27
Disentangling the Direct and Indirect Effects of the Initiative Process John G. Matsusaka University of Southern California A large literature documents that initiative states choose different policies than noninitiative states, but little is known about why the policies are different. Do initiatives change policy directly by overriding representatives, or indirectly by providing a threat that alters legislative behavior? This paper develops an empirical strategy to quantify the direct and indirect effects by comparing policy choices in states that actually pass initiatives with policies in states where the initiative is available but none are approved by voters. Evidence from all 50 states on nine separate issues suggests that both direct and indirect effects are important, but the direct effect is several times as large as the indirect effect. June 2007 Comments welcome. The author is a professor in the Department of Political Science, Gould School of Law, and Marshall School of Business, all at USC. Please contact the author at USC Marshall School of Business, Los Angeles, CA 90089-1427, [email protected] . I thank USC for financial support.

Disentangling the Direct and Indirect Effects of the Initiative

  • Upload
    others

  • View
    3

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Disentangling the Direct and Indirect Effects of the Initiative

Disentangling the Direct and Indirect Effects of the Initiative Process∗

John G. Matsusaka

University of Southern California

A large literature documents that initiative states choose different policies than

noninitiative states, but little is known about why the policies are different. Do initiatives

change policy directly by overriding representatives, or indirectly by providing a threat

that alters legislative behavior? This paper develops an empirical strategy to quantify the

direct and indirect effects by comparing policy choices in states that actually pass

initiatives with policies in states where the initiative is available but none are approved by

voters. Evidence from all 50 states on nine separate issues suggests that both direct and

indirect effects are important, but the direct effect is several times as large as the indirect

effect.

June 2007

∗ Comments welcome. The author is a professor in the Department of Political Science, Gould School of

Law, and Marshall School of Business, all at USC. Please contact the author at USC Marshall School of

Business, Los Angeles, CA 90089-1427, [email protected]. I thank USC for financial support.

Page 2: Disentangling the Direct and Indirect Effects of the Initiative

1

1. Introduction

More than a decade of rigorous empirical research has shown that states where

citizens can bypass their elected representatives and pass laws directly through the

initiative process choose different policies than noninitiative states (Lupia and

Matsusaka, 2004; Matsusaka, 2005).1 Moreover, recent research suggests that the policies

are different in a particular way: initiative states tend to choose policies that are more

congruent with public opinion than noninitiative states.2 Yet exactly how these policy

changes come about is something of a mystery. In theory, the initiative process can

influence policy directly, by creation of new laws, and also indirectly, by changing the

legislature’s behavior. An indirect effect can appear if the legislature chooses different

policies when initiatives are available in order to deter the threat of being overridden

(Gerber, 1996; Matsusaka and McCarty, 2001) or in response to new information that is

revealed by election returns, even from an unsuccessful measure (Gerber, 1999;

Matsusaka, 2004, Ch. 9). There is some anecdotal evidence of indirect effects (e.g. Key

and Crouch, 1938; Gerber, 1998), and casual observation of state politics suggests that

direct effects are important, but there is no quantitative or statistical evidence on the

relative importance of the different channels of influence.

This purpose of this paper is provide what I believe is the first quantitative

evidence that attempts to disentangle the direct and indirect effects of the initiative. The

standard research design in this literature is to compare policies in initiative and

noninitiative states, typically using state-level cross-sectional or panel data and regressing

policy outcomes on a dummy variable for initiative availability.3 The coefficient on the

1 Following standard (but not universal) practice, I define the “initiative” to be a form of direct democracy

in which a new law can be proposed by citizens and placed on the ballot by petition. Other forms of direct

democracy are the “referendum,” a proposal to repeal an existing law, also placed on the ballot by petition,

and the “legislative” or “referred” measure, a new law placed on the ballot by the legislature. 2 See Tolbert and Smith (2006) and Matsusaka (2004, 2007a) for evidence and reviews of the literature. 3 The literature that uses the dummy variable is now quite large. Recent papers that can be used as

beachheads into the literature are Persily and Anderson (2005) and Matsusaka (2006) on election law,

Matsusaka (forthcoming) on the executive branch, Gerber and Hug (2002) on minority rights, Matsusaka

(2007c) on public employment, Matsusaka (2004) and Tolbert (1998) on fiscal policy and fiscal

Page 3: Disentangling the Direct and Indirect Effects of the Initiative

2

initiative dummy reveals whether or not the initiative matters, but it does not separate the

initiative effect into direct and indirect channels. The methodological innovation of this

study is to decompose the initiative dummy into pieces that represent direct and indirect

effects so the separate channels of influence can be isolated. The direct effect of the

initiative on a particular issue can be captured by a dummy variable equal to one for

states in which an initiative on the issue appeared on the ballot and was approved, while

the indirect effect can be captured by a dummy variable equal to one for states in which

the initiative process was available but a measure did not appear on the ballot or was not

approved.

I employ this methodology in a sample that contains information for all 50 states

on nine different issues. The estimates suggest that both direct and indirect effects matter,

but the direct effect is about five times as large as the indirect effect. Initiative states are

about 20 percent more likely than noninitiative states to choose a policy congruent with

public opinion, but the difference is about 61 percent when initiatives are actually

approved (direct effect) compared to about 11 percent when the initiative is available but

not used (indirect effect). Focusing on the ideological orientation of policy choices rather

than congruence, I find that initiative and noninitiative states are about 17 more likely

than noninitiative states to choose a conservative policy, but the difference is 38 percent

when an initiative actually appears on the ballot compared to 13 percent when the

initiative remains purely a threat.

The paper is intended to advance the literature in several ways. Methodologically,

it offers a strategy to disentangle the different effects of the initiative that may be useful

for analyzing the workings of other institutions that can display both direct and indirect

effects, such as the executive veto (McCarty, 2000). In terms of theory, the findings

provide support for game theoretic models of the initiative process such as Gerber (1996)

and Matsusaka and McCarty (2001) that predict the initiative influences policy indirectly

by providing a threat. Similarly, the evidence lends support to a central message of these

models that it is important to consider the strategic responses of political actors to

institutional opportunities. Finally, the evidence highlights the importance for empirical

institutions, and Matsusaka (2007b) on social issues. Bowler and Donovan (2004) contains a general

discussion and raises some issues.

Page 4: Disentangling the Direct and Indirect Effects of the Initiative

3

research of considering both direct and indirect effects of the initiative. Gerber (1998, p.

192) argues, “Studies that focus solely on direct influence [ballot propositions that are

actually passed] are likely to seriously underestimate the influence of groups that use

initiatives to achieve indirect influence.” The findings reinforce Gerber’s claim, and by

quantifying the direct and indirect effects, suggests that the amount of underestimation

can be large if the indirect effects are ignored.

2. Channels of Influence

This section reviews existing theories on direct and indirect effects, and discusses

approaches to testing them.

Direct Effect: Override

The direct channel is the best known way the initiative can influence policy. When the

Progressives agitated for the initiative and referendum at the turn of the nineteenth

century, they emphasized how direct democracy would allow the people to break the grip

of special interests on the legislature:

“The opponents of the referendum and initiative, therefore, would do well

to remember that the movement in favor of the two is largely due to the

failure of the representative bodies really to represent the people. There

has been a growing feeling that there should be more direct popular action

as an alternative, not to the action of an ideal legislative body, but to the

actions of legislative bodies as they are now too often found in very fact to

act.” (Theodore Roosevelt, 1912, p. 62)

“If we felt that we had genuine representative government in our state

legislatures no one would propose the initiative and referendum in

America. They are being proposed now as a means of bringing our

representatives back to the consciousness that what they are bound in duty

and in mere policy to do is to represent the sovereign people who they

profess to serve and not the private interests which creep into their

Page 5: Disentangling the Direct and Indirect Effects of the Initiative

4

counsels by way of machine orders and committee conferences.”

(Woodrow Wilson, 1912, pp. 87-88)

The theoretical rationale for this channel of influence is thus grounded in the idea

that legislators are not perfectly representative. Several formal models that view voters

and representatives as being involved in a principal-agent relationship, notably Barro

(1973) and Ferejohn (1986), show how representation can break down: elections give

officials an incentive to pursue constituent interests, but elections are crude enough tools

to allow some slack for representatives to choose policies that they favor for personal

reasons or to help key supporters. Representation failures create an opportunity for

initiatives to improve outcomes. There are also models that show how even legislators

who faithfully pursue their constituent interests might end up choosing collectively

inefficient outcomes if they are elected from geographically based constituencies and

engage in logrolling (Weingast et al., 1981). Initiatives provide voters a direct mechanism

to repair breakdowns in the representation process that occur from logrolling.4

Testing for direct effects is not as simple as it might seem at first. The fact that an

initiative appears on the ballot and is approved does not necessarily mean that the policy

is different than it otherwise would have been. The initiative may be no more than

window dressing if the legislature would have approved the policy anyway. For example,

while initiatives were used to ban gay marriage in 11 states (through 2006), legislatures

placed similar measures on the ballot in 16 other states, raising the possibility that the end

result would have been the same with or without the initiatives. In addition, as Gerber et

al. (2001) show, even when voters approve an initiative, it may never go into effect

because of court challenges or lack of enforcement. Thus, it is ultimately an empirical

4 The initiative can also improve the efficiency of elections by reducing the number of issues controlled by

elected officials (Matsusaka, forthcoming). The empirical study of legislator responsiveness to constituency

interest goes back at least to Miller and Stokes (1963). Well known studies finding that legislators

sometimes do not pursue constituent interests include Kau and Rubin (1979), Kalt and Zupan (1984), and

Peltzman (1984). For evidence that logrolling leads to collectively inefficient policies, see Gilligan and

Matsusaka (1995, 2001), DelRossi and Inman (1999), and Chen and Malhotra (2007).

Page 6: Disentangling the Direct and Indirect Effects of the Initiative

5

question whether states that pass an initiative on a particular issue actually end up with

different policies than states that do not pass an initiative on that issue.

Indirect Effect: Threat

The indirect threat effect of the initiative is a central theme of game theoretic models

(Gerber, 1996; Matsusaka and McCarty, 2001).5 In these models, the legislature has a

policy preference that may not coincide with the median voter’s preference. If an interest

group emerges that wants a change in policy, a forward-looking legislature may

accommodate the group by moving the policy closer to what the group wants in order to

deter the group from paying the cost to place a measure on the ballot. As a result, the

policy outcome may be different due the availability of the initiative even though a

proposition does not appear on the ballot. With complete information, the policy changes

brought about by the threat can only help the median voter because only a threat to move

toward the median voter is credible enough to engender a legislative response. However,

with incomplete information about voter preferences or policy consequences, legislatures

may move away from the median voter by accommodating even an extreme interest

group (Gerber and Lupia, 1995; Matsusaka and McCarty, 2001). Thus, these formal

models make it clear that the changes brought about by the initiative may not necessarily

help the majority of voters, even if all actors are rational and forward-looking.

Gerber (1998) discusses specific cases where initiative threats appeared to prompt

policy changes by the legislature, and other isolated examples can be found, but the

relevance of the threat effect has not been quantified or demonstrated rigorously.

Evidence on the threat effect is particularly important because game theoretic models

suggest that the threat effect is the primary way the initiative would change policy.

Indeed, in models with complete information, the initiative influences policy only by

providing a threat – initiatives never actually reach the ballot because a forward-looking

legislature always accommodates groups with credible initiative threats so as to make

them unwilling to pay the costs of a campaign. I will test for indirect threat effects by

studying whether states that have the initiative process but do not pass a measure on a

particular issue end up with different policies than states without the initiative process. 5 All of these models can be traced back to the agenda-setter framework of Romer and Rosenthal (1979).

Page 7: Disentangling the Direct and Indirect Effects of the Initiative

6

Indirect Effect: Signaling

Another channel of influence, less prominent than the override and threat channels, is

what I will call “signaling,” following Gerber (1998). The signaling channel is premised

on the idea that representatives want to follow public opinion, at least to some degree, but

may lack information about the preferences of citizens and the consequences of policies.

In such a context, election returns on a ballot proposition can convey useful information

to representatives that may lead them to change their policy decisions. The idea that

representatives may make “honest mistakes” that can be corrected by ballot propositions

is explored in Gerber (1998), Lupia (2001), and Matsusaka (1992, 2004). Signaling is

presumably the underlying rationale for the practice in many cities and some states of

holding nonbinding advisory votes on specific issues.

Many observers have suggested that signaling effects may be important. Key and

Crouch (1938, p. 457) observed that: “the initiative occasionally plays a role of varying

importance in bringing about new legislation through the ordinary procedures of

lawmaking. The campaign in support of the initiative may demonstrate to the legislature

that, with certain alteration, the program would be in accord with public opinion; it may

bring to the public attention abuses requiring correction; it may bring opposing groups to

recognition of the futility of demanding enactment of their unaltered ideas, thereby

facilitating compromise.” In support of the idea that legislators respond to information

about changing voter preferences, Kousser et al. (forthcoming) show that California

legislators began to vote more conservatively following the recall of Democratic

governor Gray Davis in 2003. Reinforcing this conclusion, in a survey of initiative

proponents, Gerber (1999, Ch. 5) finds that the primary reason businesses and other

groups put measures on the ballot is not to secure their passage, but to signal support to

the legislature.6 To assess the importance of signaling, I will test if states with ballot

propositions that do not pass (that is, can only be viewed as signaling information)

choose different policies than states without the initiative process.

6 A related, but somewhat different channel is suggested by the findings of Smith and Tolbert (2004) that

the presence of initiatives on the ballot changes what citizens know about politics, how involved they are in

politics, and the number of organized interest groups. Educated citizens would presumably choose elect

different representatives, resulting in different policy outcomes.

Page 8: Disentangling the Direct and Indirect Effects of the Initiative

7

3. Empirical Strategy

Methods

The empirical strategy is to analyze a set of 9,...,1=n policies in 50,...,1=s states using a

logistic regression framework of the form

(1) ,)0Pr()1Pr(ln nsns

ns

ns uayy

+++=⎟⎟⎠

⎞⎜⎜⎝

⎛== cXbIns

where y is a dichotomous policy outcome, I is vector-valued variable capturing initiative

channels of influence, X is a vector of control variables, u is an error term, and a, b, and c

are parameters to be estimated. In much of the previous literature, I is specified as a

dummy variable equal to 1 if a state allows the initiative and 0 otherwise. With such a

specification, the coefficient on the dummy variable absorbs all of the different initiative

effects, and provides a good summary indicator of the overall effect of the initiative

(Matsusaka, 1995).

In order to disentangle the different channels of influence, I introduce multiple

dummy variables. The simplest specification distinguishes direct (override) from indirect

(threat + signal) effects by using two dummy variables:

(2) DIRECTns

AVAILABLEs IbIb 10 +=nsbI ,

where

1=AVAILABLEsI if the initiative was available in state s, and 0 otherwise;

1=DIRECTnsI if an initiatives on issue n was approved in state s, and 0 otherwise.

Since initiative availability is captured with AVAILABLEsI , the direct effect of the

initiative is given by the coefficient 1b . The indirect effect is given by 0b . The full effect

of the initiative would be 10 bb + in a state where voters approved a proposition, and 0b

Page 9: Disentangling the Direct and Indirect Effects of the Initiative

8

in a state where voters did not approve an initiative. The questions of interest are whether

0b and 1b are different from zero, that is, whether each of the effects is important, and

also how 0b and 1b compare to each other.

The indirect channel can be further disentangled between signaling and threat

effects by introducing a third dummy variable:

(3) SIGNALns

DIRECTns

AVAILABLEs IbIbIb 210 ++=nsbI ,

where

1=SIGNALnsI if an initiative on issue n appeared on the ballot in state s, and 0 otherwise.

In this case, the direct effect continues to be captured by 1b . The indirect effect from

communication is captured by 2b and the indirect threat effect is captured by 1b .7 The

full effect of the initiative in a state where voters approved a measure would be

210 bbb ++ in a state where voters approved a measure because all three channels of

influence would be at work.

Data

The analysis is built around the analysis of nine different policies for which state-by-state

opinion data are available in the American National Election Studies (ANES) during the

period 1988-2004. Each of the policies was treated by the ANES as having a

dichotomous outcome (for example, allow or prohibit the death penalty), which means

that there is a clear majority position on each issue (which would not be the case if the

policy had multiple outcomes). The nine issues and some summary information are

7 In principle, the measurement of the communication effect could be refined by allowing it to vary by vote

totals. For example, a one-sided vote in an election with high turnout might convey more information than

a 50-50 vote in a low-turnout election. Unfortunately, there are too few state-issues in my sample to

squeeze such information from the data.

Page 10: Disentangling the Direct and Indirect Effects of the Initiative

9

reported in Table 1. Information on initiative institutions is taken from Matsusaka (2004,

Appendix 1), and demographic information is taken from the Census Bureau.8

I approach the central model (1) in two ways. The first is to treat the dependent

variable as a measure of the congruence between policy and public interest. For issue n

and state s, “congruence” is defined as

(4) ⎩⎨⎧= otherwise.0

; issueon majority by the preferred outcome thechooses state if1 nsyns

Thus, 1=nsy means that state s has adopted the majority position (which is also the

median position) on issue n, while 0=nsy means the state has adopted the minority’s

choice, which is not the median position. The various dummy variables in (2) and (3)

then reveal how the different channels of influence affect the congruence or

responsiveness of policy to opinion.

The other approach to model (1) is to treat the dependent variable as a policy

outcome. In order to combine all nine policies in a single metric, the outcomes have to be

expressed in a common metric. Accordingly, for the second set of estimates I define nsy

in terms of the ideological orientation of the outcome:

(5) ⎩⎨⎧= otherwise.0

; issueon outcome veconservati thechooses state if1 nsyns

Conventional classifications were used to assign each outcome as conservative or

liberal. That is, “conservative” outcomes were defined as opposition to “partial-birth”

abortion, opposition to public funding of abortion, opposition to anti-discrimination laws

based on sexual orientation, opposition to same-sex marriage, opposition to the estate tax,

support for parental notification for abortion, support for the death penalty, support for

English-only laws, and support for term limits. These assignments are not ironclad – 8 More detailed information on the data is contained in the appendix to Matsusaka (2007a). Following

standard practice, I classify Illinois as a noninitiative state because courts have held that its initiative

process can only be used to modify one narrow section of the state constitution.

Page 11: Disentangling the Direct and Indirect Effects of the Initiative

10

some conservatives may favor anti-discrimination laws and some liberals may favor term

limits, for example.

The empirical strategy requires distinguishing states that use their initiative

process and states that do not. I collected this information by searching the Initiative and

Referendum Institute’s Initiatives Historical Database (version 2007-2) that lists and

describes every statewide initiative that has been decided since the first one in Oregon in

1904. I extracted every initiative related to one of the nine issues identified from the

ANES. For each state and each issue, I determined whether or not an initiative had

appeared on the ballot, and whether or not an initiative had passed. Table 1 shows the

number of states that had initiatives on each subject and how many passed.9 As can be

seen, the most common issue was term limits, which appeared on the ballot in 22 states.

By way of comparison, 23 states allow the initiative. Same-sex marriage initiatives were

also common, appearing in 11 states. Initiatives concerning partial-birth abortion were

the rarest, appearing in only two states. The approval rates varied by issue, with 91

percent of same-sex marriage initiatives and 86 percent of term limits initiatives passing

at the high end, compared to no partial-birth abortion laws passing at the low end. For the

nine issues overall, 14 percent of states had a related initiative on the ballot at some point,

and 10 percent of states approved a measure. Only 4 percent of states had a measure on

the ballot but did not approve one. Because there are so few of these states that are used

to identify the signaling effect, the estimates for the signaling channel turn out to be

imprecise.

4. Empirical Results

Congruence

Table 2 summarizes the percent of outcomes that are congruent across all states

and issues, distinguishing according to whether the initiative is available or not, and

whether it is used. When the initiative is not available, only 51.9 percent of state-issues

9 I included initiatives requiring parental notification in the “parental consent” category. The results are

quite similar if they are excluded instead. The “term limits” category does not include initiatives that

allowed representatives to take a non-binding term limits pledge.

Page 12: Disentangling the Direct and Indirect Effects of the Initiative

11

reflect the majority’s preference. Since choosing a policy by flipping a coin would yield

50 percent congruence, it seems that public opinion is not very important for policy

choice in noninitiative states on these issues. In states where the initiative is available but

has at most an indirect influence, congruence is 60.1 percent, 8.2 percent higher than

states where the initiative is not available. In states where the initiative operates directly,

that is, where a ballot proposition is actually approved by the voters, congruence is 78.7

percent. Table 2 shows that initiative is associated with congruence, and suggests that

both indirect and direct effects matter, with direct effects the most important.

The next step is to put this conclusion on a sounder footing by controlling for

other factors that influence congruence using model (1). Each column of Table 3 reports

logistic regressions of model (1) in which the dependent variable is the likelihood of a

congruent outcome (as defined in (4)). The regressions include several control variables

that for the most part are standard in the literature, as well as nine issue-specific dummy

variables (and no intercept) whose coefficients are not reported.

Column (A) reports a benchmark regression where the dummy variable

distinguishes between initiative and noninitiative states, without separating direct and

indirect effects. The positive and significant coefficient of 0.85 on the initiative dummy

indicates that policies are more likely to be congruent with majority opinion in initiative

than noninitiative states. When the control variables take on their mean values, the

estimates imply that policies are 5.7 to 21.2 percent more likely to be congruent in

initiative than noninitiative states, depending on the issue, with a difference of 19.9

percent for the median issue (partial-birth abortion ban).10 Consistent with other evidence

(Matsusaka, 2004, 2007a, 2007b), the initiative appears to make policy significantly more

responsive to public opinion.

Column (B) reports a regression that separates direct and indirect effects using the

approach in equation (2), one dummy variable for initiative availability and one dummy

variable for states where voters actually approved an initiative on the issue in question.

10 These (marginal) effects of the initiative and others like them throughout the paper are calculated with all

explanatory variables set at their mean values. Because the models include separate intercepts for each

issue, the effect of the initiative depends on what the issue is. The ranges reported here and below are

across the nine issues, and the median is the median effect across the nine issues.

Page 13: Disentangling the Direct and Indirect Effects of the Initiative

12

Both initiative coefficients are positive and different from zero at the 10 percent level of

statistical significance or better, suggesting that both direct and indirect effects are

important. The marginal effects appear to be quantitatively important as well. With

control variables set at their mean levels, the indirect effect is associated with 1.7 to 12.5

percent greater congruence, depending on the issue, with a value of 11.1 for the median

issue (partial-birth abortion ban). Because the dummy variable for initiative availability

captures the indirect effect, the coefficient on the direct effect captures the additional

effect of having a measure approved by the voters. The direct effect is quantitatively

important, associated with 9.9 to 56.5 percent greater congruence compared to an

initiative state with only an indirect effect, and the direct effect is 51.7 percent for the

median issue (abortion parental consent). In a state where both direct and indirect effects

operate, the full effect of the initiative is to increase the probability of congruence by 9.2

to 69.0 percent, with a median of 61.2 percent. The estimates suggest that the direct effect

is greater than the indirect effect, and the hypothesis that the coefficients are the same can

be rejected at the 1 percent level ( 011.=p ).

A potentially confounding issue in estimating model (1) is the endogeneity of the

number of initiatives appearing on the ballot and approved by voters. In particular, we

might expect to see more initiatives when legislators fail to respond to public opinion,

that is, when congruence is low. Such an effect would create a negative relation between

the number of initiatives and congruence, biasing down the initiative coefficient. To

address this issue, I employ a two-stage framework in which I first generate predicted

values for the probability that a state adopts an initiative on a given issue, and then use

the predicted values in the second stage estimate of model (1). The first stage model is

(6) ,adoptedYear required Signature)0Pr()1Pr(ln nsssDIRECT

ns

DIRECTns udca

II

+⋅+⋅++=⎟⎟⎠

⎞⎜⎜⎝

⎛==

nsbX

where X includes the standard control variables. Two new variables are introduced in the

first stage model (6) to predict whether or not an initiative was adopted. Signatures

required to qualify an initiative for the ballot captures the cost of making a proposal. The

year the state adopted the initiative process indicates how much time the state has had to

Page 14: Disentangling the Direct and Indirect Effects of the Initiative

13

adopt an initiative. Values for both are taken from the appendix of Matsusaka (2004). The

predicted values of DIRECTnsI are then included in model (1) in place of the actual value.

The results of the second stage regression are reported in column (C) of Table 3.

The most noticeable difference is that the coefficient on the direct effect increases while

the coefficient on the indirect effect decreases. This lends some support to the idea that

the direct initiative dummy is biased downward in column (B), and further emphasizes

the importance of the direct channel of influence. Indeed, while the coefficient on the

indirect initiative dummy remains positive, it declines in magnitude, and cannot be

distinguished from zero at conventional levels of statistical significance. This

specification raises the possibility that there may not be an important indirect effect.

Column (D) reports a regression that allows for a separate indirect signaling effect

by including a dummy variable for states that had an initiative on the ballot. Because

initiative availability and initiative approval are controlled by other dummy variables, the

coefficient on the INDIRECT/SIGNAL dummy indicates the effect of having an

initiative on the ballot that does not pass. Column (E) reports the same regression

estimated using the two-stage approach for both DIRECT and INDIRECT/SIGNAL

variables. The estimates in columns (D) and (E) provide little evidence of an indirect

signaling channel. The coefficient on the signaling dummy variable is never statistically

significant and changes sign from (D) to (E). Because there are so few observations that

fit into the signaling category (only 4 percent of the total), it is hard to tell if the weak

results are due to a lack of data or because there simply is no signaling effect. However,

the fact that there are so few cases in this category itself suggests that this channel of

influence is not very important for the issues studied here. The indirect threat effect,

given by the coefficient on the dummy variable on initiative availability, remains

positive, but the magnitude is modest and cannot be distinguished from zero at the 10

percent level.

Turning to the control variables, the most important is the size of the majority. As

the majority becomes larger, a congruent outcome is more likely. The majority apparently

gets its way when it becomes large enough. The state’s population (as a logarithm) is

negatively related to congruence. It may be more difficult to monitor elected officials in

large states than small states because of greater free rider problems, and because

Page 15: Disentangling the Direct and Indirect Effects of the Initiative

14

politicians might be find it more difficult in large states to determine public preferences

due to the greater distance between representatives and constituents. The fraction of the

state’s adult population with a high school degree is negatively related to congruence, but

never different from zero at conventional levels of significance. We might have expected

that educated voters would be more able to control policymaking in their state, but this

does not appear to be the case. Southern states and older states are significantly more

likely to choose congruent outcomes. These variables may be capturing an otherwise

unmeasured regional feature of political culture or institutions. The last control variable is

a dummy equal to one if a state’s judges must stand for re-election instead of being

appointed for life or reappointed by the governor, legislature, or a commission. As shown

in Hanssen (1999) and Besley and Payne (2006), elected judges tend to make different

decisions than appointed judges, and Matsusaka (2007a) shows that judicial retention

procedures are important for congruence.11

Conservative versus Liberal Outcomes

The preceding section explores how the initiative affects the congruence between

policy and public opinion. This section explores how the initiative affects the ideological

direction of policy choices. Since the policy is dichotomous, this boils down to

explaining whether a “conservative” or “liberal” law is chosen. Table 4 reports

coefficients from logistic regressions in which the dependent variable is the likelihood of

a state choosing a conservative law (model (1) with dependent variable (5)). The control

variables are the same as in Table 3, except that public opinion is measured as the percent

of the population favoring the conservative outcome.

As before, the first regression provides a benchmark for the average initiative

effect, without distinguishing channels of influence. The point estimate of 1.23 is positive

11 I also estimated the regressions with other control variables, including income, racial composition,

partisan composition of the legislature, and legislative professionalism, none of which alter the main

conclusions. Because the initiative process is more common among Western states, one concern is that the

initiative variables are actually capturing an otherwise unmeasured Western effect. To check for this, I re-

estimated all of the regressions in the paper with the Western states deleted. The findings were not

materially different.

Page 16: Disentangling the Direct and Indirect Effects of the Initiative

15

and different from zero at better than the 1 percent level, indicating that initiative states

choose more conservative laws than noninitiative states on these nine issues. When

control variables are set at their mean values, the estimates imply that the initiative

increases the probability of a conservative outcome by 4.3 to 30.7 percent depending on

the issue, with an effect of 17.4 percent for the median issue (abortion parental consent).

The regression in column (B) introduces a dummy variable for state-issues where

a ballot measure was approved by the voters. As before, the dummy on this variable

indicates the direct effect of the initiative, while the dummy on initiative availability

indicates the indirect effect. Both direct and indirect effects are positive and different

from zero at better than the 1 percent level, and again, the direct effect is much larger

than the indirect effect. The direct and indirect effect coefficients are different from each

other at the 1 percent level ( 012.=p ). At mean values for the control variables, the

indirect effect ranges from 1.5 to 20.8 percent, depending on the issue, with a median of

12.5 percent. The direct effect adds from 3.2 to 67.3 percent to the probability of a

conservative outcome, depending on the issue, with a median of 52.8 percent. Column

(C) reports estimates from a regression based on the two-stage procedure discussed above

that uses predicted values for the DIRECT initiative dummy instead of the actual values.

The rationale for the two stage procedure is less compelling here than for congruence

because there is no obvious reason to believe that initiatives would be more likely in

conservative than liberal states, but the changes in coefficients suggests that endogeneity

may be an issue. In column (C), both direct and indirect effects are still positive, but the

indirect effect falls in magnitude and cannot be distinguished from zero at conventional

levels of significance, while the direct effect becomes larger. There is fairly convincing

evidence that the direct effect of the initiative is important and pushes policy in a

conservative direction, but the case of indirect effects is mixed.

The regressions in columns (D) and (E) attempt to separate the indirect effect into

a threat component (given by the initiative availability dummy) and a signaling

component (given by the INDIRECT/SIGNAL dummy). Column (E) uses predicted

values of the DIRECT and INDIRECT/SIGNAL initiative variables, while column (D)

uses the actual values. The clearest finding continues to be the importance of the direct

effect, which shows up as a positive and highly significant coefficient on the DIRECT

Page 17: Disentangling the Direct and Indirect Effects of the Initiative

16

dummy in both columns. The model in column (D) suggests a negative effect of the

signaling channel – having an unsuccessful initiative on the ballot reduces the likelihood

of a conservative outcome – but the coefficient is not statistically different from zero in

column (E) where the two-stage procedure is used. As noted above, one reason for the

unreliable estimates on the INDIRECT/SIGNAL variable is probably the paucity of

observations in this category. The coefficient on the initiative availability variable

remains positive but is significant only in column (D), providing modest support for a

threat effect.

As for the control variables, conservative policy choices are more likely as the

percent of the population in support of conservative outcomes rises, as in Erikson et al.

(1993). Populous states tend to adopt more liberal policies. States with more educated

residents also tend to adopt liberal laws, but the coefficients are not statistically different

from zero in any specification. Southern states and newer states are significantly more

likely to adopt conservative outcomes. States with elected judges also are more likely to

choose conservative laws.

5. Implications

A healthy scholarly literature generally agrees that the initiative process changes

policy outcomes, and many studies find that it makes laws more congruent with public

opinion, and tilts them in a conservative direction.12 However, little is known about how

the initiative brings about policy changes. This paper develops an empirical strategy to

measure the impact of three potential channels of influence that have been emphasized in

the literature: an indirect “threat” channel, an indirect “signaling” channel, and a direct

“override” channel. The strategy is to compare policy choices in states that use only the

direct channel, states that use the indirect channels, and states that do not have the

initiative process at all. Because the approach requires identification of states that have

actually passed initiatives on specific issues, it can only be implemented by focusing on

specific issues rather than broad policy indexes. This study focuses on nine specific

policy issues for which state-by-state public opinion data are available.

12 The conservative tilt emerges from studies that cover the last several decades. Research from the early

twentieth century finds a liberal tilt associated with the initiative process (Matsusaka, 2000).

Page 18: Disentangling the Direct and Indirect Effects of the Initiative

17

A central finding is that the direct effect turns out to be very important, both

quantitatively and statistically. In all specifications, states that actually pass initiatives on

specific issues choose policies that are more congruent with public opinion and more

conservative than states where the initiative is available but not used or states where the

initiative is unavailable. For the median issue, having a successful initiative on the ballot

increases the probability of congruence by 51.7 percent. There is also evidence that the

initiative affects policy indirectly. States that have the initiative process available but did

not approve an initiative are more likely to choose congruent policies than noninitiative

states. However, the indirect effect is much smaller in magnitude than the direct effect,

and in some specifications cannot be distinguished from zero at conventional levels of

statistical significance. The evidence for an indirect “signaling” effect is weak and

unreliable, perhaps due in part to the relatively small number of observations in this

category. In any case, the rarity of cases where states voted on an initiative but did not

approve it together with the weak results for the signaling effect suggests that the

signaling channel was not of primary importance for the issues studied here.

These findings confirm one of the central insights of game theoretic models of the

initiative process, that the process can influence policy by providing a threat, even if it is

not used (Gerber, 1996; Matsusaka and McCarty, 2001). However, the evidence suggests

that the threat effect is of secondary importance. Perfect information models in which the

initiative matters only through the threat effect appear to be missing most of the story.

This suggests that more research would be useful on incomplete information models, and

more generally, models in which the initiative matters directly by overriding the

legislature rather than through threats. Gerber (1998) offers some thoughts on why

indirect effects may be difficult for groups to exploit, and further research along these

lines would seem to be in order.

The evidence also provides some guidelines for empirical research. The finding

that indirect channels are important for policy choices implies that empirical studies

focusing only on the direct channel will be missing a potentially important part of the

story. Specifically, studies that examine only initiatives that reach the ballot or only

initiatives that are approved will not capture the threat effect. To determine the full effect

Page 19: Disentangling the Direct and Indirect Effects of the Initiative

18

of the initiative, it is necessary to adopt a methodology that captures both direct and

direct effects.

Finally, the paper offers a strategy for separating the direct and indirect channels

of initiative influence. I believe this is the first such study that tries to disentangle the

different channels of influence. The finding that more than one channel is important

suggests it might be worthwhile to further refine the approach. The empirical strategy

may also be useful in studying other political institutions that are likely to have both

direct and indirect effects, such as the executive veto.

Page 20: Disentangling the Direct and Indirect Effects of the Initiative

19

References

Barro, Robert J., “The Control of Politicians: An Economic Model,” Public Choice,

March 1973, Vol. 14(1), 19-42.

Besley, Timothy and A. Abigail Payne, “Implementation of Anti-Discrimination Policy:

Does Judicial Selection Matter?,” Working paper, London School of Economics

and McMaster University, May 2006.

Bowler, Shaun and Todd Donovan, “Measuring the Effects of Direct Democracy on State

Policy: Not All Initiatives Are Created Equal,” State Politics and Policy Quarterly,

Fall 2004, Vol. 4(3), 345-363.

Chen, Jowei and Neil Malhotra, “The Law of k/n: The Effect of Chamber Size on

Government Spending in Bicameral Legislatures,” Working Paper, Stanford

University, 2007.

DelRossi, Alison F. and Robert P. Inman, “Changing the Price of Pork: The Impact of

Local Cost Sharing on Legislators’ Demands for Distributive Public Goods,”

Journal of Public Economics, February 1999, Vol. 71(2), 247-273.

Erikson, Robert S., Gerald C. Wright, and John P. McIver, Statehouse Democracy:

Public Opinion and Policy in the American States, Cambridge, UK: Cambridge

University Press, 1993.

Ferejohn, John, “Incumbent Performance and Electoral Control,” Public Choice, January

1986, Vol. 50(1-3), 5-25.

Gerber, Elisabeth R., “Legislative Response to the Threat of Popular Initiatives,”

American Journal of Political Science, February 1996, Vol. 40(1), 99-128.

Gerber, Elisabeth R., “Pressuring Legislatures through the Use of Initiatives: Two Forms

of Indirect Influence,” in Citizens as Legislators: Direct Democracy in the United

States, edited by Shaun Bowler, Todd Donovan, and Caroline J. Tolbert,

Columbus, OH: Ohio State University Press, 1998.

Gerber, Elisabeth R., The Populist Paradox: Interest Group Influence and the Promise of

Direct Legislation, Princeton, NJ: Princeton University Press, 1999.

Gerber, Elisabeth R. and Simon Hug, “Minority Rights and Direct Legislation,” Working

Paper, February 2002.

Page 21: Disentangling the Direct and Indirect Effects of the Initiative

20

Gerber, Elisabeth R. and Arthur Lupia, “Campaign Competition and Policy

Responsiveness in Direct Legislation Elections,” Political Behavior, September

1995, Vol. 17(3), 287-306.

Gerber, Elisabeth R., Arthur Lupia, Mathew D. McCubbins, and D. Roderick Kiewiet,

Stealing the Initiative: How State Government Responds to Direct Democracy,

Upper Saddle River, NJ: Prentice Hall, 2001.

Gilligan, Thomas W. and John G. Matsusaka, “Systematic Deviations from Constituent

Interests: The Role of Legislative Structure and Political Parties in the States,”

Economic Inquiry, July 1995, Vol. 33(3), 383-401.

Gilligan, Thomas W. and John G. Matsusaka, “Fiscal Policy, Legislature Size, and

Political Parties: Evidence from the First Half of the Twentieth Century,”

National Tax Journal, March 2001, Vol. 54(1), 57-82.

Hanssen, F. Andrew, “Appointed Courts, Elected Courts, and Public Utility

Commissions: Judicial Independence and the Energy Crisis,” Business and Politics,

April 1999, Vol. 1(2), 179-201.

Kalt, Joseph P. and Mark A. Zupan, “Capture and Ideology in the Economic Theory of

Politics,” American Economic Review, 1984, Vol. 74, 279-300.

Kau, James B. and Paul H. Rubin, “Self-interest, Ideology and Logrolling in

Congressional Voting,” Journal of Law and Economics, 1979, Vol. 22, 365-384.

Kousser, Thad, Jeffrey Lewis, and Seth Masket, “Ideological Adaptation? The Survival

Instinct of Threatened Legislators,” Journal of Politics, forthcoming.

Key, V. O., Jr. and Winston W. Crouch, The Initiative and Referendum in California,

Berkeley, CA: University of California Press, 1938.

Lupia, Arthur, “Dumber than Chimps? An Assessment of Direct Democracy Voters,” in

Dangerous Democracy? The Battle over Ballot Initiatives in America, edited by

Larry J. Sabato, Howard R. Ernst, and Bruce A. Larson, Lanham, MD: Rowman

and Littlefield Publishers, 2001.

Lupia, Arthur and John G. Matsusaka, “Direct Democracy: New Approaches to Old

Questions,” Annual Review of Political Science, 2004, Vol. 7, 463-482.

Matsusaka, John G., “Economics of Direct Legislation,” Quarterly Journal of

Economics, May 1992, Vol. 107(2), 541-571.

Page 22: Disentangling the Direct and Indirect Effects of the Initiative

21

Matsusaka, John G., “Fiscal Effects of the Voter Initiative: Evidence from the Last 30

Years,” Journal of Political Economy, June 1995, Vol. 103(3), 587-623.

Matsusaka, John G., “Fiscal Effects of the Voter Initiative in the First Half of the 20th

Century,” Journal of Law and Economics, October 2000, Vol. 43(2), 619-648.

Matsusaka, John G., For the Many or the Few: The Initiative, Public Policy, and

American Democracy, Chicago, IL: University of Chicago Press, 2004.

Matsusaka, John G., “Direct Democracy Works,” Journal of Economic Perspectives,

Spring 2005, Vol. 19(2), 185-206.

Matsusaka, John G., “Direct Democracy and Electoral Reform,” in The Marketplace of

Democracy: Electoral Competition and American Politics, edited by Michael P.

McDonald and John Samples, Washington, D.C.: Brookings Institution Press,

2006.

Matsusaka, John G., “Institutions and Popular Control of Public Policy,” Working Paper,

USC Marshall School of Business, 2007a.

Matsusaka, John G., “Direct Democracy and Social Issues,” Working Paper, USC

Marshall School of Business, 2007b.

Matsusaka, John G., “Direct Democracy and Public Employment,” Working Paper, USC

Marshall School of Business, 2007c.

Matsusaka, John G., “Direct Democracy and the Executive Branch,” in Directing

Democracy, edited by Shaun Bowler and Amihai Glazer, Palgrave Macmillan,

forthcoming.

Matsusaka, John G. and Nolan M. McCarty, “Political Resource Allocation: Benefits and

Costs of Voter Initiatives,” Journal of Law, Economics, and Organization, October

2001, Vol. 17(2), 413-448.

McCarty, Nolan M., “Proposal Rights, Veto Rights, and Political Bargaining,” American

Journal of Political Science, July 2000, Vol. 44(3), 506-522.

Miller, Warren E. and Donald E. Stokes, “Constituency Influence in Congress,”

American Political Science Review, March 1963, Vol. 57(1), 45-56.

Persily, Nathanial and Melissa Cully Anderson, “Regulating Democracy through

Democracy: The Use of Direct Legislation in Election Law Reform,” Southern

California Law Review, May 2005, Vol. 78(4), 997-1034.

Page 23: Disentangling the Direct and Indirect Effects of the Initiative

22

Peltzman, Sam, “Constituent Interest and Congressional Voting,” Journal of Law and

Economics, April 1984, Vol. 27(1), 181-210.

Romer, Thomas and Howard Rosenthal, “The Elusive Median Voter,” Journal of Public

Economics, October 1979, Vol. 12, 143-170.

Roosevelt, “Nationalism and Popular Rule,” in The Initiative, Referendum, and Recall,

edited by William Bennett Munro, New York, NY: D. Appleton and Company,

1912.

Smith, Daniel A. and Caroline J. 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, 2004.

Tolbert, Caroline J., “Changing Rules for State Legislatures: Direct Democracy and

Governance Policies,” Chapter 8 in Citizens and Legislators: Direct Democracy

in the United States, edited by Shaun Bowler, Todd Donovan, and Caroline J.

Tolbert, Columbus, OH: Ohio State University Press, 1998.

Tolbert, Caroline J. and Daniel A. Smith, “Representation and Direct Democracy in the

United States,” Representation, 2006, Vol. 42(1), 25-44.

Weingast, Barry R., Kenneth A. Shepsle, and Christopher Johnsen, “The Political

Economy of Benefits and Costs: A Neoclassical Approach to Distributive Politics,”

Journal of Political Economy, 1981, Vol. 93(4), 642-664.

Wilson, Woodrow, “The Issues of Reform,” in The Initiative, Referendum, and Recall,

edited by William Bennett Munro, New York, NY: D. Appleton and Company,

1912.

Page 24: Disentangling the Direct and Indirect Effects of the Initiative

Table 1. Issues and Number of Initiatives

Issue Question Mean % in favor Years

Number of states with initiatives on issue

Number of states with successful initiatives

Abortion, late term / partial birth

“There has been discussion recently about a proposed law to ban certain types of late-term abortions, sometimes called partial birth abortions. Do you favor or oppose a ban on these types of abortions?”

67.2 1998, 2000, ‘04 2 0

Abortion, parental consent

“Would you favor or oppose a state law that would require parental consent before a teen-ager under 18 could have an abortion?”

74.4 1988, ‘90, 2000 3 1

Abortion, public funding

“Would you favor or oppose a law in your state that would allow the use of government funds to help pay for the costs of abortion for women who cannot afford them?”

49.4 1988 6 2

Death penalty

“Do you favor or oppose the death penalty for persons convicted of murder?”

78.0 1988, ‘90, ‘94, ‘96, ‘98, 2000,

‘04

4 4

English only

“Do you favor a law making English the official language of the United States, meaning government business would be conducted in English only, or do you oppose such a law?’

70.9 1990, ‘98, 2000 6 6

Estate tax

“There has been a lot of talk recently about doing away with the tax on large inheritances, the so-called ‘[estate/death]’ tax. Do you favor or oppose doing away with the [estate/death] tax?”

72.5 2002 5 3

Job discrimination, homosexuals

“Do you favor or oppose laws to protect homosexuals against job discrimination?”

65.4 1990, ‘94, ‘96, 2000, ‘04

6 2

Same-sex marriage

“Should same-sex couples be allowed to marry, or do you think they should not be allowed to marry?” Responses: 1=Allowed. 5= Not allowed. 7=Not allowed to marry, but civil unions allowed. (“In favor” = response 1)

32.9 2004 11 10

Term limits “A law has been proposed that would limit members of Congress to no more than 12 consecutive years of service in that office. Do you favor or oppose such a law?”

78.5 1990, ‘94, ‘96, ‘98

22 19

Note. “Question” is the precise question asked in the American National Election Studies survey. “Year” is the study year, except that 1988 refers to the 1988-1992 ANES Pooled Senate File. Statistics for “Percent in Favor” was computed with the state as the unit of observation. States were counted as having an initiative if one or more appeared on the ballot in any year. A successful initiative was one that was approved by the voters.

Page 25: Disentangling the Direct and Indirect Effects of the Initiative

Table 2. Congruence across All States and Issues Initiative status Percent congruent Observations

NO INITIATIVE: Initiative not available in state 51.9 243

INDIRECT: Initiative available in state but no ballot proposition approved 60.1 160

DIRECT + INDIRECT: Initiative available in state and ballot proposition approved 78.7 47

Note. The unit of observation is an issue-state. The table reports the percent of observations in which policy is consistent with the views of a majority of its people. The initiative is “available” if the state is one of the 23 that allows citizens to propose and approve laws by petition.

Page 26: Disentangling the Direct and Indirect Effects of the Initiative

Table 3. Logistic Regressions Explaining Likelihood of Congruence 2-stage 2-stage (A) (B) (C) (D) (E)

INDIRECT: Dummy = 1 if initiative available

0.85*** (0.27)

0.50* (0.29)

0.13 (0.32)

0.47 (0.30)

0.20 (0.37)

DIRECT: Dummy = 1 if initiative available & at least one approved

… 2.26*** (0.57)

3.94*** (0.95)

2.08*** (0.73)

4.52** (1.94)

INDIRECT/SIGNAL: Dummy = 1 if initiative available & at least one on ballot

… … … 0.24 (0.58)

-0.65 (1.87)

Public opinion: size of majority % 13.49*** (2.16)

14.17*** (2.23)

14.32*** (2.21)

14.22*** (2.23)

14.23*** (2.22)

Population, log -0.13 (0.13)

-0.22 (0.14)

-0.28** (0.14)

-0.22 (0.14)

-0.27* (0.14)

Education, percent with high school diploma

-3.71 (3.68)

-4.61 (3.80)

-4.33 (3.79)

-4.60 (3.80)

-4.37 (3.80)

Dummy = 1 for Southern state 0.90*** (0.34)

0.99*** (0.35)

1.07*** (0.36)

0.99*** (0.35)

1.06*** (0.36)

Years since state entered Union 0.45 (0.34)

0.58 (0.35)

0.61* (0.35)

0.59* (0.36)

0.59 (0.36)

Dummy = 1 if judges stand for re-election

0.64* (0.33)

0.68** (0.33)

0.67** (0.33)

0.69** (0.33)

0.66* (0.34)

Note. Each column report estimates from a logistic regression to explain the probability of a congruent outcome, defined as a policy outcome supported by a majority of citizens. Standard errors are in parentheses beneath the coefficient estimates. The unit of observation is a state-issue. Each regression is based on nine issues in 50 states (450 observations), and includes separate intercepts for each issue. In columns (C) and (E) the DIRECT and INDIRECT/COMMUNICATION dummies are predicted values from a first-stage logistic regression with signature requirements and years since adoption of the initiative as explanatory variables. Significance levels are indicated: *=10%, **=5%, ***=1%.

Page 27: Disentangling the Direct and Indirect Effects of the Initiative

Table 4. Logistic Regressions Explaining Likelihood of Conservative Law 2-stage 2-stage (A) (B) (C) (D) (E)

INDIRECT: Dummy = 1 if initiative available

1.23*** (0.29)

0.84*** (0.31)

0.11 (0.30)

1.08*** (0.32)

0.30 (0.34)

DIRECT: Dummy = 1 if initiative available & at least one approved

… 2.75*** (0.64)

3.72*** (0.91)

4.70*** (0.96)

5.58*** (1.82)

INDIRECT/SIGNAL: Dummy = 1 if initiative available & at least one on ballot

… … … -2.36*** (0.81)

-1.96 (1.64)

Public opinion: percent in favor of conservative law

9.44*** (2.05)

10.10*** (2.13)

3.48** (1.75)

9.18*** (2.15)

2.66 (1.88)

Population, log -0.42*** (0.14)

-0.55*** (0.15)

-0.25* (0.13)

-0.52*** (0.15)

-0.21 (0.14)

Education, percent with high school diploma

-5.11 (3.79)

-6.32 (3.98)

-2.98 (3.63)

-6.38 (4.03)

-3.01 (3.64)

Dummy = 1 for Southern state 1.71*** (0.39)

1.91*** (0.42)

0.82** (0.34)

1.95*** (0.42)

0.83** (0.34)

Years since state entered Union 0.91** (0.38)

1.11*** (0.40)

0.63* (0.33)

1.01** (0.40)

0.56 (0.34)

Dummy = 1 if judges stand for re-election

1.22*** (0.37)

1.31*** (0.38)

0.72** (0.34)

1.30*** (0.38)

0.72** (0.34)

Note. Each column reports estimates from a logistic regression to explain the probability of a conservative outcome. Standard errors are in parentheses beneath the coefficient estimates. The unit of observation is a state-issue. Each regression is based on nine issues in 50 states (450 observations), and includes separate intercepts for each issue. In columns (C) and (E) the DIRECT and INDIRECT/SIGNAL dummies are predicted values from a first-stage logistic regression with signature requirements and years since adoption of the initiative as additional explanatory variables. Significance levels are indicated: *=10%, **=5%, ***=1%.