variables are unavailable for that year. Including the 1985 GSS data would not change any of our findings: political efficacy is unrelated to direct democracy in that survey. 14 In the course of specifying our models, we ran separate models on each of the individual questions summed to the index. Absent any major differences in interpretation, we present the analysis of the indexed measures only.

16 Table 2 Direct democracy and political efficacy, general social survey, 1996 Independent variables Initiative average models Initiative dummy models External efficacy Internal efficacy External efficacy Internal efficacy b SE b SE b SE b SE Direct democracy use State racial diversity Divided state government State political culture Age Female African American Education Income Strong democrat Strong republican Pure independent Constant N R Bolded coefficients are significant at p \.05, 1 tailed test. Beta values are from OLS regression; standard errors are robust, clustered by state criticisms of our analysis based on use of PSU sampled data would be equally applicable to the earlier, positive findings. Finally, we turn to the 1997 PEW Trust in Government Study. This data set has the virtue of allowing us to follow the third guiding principle of our research design: to corroborate findings from PSU-sampled data with data from a random sample as an additional reliability check. 15 The sample size of the survey is also rather large, leaving over 1700 cases available for analysis after missing data are excluded. The PEW study had a single question on external efficacy, measured on a fourpoint Likert scale which asked respondents to either agree, mostly agree, mostly disagree or disagree with the statement: Public officials don t care what people like me think. Models presented in Table 3 were estimated using ordered logit with robust standard errors clustered on state. 16 Once again, direct democracy usage appears unrelated to external efficacy. If anything, Model 2 in Table 3 suggests that increasing usage of direct democracy is negatively related to external efficacy! Looking at the control variables, we find again that higher efficacy exists among partisans, the educated, non-black respondents, and those who evaluate current incumbents and the economy/country positively. 15 The PEW data were collected using a random-digit-dial telephone sample. 16 A likelihood-ratio test determined that the proportional odds assumption was not violated in any of the models presented.

17 Table 3 Direct democracy and external efficacy, PEW trust in government study, 1997 Independent variables Initiative average models Initiative dummy models Model 1 Model 2 Model 3 Model 4 b SE b SE b SE b SE Direct democracy use State racial diversity Divided state government State political culture Age Female African American Education Income Democrat Republican Present national situation Clinton favorability Regular voter Political information Cut Cut Cut Prob [ wald chi-squared N Bolded coefficients are significant at p \.05, 1 tailed test. Beta values are from ordered logit maximum likelihood estimation; standard errors are robust, clustered by state Probing for Possible Interactions Someone questioning our null findings might note that when examining the different specifications presented in Tables 1, 2 and 3, the size and direction of the coefficients on direct democracy use occasionally change. While this is potentially the result of random error, it is also possible that the direct democracy measure is interacting with some of the variables in the model. In the theoretic section, we suggested two possible interactive scenarios for both internal and external efficacy. The first is voting: in order for citizens to feel externally efficacious because of direct democracy, they presumably need to have participated. The mere existence of ballot measures is insufficient to transfer stronger feelings of efficacy. The second interactive scenario is that being better informed allows one to connect voting on ballot measures with government responsiveness and hence leads to higher levels of efficacy. Note that the expectation for external efficacy is that the usage of direct legislation should simply have no effect on those who are unable to connect the dots: low-information non-voters. With regards to internal efficacy, we

18 expect those with low information to be negatively affected by the frequent usage of ballot initiatives, as the task of voting may seem more daunting to those with fewer political resources. Accordingly, we probed for such possible interaction effects using the ANES cumulative data. Table 4 contains the key results. The Table 4 results do not support broad claims about interaction effects in the area of external efficacy. Findings from Model 1 (first column) indicate that while voters generally have a greater sense of external efficacy than non-voters, voters with more exposure to direct democracy feel no more efficacious than those with less exposure. Similarly, results from Model 2 indicate that while better informed respondents are more likely to feel externally efficacious, there is no difference in the effect of direct democracy usage on efficacy by information level. Note that an interaction effect does appear if the sample is limited to voters, as in Model 3. In this case we find that among uninformed voters there is a significant positive relationship between direct democracy usage and external efficacy. This effect, however, is largely mitigated by information as indicated by the negative sign on the interaction term in Model 3. Model 4 clarifies the result from Model 3 demonstrating there is in fact a non-monotonic information effect with voters at the middle levels of awareness becoming more externally efficacious due to exposure to direct democracy. For internal efficacy, there is stronger confirmation of our hypotheses. Nonvoters and the politically uninformed feel less internally efficacious as direct democracy usage increases, while voters and those with higher levels of political awareness feel more internally efficacious as direct democracy usage increases (Model 6). Taken with the results from Table 1 regarding the race interaction, it appears as though lower resource voters sense of their own political abilities is negatively affected by the presence of direct legislation. At the minimum, the regression results using interaction terms suggest that the relationship between efficacy and direct democracy is much more complicated than suggested by earlier research. Direct democracy appears to increase efficacy for some, decrease it for others, and have no effect on others and the patterns across categories and models vary. Generally, we find that if positive effects of direct democracy on external efficacy exist, they exist only amongst moderately informed voters. Furthermore, the usage of the direct initiative also has the potential to harm the internal efficacy of low resource voters: those with low levels of political awareness, non-voters, and non-white respondents. We also note that none of the findings from models using interaction variables undermine the conclusions reached from results of general models presented in Tables 1, 2 and 3. Overall, the weight of the evidence indicates that direct democracy does not have uniform effects on the internal or external efficacy of Americans. Discussion We began our analysis with concerns about both the logic and empirical methods supporting the conclusion that direct democracy has a positive impact on political