453592
tate Politics & Policy QuarterlyDamore et al.
SPA12410.1177/1532440012453592S
Agenda Setting by Direct
Democracy: Comparing
the Initiative and the
Referendum
State Politics & Policy Quarterly
12(4) 367­–393
© The Author(s) 2012
Reprints and permission:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/1532440012453592
http://sppq.sagepub.com
David F. Damore1, Shaun Bowler2,
and Stephen P. Nicholson3
Abstract
Using research on the initiative as a point of comparison, we consider how frequently
and for what ends state legislators use the referendum. Akin to initiative use, we
find that legislators are constrained by procedural hurdles in their ability to place
referendums on the ballot. However, in contrast to research on the initiative, which
emphasizes the role of interest groups as the drivers of initiatives, our analysis
suggests that referendum use is motivated by partisan legislative majorities seeking to
achieve a mix of political and policy goals.
Keywords
direct democracy, legislative referendum, state political institutions, state legislatures
The literature on the initiative offers a well-developed understanding of who uses the
process, for what ends, and under what conditions. By comparison, the study of referendum use by state legislatures lags significantly behind (for exceptions, see Gazey
1971; Zimmerman 2001). Indeed, beyond a general sense that legislators in some
instances are obligated to place referendums before voters, there is little scholarship
examining how the referendum is used in practice and for what purposes.1 This dearth
1
University of Nevada, Las Vegas, USA
University of California, Riverside, USA
3
University of California, Merced, USA
2
Corresponding Author:
David F. Damore, Department of Political Science, University of Nevada, Las Vegas,
505 Maryland Parkway, Box 455029, NV 89154, USA
Email: [email protected]
368
State Politics & Policy Quarterly 12(4)
of research is surprising given that referendums are used in all but one state and they
appear on the ballot twice as often as initiatives (Magleby 1984).
As a consequence, besides the fact that both types of measures are either passed or
rejected by state electorates, we know little about the differences and similarities in
how these two institutions of collective policy making operate—often side by side.
Yet, because initiatives and referendums travel different paths to the ballot, there are
reasons to expect important differences between the two, particularly with respect to
where agenda setting power lies. Research on the initiative posits that interest groups
often use the process to circumvent, weaken, or constrain state governing institutions
(Banducci 1998; Boehmke 2005a; 2005b; Gerber 1999; Matsusaka 1992; Matsusaka
and McCarty 2001; Price 1975). In contrast, because the referendum is under the purview of state legislatures, its use is likely governed by the preferences and goals of
legislators.
The purpose of this article, then, is to compare how different variants of direct
democracy—the referendum and initiative—are used to set the agenda for policy
change in the American states. In the next section, we identify the political actors who
use these processes and for what ends, as well as the institutional and political constraints under which they operate. We then test hypotheses derived from this framework using data for all initiative and discretionary referendums put before voters in the
American states between 1990 and 2008. In the third section, we present the results of
our analysis suggesting that the referendum and initiative work in fundamentally different ways. Most notably, although both are constrained by qualification requirements, we find in contrast to the initiative that interest groups are not a significant
factor in qualifying referendums for the ballot. Instead, the partisan composition of
state governing institutions affects referendum use, as well as the policy goals imbued
in these proposals. We conclude by discussing the implications that our effort has for
the literature and suggesting directions for future research.
Policy Making at the Ballot Box:
Comparing the Initiative and Referendum
The study of direct democracy is largely synonymous with analyses of the initiative.
Less well understood is the more commonly used (Magleby 1984) legislative referendum; the process whereby legislatures in 49 states are able to place either statutory or
constitutional proposals before state electorates. Although both forms of direct
democracy allow voters to have a voice in state policy making, there are important
differences between the referendum and the initiative not least of which is what actors
use each process to set the agenda for policy change. In the discussion that follows,
we compare the initiative and referendum by considering the motivations that underlie
why these processes are used to facilitate policy change and the political and institutional factors that constrain political actors’ ability to use these processes.
Although labeled the “citizen” initiative and intended to break the influence of
“special” interests, research on the initiative challenges the idea that the initiative is a
Damore et al.
369
tool of “grass roots” politics. In particular, the research demonstrates that organized
interests are vital to the process, particularly in terms of qualifying proposals for
the ballot (Banducci 1998; Boehmke 2005a; 2005b; Gerber 1999; Matsusaka 1992;
Matsusaka and McCarty 2001; Price 1975). Others have found that political parties
use the initiative for agenda setting in elections (e.g., Nicholson 2005).
In many instances, the policy goals that initiative proponents seek are designed to
weaken state legislatures by imposing budgetary constraints or other restrictions such as
term limits (Tolbert 1998; 2003). Groups also may use the initiative to raise the profile
of social issues that may get little traction in the legislature but may resonate in the electorate (Damore and Nicholson 2011). In addition, organized interests may participate in
initiative politics for reasons of organizational maintenance such as advertising the
group and its agenda, and networking with like-minded interests (see Gerber 1999, 82).
Whether initiatives are inherently skewed liberally or conservatively is less certain
and a debate exists over whether the process produces outcomes closer to—or further
from—a state’s median voter (e.g., Besley and Case 2003; Camobreco 1998; Lascher,
Hagen, and Rochlin 1996; Lax and Phillips 2009; Matsusaka 2001; 2004). Regardless,
the initiative process can see most any issue raised provided that proponents can overcome qualification hurdles. On this point, prior work emphasizes the role of procedural
constraints such as signature requirements in limiting the number of proposals that
qualify for the ballot (Banducci 1998; Boehmke 2005b; Bowler and Donovan 2004).
In contrast, because the referendum originates within the legislature, we expect that
the preferences of legislators, as opposed to other actors such as interest groups, will
be preeminent. To be sure, legislators do not have free reign over referendum use as
there are a number of instances when referendums must be put before voters. With the
exception of Delaware, changes to state constitutions must be ratified by voters
(Colantuono 1987; Fisch, 2006; Krislov and Katz 2008; Schmidt 1989; Zimmerman
2001).2 Some state constitutions require that proposals be periodically referred to the
ballot asking voters whether they wish to hold constitutional conventions. On financial
matters, state legislators must place full-faith-and-credit obligation bonds before the
voters (Kiewiet 1995; Kiewiet and Szakaty 1996).3
In these cases, referendum use in the American states is similar to the referendum
process elsewhere in the world (see LeDuc 2003). However, if we take a step back in
the causal process, except for instances where state constitutions mandate that voters
be asked whether they want to hold a constitutional convention, all other referendums
originate at the discretion of the legislature. So, although bond proposals may require
popular consent, it is a legislative prerogative to seek bond funding in the first place.
Similarly, although referendums that seek to amend state constitutions must gain popular passage, it is the volition of the legislature that causes such proposals to be placed
on the ballot.
A recent example may help to illuminate these dynamics. In 2010, the Oklahoma
Legislature referred State Question 755 to the ballot. Better known as the “Sharia Law
Amendment,” the measure sought to prohibit Oklahoma courts from drawing on
Sharia law when making rulings. Because the proposal would amend the Oklahoma
370
State Politics & Policy Quarterly 12(4)
Constitution, a referendum was required. This, however, does not explain why
Oklahoma legislators pursued the policy; only that they sought to achieve their policy
goals via the referendum. Moreover, the content of the measure belies the notion that
the referendum is only used for issues that are unlikely to engender disagreement.
Indeed, central to developing our understanding of the referendum process is having some sense of the goals that legislators are likely to pursue via the referendum. To
the extent that there is a literature on referendum use, it suggests that politicians may
have political motivations for using them. Gazey (1971, 131), for example, argues that
legislators will use the referendum to “pass the buck” to the electorate on tax increases
to avoid politically difficult decisions. This view is in keeping with the intuition of
Romer and Rosenthal’s (1979) model of agenda setting where politicians use the referendum for budget maximization. A particularly interesting motivation for referendum use turns on the fact that the process is often used to alter constitutions as in the
case of Oklahoma’s State Question 755 discussed above. By packaging preferred policies as legislatively referred constitutional amendments, legislators can commit their
states to a policy that is procedurally more difficult to alter in the future.
At the same time, if referendum use is driven by a mix of policy goals and political
motivations, legislators should be constrained in their ability to use the process.
Earlier, we mentioned the procedural barriers that hinder initiative use. Similar hurdles
exist for the referendum. Specifically, qualifying most referendums necessitates a high
degree of coordination as many states require the support of super-majorities in both
chambers, often in successive sessions.4 The ability to overcome these hurdles should
be greater in contexts where the same party controls both legislative chambers.
Thus, whereas prior research suggests that the presence of divided government may
encourage initiative use as interest groups turn to the initiative to circumvent policy
gridlock (Banducci 1998; see also Gerber 1996; 1999), the effects of divided government for referendum use may be more nuanced. If the chambers of a state legislature
are under the control of different parties, then it may be more difficult to muster the
necessary support to qualify the proposal for the ballot, particularly if the policy goals
being sought are partisan or ideological. However, if the two Houses are under the
control of one party and the governorship is under the control of the other, then the
legislature may use the referendum to circumvent a recalcitrant governor.
Thus far, our discussion has focused on the politics that underlie referendum use
and the political and institutional factors that are likely to impede legislators’ ability to
place such measures on the ballot. Qualifying a referendum, of course, is no guarantee
that the proposal will be approved at the ballot box. More generally, measure passage
offers another important point of comparison between the initiative and the referendum.5 The literature on the initiative suggests that interest groups may benefit even if
a proposal does not pass. As noted above, organized interests may have goals (i.e.,
mobilization, networking, membership growth, etc.) besides policy change for qualifying initiatives. Interest groups may, in a sense, “win” even if a proposal loses at the
ballot.
Damore et al.
371
But the same is unlikely to hold for the referendum. Because of the degree of coordination required to qualify referendums, legislators may be hesitant to use the process
unless there is a reasonable prospect that the proposal will pass. Moreover, because
legislators have leeway over the content and timing of referendums, the ability to use
the referendum successfully may reflect legislative capabilities as much as intent.
There exists a rich vein of research (e.g., King 2000; Squire 1992; 2007) distinguishing between amateur and professional legislatures, and the consequences that professionalism has for representation, turnover, and decision making. The proliferation of
term limits has further complicated these dynamics (Kousser 2005; Moncrief, Niemi,
and Powell 2004). All else equal, the skills needed to successfully package proposals
may be less prevalent in amateur or term-limited legislatures, and as a consequence,
these legislatures may produce fewer referendums that gain the support of the
electorate.
One policy area where the prospects for passage may loom large is for proposals
seeking to expand the scope of state government.6 Literature on budget maximizing
(Romer and Rosenthal 1979) and buck passing (Gazey 1971) suggests that legislators
may use the referendum to propose tax increases to sidestep responsibility for such
increases. At the same time, even the most well-crafted tax measure may fail given the
electorate’s antipathy toward higher taxes. In this regard, the referendum affords budget maximizing legislators (e.g., Democrats) another type of buck passing. By packaging referendums that expand the scope of government not as tax increases, but as bond
proposals, legislators can achieve their preferred policy goals while avoiding the
super-majority thresholds that exist in some states for such increases.7 Moreover,
unlike tax increases with their immediate costs, bonds provide increased spending in
the short term with the costs pushed into the future.
In sum, the above discussion has designated the primary political actors who use
the initiative (interest groups) and referendum (state legislators) to pursue policy
change at the ballot, posited a set of political and policy motivations to explain why
these actors seek to make collective decisions via direct democracy, and identified the
political and institutional constraints under which these actors operate. In the case of
the referendum, we have also considered factors affecting the likelihood of passage.
Below, we turn our attention to the empirical analysis of these arguments.
Research Design
To evaluate hypotheses derived from the above discussion, we use data from the
National Conference of State Legislatures’ (NCSL) Ballot Measure Database for the
years 1990 through 2008.8 Below, we discuss the measurement of the dependent variables and the methodology used to estimate the statistical models. Appendix A details
the measurement and data sources for the independent variables included in these
analyses.
We present our analysis in three stages. In the first stage, we examine the frequency
of referendum and initiative use. The dependent variables for these models are counts
372
State Politics & Policy Quarterly 12(4)
of the number of such proposals appearing on state ballots between 1990 and 2008. In
operationalizing the dependent variable for the referendum analysis, we only consider
referendums that appear on the ballot at the discretion of the legislature. Omitted then
are all automatically referred referendums (the constitutions in 14 states require periodic referendums asking the citizenry whether they wish to hold a constitutional convention), referendums proposed by entities other than state legislatures (such as the
Arizona Commission on Salaries for Elective Office or Florida’s Constitutional
Revision Commission), and popular referendums (a process used in 23 states that
allow citizens to invalidate legislative actions).9 For the initiative analysis, data are
included for the 24 initiative states. For the referendum model, data are included for
all states except for Delaware, which does not have the referendum process, and
Nebraska. Nebraska is excluded because the state’s unicameral and nonpartisan legislature yields no values for the size of the majority variables and precludes testing our
hypothesis regarding the effects of divided partisan legislative control.
Consistent with Banducci (1998) and Boehmke (2005b), we collapse the data
over 2-year periods to account for off-year elections, as well as to make the data
consistent with state legislative terms in most states. Thus, for each state, there are
10 observations (48 states by 10 observations for the referendum model, N = 480,
and 24 states by 10 observations for the initiative model, N = 240). Following
Boehmke (2005b), the models are estimated as negative binomial regressions, with
robust clustered (by state) standard errors (RCSEs) to control for cross-case dependency. Recent work by Harden (2011), however, suggests that RCSEs may be biased
downward, increasing the likelihood of Type I errors. Harden recommends using
bootstrap cluster standard errors (BCSEs) with data such as ours. The BCSE estimates for these models are presented in Appendix B. Appendix B also provides an
extension of our first stage analysis that addresses any dependency between the
referendum and initiative processes that may exist in the initiative states (sans
Nebraska) by simultaneously analyzing initiative and referendum use via a seemingly
unrelated estimation strategy.
In the second stage, we consider some of the different purposes that may motivate
referendum use by presenting biennial count models where the dependent variables
are the number of referendums packaged as bond proposals and constitutional amendments using data from all states except for Delaware and Nebraska. These models are
also estimated as negative binomial regressions with state-specific RCSE. Appendix B
presents the BCSE estimates for these models.
In the final stage of our analysis, we examine the conditions that promote successful agenda setting by presenting a model of referendum passage. For this model, the
unit of analysis is the referendum and the model is estimated using logit with RCSE
for each state election year. To address selection effects between the agenda setting
and passage stages, we examine both processes using a Heckman selection model.
Appendix B presents the results of this analysis and the BCSE estimates for the passage model.
Damore et al.
373
Results
Frequency of Referendum and Initiative Use
In our first stage analysis considering the frequency of referendum and initiative use,
the main variables of interest capture the effects of institutional and political considerations on the frequency of referendum and initiatives appearing on statewide ballots.
For institutional constraints, we include Initiative Qualification Difficulty and
Referendum Qualification Difficulty and expect these variables’ coefficients to sign
negatively. For political constraints, we include a measure of split partisan control of
the legislature (Divided Legislature) and a measure of interbranch partisan division
(Divided Government). Because prior research (see Banducci 1998; Boehmke 2005b)
suggests the importance of interest groups in promoting initiative use, we include a
measure for the density of state interest group populations (Interest Groups) in both
models.
Following Banducci (1998) and Boehmke (2005b), we include controls for state
partisanship (Republican Strength), ideology (State Liberalism), aspects of state legislative context (Legislative Professionalism and Term Limits), population (Population),
and population change (Population Change). Smith and Fridkin (2008) find that in
states with less interparty legislative competition, legislators were less likely to extend
to voters the choice to adopt the initiative. To account for this, we include two additional control variables, Majority in Lower Chamber and Majority in Upper Chamber,
measuring the size of the majority party’s advantage in each chamber. The results for
the model of initiative use are presented in the first column of Table 1, and the second
column does the same for all discretionary referendums.
Beginning with the benchmark initiative analysis, our results replicate the key findings of prior research (e.g., Banducci 1998; Boehmke 2005b). The positive and statistically significant coefficient for Interest Groups indicates that as the density of a
state’s interest group population increases so does the volume of initiatives. With
respect to divided government, consistent with Banducci (1998), the results indicate
that split partisan control of the executive and the legislative branches is a significant
predictor of initiative use as Divided Government is positive and statistically significant, whereas the coefficient for Divided Legislature is statistically insignificant. Last,
the coefficient for Initiative Qualification Difficulty is significant and positive, indicating that more stringent qualification requirements decrease the number of initiatives
reaching the ballot.
The model of referendum use presented in the second column of Table 1 suggests
a process with some similarities to the initiative, but also important differences. Most
notably, whereas initiative use increases in contexts with higher interest group density,
referendum frequency appears to be shaped by the partisan characteristics of a state’s
legislature. Specifically, although Interest Groups is insignificant, the coefficient for
Divided Legislature is negative and statistically significant suggesting that referendums are more likely to emerge in contexts where one party controls both legislative
374
State Politics & Policy Quarterly 12(4)
Table 1. Negative Binomial Regression Analysis of Initiative and Discretionary Legislative
Referendum Use, 1990 to 2008
Independent variables
Initiatives
Referendums
Bond proposals
Divided Government
0.228* (0.132)
0.064 (0.162)
0.071 (0.323)
Divided Legislature
−0.103 (0.204) −0.549* (0.293)
—
Republican Legislature
—
—
−0.642* (0.359)
Democratic Legislature
—
—
1.10* (.378)
Interest Groups
0.831* (0.469)
0.279 (0.471)
0.363 (0.773)
Initiative Qualification
−0.337* (0.095)
—
—
Difficulty
Referendum
—
−0.483* (0.223)
—
Qualification Difficulty
Majority in Lower
—
−0.818 (0.752)
−1.69 (1.71)
Chamber
Majority in Upper
—
−0.292 (0.783)
0.820 (1.55)
Chamber
Legislative
1.19 (1.31)
−0.826 (1.35)
1.32 (2.47)
Professionalism
Term Limits
—
0.332* (0.197) 0.774* (0.304)
Prior Amendments
—
—
—
Republican Party
−0.009 (0.015)
0.003 (0.015)
0.045 (0.033)
Strength
State Liberalism
0.003 (0.007) −0.005 (0.009) 0.051* (0.018)
Population
0.385 (0.324)
0.244 (0.287)
0.140 (0.614)
Population Change
0.188* (0.081) −0.059 (0.071)
0.065 (0.117)
Constant
7.78* (3.25)
3.64 (3.63)
−3.60 (5.36)
N
240
480
480
Log likelihood
−498.97
−1,064.12
−365.80
Wald χ2
47.51*
24.74*
55.19*
Amendments
0.118 (0.170)
—
0.365 (0.253)
0.429 (0.356)
0.386 (0.534)
—
−0.506* (0.246)
−0.769 (0.747)
−0.603 (1.01)
−1.03 (1.02)
0.478* (0.195)
0.002* (0.001)
0.018 (0.020)
−0.009 (0.009)
0.283 (0.328)
−0.081 (0.121)
3.06 (4.21)
480
−970.04
55.14*
Notes: Robust standard errors clustering on each state are shown in parentheses. Dependent variables
are biennial counts.
*p < .05 (one-tailed).
chambers. As expected, although we find evidence that divided government matters
for both types of measures, for initiative use the key division is between the branches,
whereas for the referendum the partisan split of consequence is within the legislature.
At the same time, we do see similarities across both processes in terms of qualification thresholds. The coefficient for Referendum Qualification Difficulty is negative
and statistically significant indicating that the more difficult the process for qualifying
referendums, the less often the process is used; a result consistent with the effects for
Initiative Qualification Difficulty from the initiative model.10 Thus, across both forms
of direct democracy, the easier the process, the more frequently it is used, and the more
difficult the process, the less often it is used. More generally, the significant effects for
Divided Legislature and Referendum Qualification Difficulty are consistent with our
contention that referendums can be used for noncontroversial ends. As we note above,
Damore et al.
375
if the referendum is used to simply set the agenda for consensual issues, then it is less
likely that these factors would systematically affect patterns of use.
The Preferences of Legislators
Although our results thus far offer useful insights into the differences and similarities
between initiative and referendum use, we have not directly explored how legislators
with different goals may utilize the process. Indirect evidence for this point can be
gleaned from the analysis of all discretionary referendums (Table 1, column 2), which
indicates that the process is more likely to be used when one party controls both legislative chambers. To evaluate party effects more rigorously, we present two additional models of referendum use. For the first model, we consider the frequency of
bond requests and here we expect that Democratic legislatures will be more frequent proposers.7 For this analysis, the specification is the same as for the model
examining all discretionary referendums with two exceptions. First, to capture the
partisan differences, we replace Divided Legislature with two dummy variables coded
one if the Democrats (Democratic Legislature) or Republicans (Republican
Legislature) controlled the legislature. Thus, the referent category is legislatures characterized by split partisan control or instances when chambers are evenly divided.
Second, we drop Referendum Qualification Difficulty. Because this variable captures
qualification requirements for referendums that are packaged as constitutional amendments10 (see Appendix A) and only 2% of the bond proposals in the data set are constitutional amendments, the variable has little theoretical bearing.
The results for the bond analysis presented in the third column of Table 1 support
our expectations about partisan differences in the frequency of bond proposals. The
positive and statistically significant coefficient for Democratic Legislature suggests
that in states where Democrats control the legislature there are significantly more bond
proposals placed on the ballot as compared with states where legislative control is split
across parties. Conversely, the negative and significant coefficient for Republican
Legislature indicates that bond proposals are less likely to emerge from legislatures
controlled by the Republican Party as compared with the referent category. In light of
these partisan differences, the performance of the control variable for state ideology
(State Liberalism) merits mention. Of the three models of referendum use presented
here, it is only for the bonds analysis that the variable is a significant predictor.
As we offer above, a key benefit for using the legislative referendum is that it can
bind legislators’ successors to the present legislators’ policy preferences via the constitutionalization of preferred policies. To assess whether either party is more-or-less
likely to use the referendum process for policy commitment, we model referendum
usage where the dependent variable is the number of proposed referendums that are
constitutional amendments. Specifically, we include Referendum Qualification
Difficulty to assess the impact of qualification thresholds and Prior Amendments to
control for the tendency of some states to frequently amend their constitutions for
minor reasons. The variable is measured as the number of constitutional amendments
prior to 1990. The results for this analysis are presented in the fourth column of Table 1.
376
State Politics & Policy Quarterly 12(4)
Here, we find no evidence of partisan differences in the propensity to use the referendum to alter state constitutions as Democratic Legislature and Republican Legislature
are statistically insignificant. Referendum Qualification Difficulty, however, is significant and in the expected direction, again demonstrating the dampening effect that more
stringent qualification requirements have on referendum use. In addition, note that the
control for prior amendments (Prior Amendments) is positive and significant.
Passage of Legislative Referendums
Of course, qualifying a referendum for the ballot does not guarantee that voters will
accept the legislators’ proposal. To evaluate whether legislators are able to skillfully
time and frame their proposals, we present a model of referendums passage where the
unit of analysis is the referendum (N = 1,536), and the dependent variable is coded
one if the referendum was passed and zero if it failed.11 This model is estimated as a
logit function with RCSE clustered for each state election year.
In specifying the model, we consider that legislators may be more-or-less capable
and face differing levels of opposition to their proposals in the broader political
environment—all of which may affect the likelihood of passage. Thus, although some
of the factors included as controls in prior models may not systematically affect the
frequency of referendum use, they may influence the frequency of success. For example, regardless of content, legislators operating in more professional legislatures may
be better able to package referendums in a manner that increases their chances of passage as compared with amateur legislatures or those where legislative service is
capped. As a consequence, we expect that Term Limits will have a negative effect and
Legislative Professionalism will have a positive effect on the likelihood of passage.
Organized interests may present another obstacle to passage. That is, although the
role of interest groups at the qualification stage may be limited, groups may use their
clout to work against passage, particularly if the referendum is used for nonconsensual
policies that target specific interests. Thus, Interest Groups should sign negatively. In
addition, as we note above, although legislators may consider the likelihood of passage when deciding whether to place discretionary referendum before voters, the ability to accurately predict electoral support may vary. Romer and Rosenthal (1979) and
Matsusaka and McCarty (2001) argue that it may be more difficult to gauge voter
receptivity in states where the population is rapidly changing. To the degree that such
uncertainty increases in fast growing states, Population Change should have a negative effect on the likelihood of passage. Outcomes also may be more difficult to predict in states where the electorate is closely divided. This dynamic may be captured
indirectly in states characterized by a partisan split between the legislature and the
governor (Divided Government). To more precisely tap electoral competitiveness, we
replace Republican Party Strength with Purple State: a dummy variable that equals
one in states where the major-party index (Ceaser and Saldin 2005) is between 45%
and 55%. We anticipate that Divided Government and Purple State will sign
negatively.
Damore et al.
377
Because success or failure also may be affected by the underlying issue, we include
six variables tapping measure content: dummy variables coded one if the measure
proposed a tax increase (Tax Increase), a tax decrease (Tax Decrease), a bond proposal (Bond Proposal), or sought to make the tax-raising process more (Taxes More
Difficult) or less difficult (Taxes Easier). We also include an interval measure capturing the complexity of the proposal as indicated by the number of issue areas associated
with the measure (Measure Complexity). Bowler and Donovan (1998) find that the
more complicated an initiative measure is, the less likely it is to pass. The same may
hold for referendums. We also control for the number of referendums on the ballot
(Number of Referendums) and for states where a super-majority is required for passage
of constitutional amendments (Super-Majority). For consistency with the prior analyses, we include controls for state ideology (State Liberalism) and population
(Population). Last, we include variables (Democratic Legislature and Republican
Legislature) to assess whether partisan differences exist in the ability to successfully
use the referendum.
Table 2 presents the results of the analysis of referendum passage. Considering first
the coefficients tapping legislative capabilities, the negative and significant coefficient
for Legislative Professionalism indicates unexpectedly that state legislatures that are in
session more often, are afforded greater resources, and are compensated more generously are less likely to see their referendums pass as compared with measures proposed
by less professionalized legislatures. The results suggest that there is no difference in
passage rates between legislatures that are not subject to term limits as compared to
those legislatures where service is capped, as Term Limits is statistically insignificant.
The results do, however, suggest that organized interests and uncertain state political environments result in fewer referendums being passed. Specifically, the significant and negative coefficient for Interest Groups suggests that as the concentration of
a state’s interest group population increases, the number of referendums that are
passed decreases. Similarly, referendums are less likely to pass in states characterized
by competitive or less certain electoral environments. Indirect support for this point
can be gleaned by the negative sign for Divided Government and more direct evidence
comes from the negative and statistically significant performance of Purple State. At
the same time, the insignificance for Population Change suggests that uncertainty,
conceptualized in terms of fast growing populations, does not systematically affect
referendum passage.
For the content variables, the results indicate that as compared with other types of
referendums, voters are less likely to support measures that seek to raise taxes or
make the tax-raising process easier or more difficult as the coefficients for Tax
Increase, Taxes Easier, and Taxes More Difficult are negative and significant.
However, voters appear to be more likely to support bond measures as compared with
some other type of referendum as the coefficient for Bond Proposal is positive and
statistically significant. The coefficients for the other two content variables (Tax
Decrease and Measure Complexity) are not significantly different from zero. The
results for the variables tapping the partisan composition of the legislature (Democratic
378
State Politics & Policy Quarterly 12(4)
Table 2. Logit Analysis of Discretionary Referendum Passage, 1990 to 2008
Independent variable
Tax Increase
Tax Decrease
Bond Proposal
Taxes Easier
Taxes More Difficult
Measure Complexity
Republican Legislature
Democratic Legislature
Divided Government
Legislative Professionalism
Term Limits
Interest Groups
Population Change
Purple State
State Liberalism
Population
Super-Majority
Number of Referendums
Constant
N
Log likelihood
Wald χ2
Referendum passage
−1.06* (0.295)
0.246 (0.223)
0.547* (0.237)
−1.25* (0.467)
−0.711* (0.417)
−0.092 (0.087)
0.646* (0.242)
0.467* (0.210)
−0.309* (0.121)
−2.68* (0.779)
0.143 (0.224)
−0.612* (0.286)
−0.218 (0.196)
−0.284* (0.156)
0.002 (0.006)
0.223 (0.180)
0.473 (0.557)
−0.012 (0.017)
−4.49* (2.02)
1,536
−821.18
84.36*
Notes: Robust standard errors clustering on each state electoral cycle are shown in parentheses.
*p < .05 (one-tailed).
Legislature and Republican Legislature) indicate that as compared with legislatures
characterized by either tied or split partisan control, unified Republican and
Democratic legislatures are systematically more likely to see their referendums pass,
with the larger coefficient for Republican Legislature suggesting that Republican
controlled legislatures are somewhat more successful agenda setters as compared to
their Democratic counterparts.
Last, the coefficients for the control variables (Number of Referendums, SuperMajority, State Liberalism, and Population) are all statistically significant. Thus, it
appears that the passage of referendums is unaffected by citizens’ ideological orientations, by state size, super-majority requirements, or the number of referendums on the
ballot.
Damore et al.
379
Discussion and Conclusion
As we noted at the outset, what we know about agenda setting and policy making via
direct democracy stems from the study of the initiative. Less well understood is the
legislative referendum although legislatively referred ballot measures are in many
ways a potentially more significant policymaking instrument. As compared with initiatives, legislative referendums appear twice as frequently on state ballots, the process
is used in more than twice as many states, and the referendum is more likely to be used
to alter state constitutions.
In examining how the initiative and legislative referendums are used to set the
agenda for policy change in the American states, we find important similarities and
differences between these two forms of direct democracy. In terms of similarities, we
find that both types of ballot measures are constrained by qualification requirements.
The two processes primarily differ with respect to where agenda setting power
resides. On this point, our analysis reaffirms the role of interest groups as a driver for
initiatives (Banducci 1998; Boehmke 2005b), presumably as means to circumvent
interbranch partisan gridlock (Banducci 1998). In contrast, although the ability of
legislators to successfully use the referendum may be constrained by interest group
opposition and electoral uncertainty (as well as policy content) once a proposal
reaches the ballot, we find that referendum use is shaped by the partisan composition
of state legislatures.
Although it may not be all that surprising that the preferences of legislators motivate referendum use and that we observe a greater number of referendums on the
ballot in states where one party controls both legislative chambers, as far as we
know, our analysis is the first to demonstrate this point. More broadly, the significant effects that we find for referendum qualification thresholds and the partisan
composition of state legislatures are consistent with our argument that the referendum can be used to advance nonconsensual policies. Underlying our expectations
and analysis is that legislators will turn to the referendum to achieve a mix of political and partisan goals that cannot be as easily realized through other policymaking
processes. Again, this point may strike some as self-evident. However, conceptualizing the referendum in this manner is at odds with a conventional wisdom that
views the process as largely a mechanical one that is used to advance noncontroversial policies.
Given the dearth of research examining referendum use in the American states,
there are a number of avenues for future research. Two seem particularly worthwhile.
First, although we use bonds and constitutional amendments as a lens through which
to glean how partisan majorities may use the process, these considerations only scratch
the surface of the politics underlying referendum use. As we discuss above, in 2010
the Oklahoma Legislature used a referendum to amend that state’s constitution to prohibit the use of Sharia law in state court decision making. Moreover in 2010, the
380
State Politics & Policy Quarterly 12(4)
Missouri General Assembly referred a measure to the ballot that sought to exempt the
state from certain aspects of the Patient Protection and Affordable Care Act. In 2011,
state legislatures across the country prepared referendums to set the agenda for a
variety of contentious issues. For example, Nevada’s Democratically controlled legislature began the process to remove from that state’s constitution a provision that caps
the rate that mining proceeds can be taxed and constrains how this revenue can be
distributed, whereas the Republican dominated Florida legislature referred a proposal
to the ballot that sought to replicate Missouri’s “Obamacare” measure and another
seeking to make significant reforms to the Florida Supreme Court. Thus, a fruitful
research project might be to assess the multiple dimensions of partisan conflict that
underlie referendum proposals.
Second, with some exceptions (e.g., Kiewiet 1995; Kiewiet and Szakaty1996;
Sears and Citrin 1985), the study of bonds and the interrelationships between bonds
and taxes has received little attention in the political science literature. Given the ubiquity of bond proposals presented to voters via the referendum, the politics underlying
this aspect of state finance may be ripe for further inquiry. For instance, pairing our
findings that in states where Democrats control the legislature and the citizenry tilts
liberally that the number of bond proposals increases with the work of Matsusaka and
McCarty (2001) and Dyck (2009) indicating that the initiative is used more frequently
in such contexts to promote conservative budgetary processes suggests an interesting
dynamic. It may be that Democratic legislators are strategically responding to signals
emanating from the initiative front (Gerber 1996) by using bonds instead of taxes to
expand state governmental capacity. Indeed, although voters are less likely to support
referendums that seek tax increases or that seek to ease the tax-raising process, state
electorates passed more than 80% of all bond proposals referred by state legislatures
between 1990 and 2008.
More generally, considering the interrelationships between citizens’ preferences
and initiative and referendum use suggests an array of interesting possibilities for
examining agenda setting and collective policy making at the state level. Damore
and Bowler (2011), for instance, examine how the initiative and referendum are
used to promote distinctively different reforms to state governance structures with
initiative proponents typically seeking to tighten the leash on policy makers, while
legislators use the referendum to propose additional powers or enhance legislative
autonomy.
In this regard, unlike any other policymaking mechanism, the legislative referendum
straddles two important policy venues—direct democracy and state legislatures—and
as such, lies at the nexus between direct and representative democracy. The referendum
allows elected representatives to forge legislation—and thus set the agenda—yet, it also
requires voter approval. In doing so, the legislative referendum is uniquely situated to
resolve the democratic tension of a largely uninformed electorate that nonetheless
wants control over political outcomes.
381
Damore et al.
Appendix A
Measurement and Data Sources for Independent Variables
Variable name
Variable summary
Dummy variable coded one if a referendum was a bond proposal.
Data from the NCSL Ballot Measure Database.
Dummy variable coded one if both chambers of state legislature
Democratic Legislaturea
are Democratically controlled (excluding ties). Data from state
election returns.
Dummy variable coded one if both chambers of legislature are
Divided Governmenta
controlled by party different from the governor. Data from state
election returns.
Dummy variable coded one if chambers of legislature are
Divided Legislaturea
controlled by different parties. Data from state election returns.
Initiative Qualification
Seven-level measure of state initiative qualification requirements
Difficulty
where larger values indicate a more difficult qualification process.
Data from Bowler and Donovan (2004).
Interest Groups
The log of the total number of interest groups in the state divided
by 2000 state population. Data from Gray and Lowery (2001).
Legislative Professionalisma Interval measure of legislative professionalism where higher values
indicate a greater degree of professionalism. Data from Squire
(2007).
Majority in Lower Chambera The percentage seat advantage of the majority party in the lower
legislative chamber. Data from state election returns.
Majority in Upper Chambera The percentage seat advantage of the majority party in the upper
legislative chamber. Data from state election returns.
Measure Complexity
The number of issue areas associated with a referendum. Data
from the NCSL Ballot Measure Database.
Number of Referendums
Number of referendums appearing on a statewide ballot. Data
from the NCSL Ballot Measure Database.
Natural log of annual state population estimates in hundreds of
Populationa
thousands. Data from the United State Census Bureau.
Biennial change in state population in hundreds of thousands. Data
Population Changea
from the United States Census Bureau.
Prior Amendments
The number of amendments to a state’s constitution prior to
1990. Data from The Book of the States.
Purple State
Dummy variable coded one for states where the major-party
index is between 45% and 55%. 1990–2003 data from Ceaser and
Saldin (2005); data from 2004 to 2008 provided to authors by
Ceaser and Saldin.
Referendum Qualification Three-level measure of state referendum/constitutional
amendment qualification requirements where zero is for states
Difficultyb
that do not require a proposed measure to be supported by
a super-majority or in two successive legislative sessions to
qualify for the ballot, one is for states requiring that a proposed
measure be supported by either a super-majority or in two
Bond Proposal
(continued)
382
State Politics & Policy Quarterly 12(4)
Appendix A (continued)
Variable name
Republican Legislaturea
Republican Party Strengtha
State Liberalisma
Super-Majority
Tax Decrease
Taxes Easier
Tax Increase
Taxes More Difficult
Term Limitsa
Variable summary
successive legislative sessions to qualify for the ballot, and two
is for states requiring that a proposed measure be supported
by a super-majority and in two successive legislative sessions to
qualify for the ballot. Data from The Book of the States.
Dummy variable coded one if both chambers of state legislature are
Republican controlled (excluding ties). Data from state election returns.
State measure of Republican Party strength in the electorate.
1990–2003 data from Ceaser and Saldin (2005); data from 2004
to 2008 provided to authors by Ceaser and Saldin.
Citizen ideology measure developed by Berry et al. (1998) where
larger values indicate a more liberal citizenry. Data from http://
www.bama.ua.edu/~rcfording/stateideology.html
Dummy variable coded one for states that require a supermajority for passage of constitutional amendments. Data from
The Book of the States.
Dummy variable coded one if referendum proposed a tax
decrease. Data from the NCSL Ballot Measure Database.
Dummy variable coded one if referendum proposed to make
the process for raising taxes easier. Data from the NCSL Ballot
Measure Database.
Dummy variable coded one if referendum proposed a tax increase.
Data from the NCSL Ballot Measure Database.
Dummy variable coded one if referendum proposed to make the
process for raising taxes more difficult. Data from the NCSL
Ballot Measure Database.
Dummy variable coded one starting with year that a state’s term
limits for legislative service went into effect. Data from the NCSL
Ballot Measure Database.
NCSL = National Conference of State Legislatures.
a
Variable is time varying.
b
The data used for this measure capture requirements for constitutional amendments advanced
through the referendum process, as opposed to referendums more generally. Thus, although Referendum
Qualification Difficulty is not a perfect indicator of the qualification requirements for all legislative
referendums in the data set, they capture the requirements for the vast majority (two thirds of all
referendums in the data set are constitutional amendments). Moreover, attempts to ascertain the
requirements for referendums that are neither constitutional amendments nor general obligation bonds
by inspecting state constitutions proved fruitless suggesting that these requirements are set by statute.
Appendix B
Alternative Model Specifications
In preliminary analyses, we estimated the models presented in Tables 1 and 2 using a
number of specifications including fixed and random effects specifications. The fixed
effects specification is problematic as in some cases the models failed to converge and
in other instances, either subsets of cases or variables were dropped. When the random
383
Damore et al.
effects specification is used, there are marginal differences in the effects for some of
the control variables. However, the main inferences that we draw from the article are
not sensitive to estimation strategy.
Another issue associated with these data is the cross-case dependency stemming
from the inclusion of multiple observations from the same states. We use robust cluster standard error (RCSE) using state as the clustering criteria; a common remedy in
the literature including the work of Boehmke (2005a), whose work on initiative use we
build on. However, using Monte Carlo simulations of an ordinary least squares (OLS)
model, Harden (2011) demonstrates that bootstrap cluster standard errors (BCSEs)
outperform RCSE and yield estimates that are less likely to be biased. Tables B1 and
B2 provide the results of our models of initiative and referendum use and referendum
passage using the BCSE estimation strategy where state is the clustering criteria and
the model is estimated so that each cluster is resampled with replacement a thousand
times. Across the four usage models, there are some differences between the RCSE
and BCSE estimations. For the models examining all referendums, Term Limits falls
out of significance and Term Limits and Republican Legislature fall out of significance for the bonds model using BCSEs. The most significant difference, although, is
for the initiative model, where the BCSE estimates for Interest Groups are just outside
the .05 threshold (p = .06, one-tailed) for statistical significance. For the passage models, there are no differences between the RCSE and BCSE specifications.
Table B1. BCSE Negative Binomial Regression Analysis of Initiative and Discretionary
Legislative Referendum Use, 1990 to 2008
Independent variables
Divided Government
Divided Legislature
Republican Legislature
Democratic Legislature
Interest Groups
Initiative Qualification Difficulty
Referendum Qualification Difficulty
Majority in Lower Chamber
Majority in Upper Chamber
Legislative Professionalism
Term Limits
Prior Amendments
Republican Party Strength
State Liberalism
Population
Population Change
Constant
N
Log likelihood
Wald χ2
Initiatives
Referendums
Bond proposals
Amendments
0.223* (0.140)
−0.102 (0.222)
—
—
0.831 (0.527)
−0.337* (0.117)
—
—
—
1.19 (1.57)
—
—
−0.009 (0.015)
0.002 (0.009)
0.385 (0.349)
0.188 (7.79)
7.78* (3.72)
240
−498.67
17.53*
0.064 (0.167)
−0.549* (0.333)
—
—
0.279 (0.567)
—
−0.483* (0.288)
−0.818 (0.840)
−0.292 (0.928)
−0.826 (1.93)
0.332 (0.246)
—
0.003 (0.017)
−0.006 (0.011)
0.244 (0.346)
−0.059 (2.02)
3.64 (4.36)
480
−1,064.12
17.61*
0.071 (0.419)
—
−0.642 (0.548)
1.10* (0.645)
0.363 (1.22)
—
—
−1.69 (2.32)
0.820 (2.08)
1.32 (4.92)
0.774 (0.777)
—
0.045 (0.043)
0.051* (0.027)
0.140 (1.03)
0.065 (3.78)
−3.60 (8.16)
480
−365.80
16.15*
0.118 (0.178)
—
0.365 (0.297)
0.429 (0.381)
0.386 (0.661)
—
−0.506* (0.307)
−0.769 (0.904)
−0.603 (1.16)
−1.03 (1.28)
0.478* (0.218)
0.002* (0.001)
0.018 (0.022)
−0.009 (0.010)
0.283 (0.397)
0.083 (3.27)
3.06 (5.08)
480
−970.04
34.27*
Notes: Bootstrapped robust standard errors clustering on each state are shown in parentheses. Dependent
variables are biennial counts.
*p < .05 (one-tailed).
384
State Politics & Policy Quarterly 12(4)
Table B2. BCSE Logit Analysis of Discretionary Referendum Passage, 1990 to 2008
Independent variable
Tax Increase
Tax Decrease
Bond Proposal
Taxes Easier
Taxes More Difficult
Measure Complexity
Republican Legislature
Democratic Legislature
Divided Government
Legislative Professionalism
Term Limits
Interest Groups
Population Change
Purple State
State Liberalism
Population
Super-Majority
Number of Referendums
Constant
N
Log likelihood
Wald χ2
Referendum passage
−1.06* (0.313)
0.246 (0.234)
0.547* (0.241)
−1.25* (0.513)
−0.711* (0.447)
−0.092 (0.090)
0.646* (0.258)
0.467* (0.222)
−0.309* (0.129)
−2.68* (0.882)
0.143 (0.241)
−0.612* (0.308)
−0.218 (1.92)
−0.284* (0.165)
0.002 (0.006)
0.223 (0.198)
0.473 (0.625)
−0.012 (0.018)
−4.49* (2.13)
1,536
−821.18
72.65*
Notes: Robust standard errors clustering on each state electoral cycle are shown in parentheses.
*p < .05 (one-tailed).
Our data also raise concerns about the potential for cross-equation correlation of the
error terms between initiative and referendum use in the 24 initiative states and between
the agenda setting and passage models of the referendum analyses. Specifically, because
the results presented in the first and second columns of Table 1 examining initiative and
referendum use differ not just in terms of their dependent variables, but also the states
being analyzed, these results may obscure differences in referendum use between states
that do and do not have the initiative process. Descriptive statistics offer some support
for such a dynamic as the correlation between the number of referendums and initiatives qualifying for the ballot in the initiative states is .41 (p = .001) and the mean
number of referendums per biennium in the initiative states is 60% greater than in the
noninitiative states (2.42 as compared with 3.92; t = −4.107, p < .001).
If there is dependency between these two agenda setting process, then the error
terms across the equations presented in the first and second columns of Table 1 may
be correlated, resulting in biased estimates. To address this concern, Table B3 presents
the results of a seemingly unrelated estimation of the initiative and referendum models
for the initiative states (excluding Nebraska). This estimation is carried out in three
385
Damore et al.
stages using STATA’s suest command. In the first stage, each model is estimated
separately with the procedure computing the robust estimator of each model’s covariance. In the second stage, both equations are simultaneously estimated with statespecific RCSE (BCSE is not an option with this estimation strategy), as is the
between-model covariances of parameter estimates. In the final stage, the equality of
the coefficients common to each equation is tested to determine their independence.
Specifically, the procedure yields a χ2 test of significance such that a significant
χ2 value means that the hypothesis of the equality of the common coefficients across
models cannot be rejected.
Table B3. Seemingly Unrelated Negative Binomial Regression Analysis of Initiative and
Discretionary Legislative Referendum Use (Initiative States Only), 1990 to 2008
Independent variables
Divided Government
Divided Legislature
Interest Groups
Initiative Qualification Difficulty
Referendum Qualification Difficulty
Majority in Lower Chamber
Majority in Upper Chamber
Legislative Professionalism
Term Limits
Republican Party Strength
State Liberalism
Population
Population Change
Constant
N
Initiatives
Referendums
0.250* (0.135)
0.087 (0.207)
0.855* (0.472)
−0.341* (0.097)
—
—
—
1.14 (1.34)
—
−0.009 (0.016)
0.002 (0.008)
0.403 (0.329)
0.189* (0.083)
7.88* (3.41)
230
−0.059 (0.097)
−0.562* (0.349)
−0.049 (0.562)
—
−0.458* (0.214)
−1.76* (0.727)
−0.040 (0.554)
−0.826 (1.35)
0.129 (0.167)
−0.006 (0.010)
−0.002 (0.010)
−0.174 (0.437)
0.097 (0.105)
2.56 (4.21)
230
Notes: Robust standard errors clustering on each state are shown in parentheses. Dependent variables
are biennial counts.
*p < .05 (one-tailed).
Inspection of the estimates presented in Table B3 indicates no substantive difference from those presented in the first and second columns of Table 1. Specifically, the
coefficients for all of the primary variables of interest (Interest Groups, Divided
Legislature, Divided Government, Initiative Qualification Difficulty, and Referendum
Qualification Difficulty) perform as before and there are minimal differences in the
magnitude of their effects. Testing the equality of the coefficients for the three independent variables of primary theoretical interest (Interest Groups, Divided Legislature,
and Divided Government) indicates that Interest Groups (χ2 = 3.60, p = .06) and
Divided Legislature (χ2 = 3.52, p = .06) are statistically independent of each other,
whereas Divided Government is not (χ2 = 4.52, p = .03).
(continued)
386
State Politics & Policy Quarterly 12(4)
Appendix B (continued)
Specifying an analysis that captures dependency between the referendum agenda
setting and passage models is less clear-cut. The most significant obstacle is that the
model of referendum use presented in the second column of Table 1 and the passage
model presented in Table 2 employ different units of analysis (biennial counts vs.
the individual measure). Each is also specified using a different estimator (negative
binomial vs. logit). Moreover, in contrast to the analysis presented in Table B3 simultaneously comparing referendum and initiative use (two processes that share time and
space), the relationship between referendum qualification and passage is sequential: A
measure has to first qualify for the ballot before it can be either passed or rejected.
Thus, a Heckman selection model, as opposed to a seemingly unrelated estimation
strategy, is a more appropriate choice as the Heckman is designed to account for processes where a selection effect is likely to produce a nonrandomly selected sample.
Failure to account for such a selection effect may yield biased results. The assumptions underlying the Heckman model as applied to our analysis would suggest that
successful agenda setting is determined in part by what occurs at the agenda setting
phase and as consequence, measures that are either passed or rejected at the ballot box
are a nonrepresentative sample of the universe of potential referendums.
We are further constrained in how we model the selection stage by the limitations
of our data. Specifically, our data allow us to construct a dependent variable for the
selection stage that captures between state differences. However, these data do not
allow us to capture within state comparisons (the number of referendums that were
proposed but were rejected within the legislature). Thus, we measure our selection
dependent variable as a dummy variable that is coded one for a biennium in which a
state legislature qualified at least one discretionary referendum and zero for a biennium in which no discretionary referendum was qualified. The dependent variable for
the passage stage of the analysis is a biennial ratio of the number of qualified referendums that passed and is a biennial ratio of the number of qualified referendums that
passed. The Heckman estimates the selection model as a Probit function and the passage model using OLS. All states except for Delaware and Nebraska are included in
the analysis and state-specific RCSEs are used to estimate the model’s standard errors.
The independent variables included in the selection stage are the same as those for
the referendum use model presented in the second column of Table 1. For the second
stage of the analysis modeling passage, the independent variables are recoded to be
consistent with the model’s unit of analysis. This primarily effects the measurement of
the measure content variables. Specifically, Tax Increase, Tax Decrease, Bond
Proposal, Taxes Easier, and Taxes More Difficult are measured for this analysis as a
count of the number of such proposals appearing on the ballot for each state for each
biennium. Measure Complexity is dropped as is Number of Referendums. The end
result is that we are able to specify a selection model that approximates the process of
interest but, because of limitations with the data, we are unable to specify a model that
exactly replicates those presented in the second column of Tables 1 and 2.
387
Damore et al.
The results for the Heckman selection analysis are presented in the first and second
columns of Table B4.The third and fourth columns in Table B4 present the results for
the selection and passage models analyzed separately. For the Heckman analysis, the
estimated correlation between the errors across equations (ρ) is .22 and a Wald test of
the independence of the error terms is insignificant suggesting that the two processes
are statistically independent. Further evidence of this conclusion can be gleaned by
comparing the Heckman selection results with the results when each model is run
separately. The only difference of note is that Term Limits pops into significance when
the passage model is run separately. Otherwise, the results are virtually unchanged.
Thus, within the limitations detailed above, the results of this analysis suggest that the
agenda setting and passage models are statistically independent.
Table B4. Heckman Selection Analysis of Legislative Referendum Use and Passage, 1990 to 2008
Heckman selectiona
Independent variable
Agenda setting
Divided Legislature
Divided Government
Interest Groups
Referendum Qualification Difficulty
Majority in Lower Chamber
Majority in Upper Chamber
Republican Party Strength
Legislative Professionalism
Term Limits
Population Change
State Liberalism
Population
Tax Increase
Tax Decrease
Bond Proposal
Taxes Easier
Taxes More Difficult
Republican Legislature
Democratic Legislature
Purple State
Super-Majority
Constant
N
Log likelihood
Wald χ2
F
R2
−0.261 (0.257)
0.050 (0.182)
−0.185 (0.399)
−0.707* (0.215)
0.136 (0.729)
−0.941 (0.785)
0.029* (0.152)
−0.440 (0.796)
1.16* (0.436)
−0.171 (0.136)
0.001 (0.008)
−0.203 (0.218)
Passage ratio
0.008 (0.043)
−0.122* (0.064)
0.051 (0.214)
−0.095 (0.061)
−0.051* (0.016)
−0.004* (0.001)
−0.013 (0.042)
−0.077* (0.030)
0.007 (0.014)
0.015* (0.008)
0.007 (0.014)
−0.109* (0.053)
−0.027 (0.061)
−0.009 (0.047)
−0.041 (0.049)
0.051 (0.214)
−1.09 (3.40)
−0.085 (0.462)
480b
−370.86
106.41
Probit
OLS
Agenda setting
Passage ratio
−0.261 (0.257)
0.037 (0.181)
0.008 (0.043)
−0.178 (0.399) −0.122* (0.064)
−0.699* (0.208)
0.092 (0.752)
−0.908 (0.801)
0.029* (0.152)
−0.418 (0.769)
0.078 (0.234)
1.14* (0.424) −0.127* (0.062)
−0.185 (0.143) −0.047* (0.016)
0.006 (0.008) −0.004* (0.001)
−0.197 (0.218) −0.019 (0.042)
−0.079* (0.031)
0.009 (0.015)
0.015* (0.008)
−0.044 (0.029)
−0.113* (0.053)
−0.033 (0.064)
−0.009 (0.044)
−0.042 (0.050)
0.032 (0.071)
−0.982 (3.40)
−0.187 (0.486)
480
321
−267.69
43.52*
6.97*
.12
Notes: OLS = ordinary least squares. Robust standard errors clustering on each state electoral cycle are shown in
parentheses.
a
Estimate for ρ for Heckman selection model is .22; Wald test of independence of equations, χ2 = 0.71 (p = .40).
b
A total of 159 censored and 321 uncensored observations.
*p < .05 (one-tailed).
388
State Politics & Policy Quarterly 12(4)
Appendix B (continued)
There are, however, some differences in these results as compared with those presented in the second column of Tables 1 and 2 that merit discussion. For the passage
model, Divided Government and Purple State are statistically insignificant, whereas
Population Change is significant and Interest Groups remain significant, suggesting
some differences in how electoral uncertainty affects passage. Perhaps the most interesting difference though is the performance of Divided Legislature, which is insignificant when the dependent variable is measured not as the number of qualified
referendums, but by whether any referendum qualified or not (note that Referendum
Qualification Difficulty remains significant). Substantively, this difference indicates
that partisan legislative majorities matter in terms of the number of referendums put
before voters, but that qualification thresholds are key to differentiating legislatures
that are able to use the referendum to agenda set for policy change from those that
are not.
Acknowledgments
We are grateful to Nate Monroe for helpful comments and suggestions. Thanks, too, to Megan
Chavez, Albert Escamilla, and Ellen Moule for assistance with data collection. We would also
like to thank the National Conference of State Legislatures and Jennie Bowser for providing the
initiative and referendum data.
Authors’ Note
A previous version of this research was presented at the Annual Meeting of the Midwest
Political Science Association, Chicago, IL, April 2 to 5, 2009. Any errors in interpretation are
the responsibility of the authors.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship,
and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of
this article.
Notes
1. There is a well-developed literature on referendum use outside the United States (see LeDuc
2003 for an overview and comparative analysis). In the American context, the legislative
referendum is often noted in analyses (e.g., Key and Crouch 1938; Magleby 1984) focusing
on how direct democracy is used by citizens to either propose policy changes (the initiative)
or invalidate legislative actions (the popular referendum).
2. The Delaware General Assembly can unilaterally amend that state’s constitution by approving a proposed amendment in two consecutive sessions.
Damore et al.
389
3. State public corporations may issue moral obligation bonds. These bonds differ from general obligation bonds in two ways. First, moral obligation bonds are backed by income from
the project being financed instead of a state government’s full-faith-and-credit promise to
use its taxation powers to pay off the bondholders. Second, moral obligation bonds are
issued without voter approval (see Markham 2001).
4. The Vermont Constitution is by far the most difficult to alter as amendments can originate only in the Senate every 4 years (prior to 1974 amendments could be proposed every
10 years). The proposal must then be supported by two thirds of the Senate and a majority
of the House. Any amendment that passes both chambers must be passed a second time by
majority votes in each chamber after a newly elected legislature is seated before it can be
brought to the electorate for ratification. As a consequence, the Vermont Constitution has been
amended just 26 times; a far cry from the 800 plus amendments to the Alabama Constitution.
5. Between 1990 and 2008, 74% of referendums and 44% of initiatives passed; see also
Magleby (1984) on this point.
6. There is an extensive literature stemming from Romer and Rosenthal’s (1979) agenda
setting model examining passage of public school budget referendums (e.g., Bergstrom,
Rubinfeld, and Shapiro 1982; Brokow, Gale, and Merz 1990; Priest and Fox 2005; Tedin,
Matland, and Weiher 2001) and other local government referendums, such as flood control (Shabman and Stephenson 1994), hospital construction (Fort 1988), and public works
(Martinez-Vazquez 1981). The primary hypotheses of interest in this literature are assessing
whether the agenda setter extracts greater spending than the median voter’s ideal (see also
Figlio and O’Sullivan 2001) and whether voters’ decision making is driven by self-interest.
Although informative, this vein of research focuses on local, as opposed to statewide referendums, and in the case of school finance referendums, the referendums are required,
which allows the agenda setter to make take-it or leave-it offers to voters; a condition that
does not apply to legislative referendums.
7. Descriptive data between 1990 and 2008 support this perspective. Across the 49 states that
use the referendum, only 60 of the 617 referendums coded as tax and revenue related by the
National Conference of State Legislatures (NCSL) proposed tax increases (148 proposed
tax decreases, 25 sought to make the tax raising easier, and 24 sought to make it more difficult). In contrast, there were 268 referendums proposing bonds during this time period with
77% of such measures being proposed by Democratic legislatures. Because legislators may
also pursue tax increases through normal legislative processes, these data do not allow us to
assess whether legislators use general obligation bonds, which must be popularly ratified,
as a substitute for tax increases, which may or may not be put before the voters, to finance
expansions of state government. These data do, however, indicate that legislators seldom
use the referendum to “pass the buck” for tax increases perhaps because the passage rate for
these measures is only 50%. In comparison, more than 80% of bond measures during this
period passed.
8. For the time period under investigation, the NCSL database contains 2,386 ballot measures
(748 initiatives, 31.35%; 1,544 legislative referendums, 64.71%; 49 popular referendums,
2.05%; and 45, 1.89%, coded as other).
9. In all but two instances, the NCSL codes these measures as “other.”
390
State Politics & Policy Quarterly 12(4)
10. The performance for Referendum Qualification Difficulty for this analysis should be interpreted with a caveat. As we note in Appendix A, the data used to measure the variable capture requirements for constitutional amendments advanced through the referendum process,
as opposed to referendums more generally. At the same time, given that two thirds of all
referendums in the data set are constitutional amendments, the variable does capture the
qualification requirements for the vast majority of referendums.
11. This analysis, too, is limited to discretionary referendums. Five referendums from Wyoming are excluded because we were unable to determine the outcome, and also excluded
is the Mississippi flag referendum where voters were asked to choose among flag designs
instead of vote “yes” or “no.”
References
Banducci, Susan A. 1998. “Direct Legislation: When Is It Used and When Does It Pass?” In
Citizens as Legislators: Direct Democracy in the United States, eds. Shaun Bowler, Todd
Donovan, and Caroline Tolbert. Columbus: Ohio State University, 109-131.
Bergstrom, Theodore C., Daniel L. Rubinfeld, and Perry Shapiro. 1982. “Explaining Voter
Behavior toward Local School Expenditures: The Impact of Attitudes.” Economics of Education Review 9(1): 31–45.
Berry, William D., Evan J. Ringquist, Richard C. Fording, and Russell L. Hanson. 1998. “Measuring Citizen and Governmental Ideology in the American States, 1960-93.” American
Journal of Political Science 41(1): 327–48.
Besley, Timothy, and Anne Case. 2003. “Political Institutions and Policy Choices: Evidence
from the United States.” Journal of Economic Literature 41(1): 7–73.
Boehmke, Frederick J. 2005a. The Indirect Effect of Direct Legislation: How Institutions Shape
Interest Groups Systems. Columbus: Ohio State University.
Boehmke, Frederick J. 2005b. “Sources of Variation in the Frequency of Statewide Initiatives:
The Role of Interest Group Populations.” Political Research Quarterly 58(4): 565–75.
Bowler, Shaun, and Todd Donovan. 1998. Demanding Choices: Opinion and Voting in Direct
Democracy. Ann Arbor: University of Michigan.
Bowler, Shaun, and Todd Donovan. 2004. “Measuring the Effect of Direct Democracy on State
Policy: Not all Initiatives Are Created Equal.” State Politics & Policy Quarterly 4(3): 345–63.
Brokow, Alan J., James R. Gale, and Thomas E. Merz. 1990. “The Effect of Tax Price on Voter
Choice in Local School Referenda: Some New Evidence from Michigan.” National Tax
Journal 43(1): 53–60.
Camobreco, John F. 1998. “Preferences, Fiscal Policies, and the Initiative Process.” Journal of
Politics 60(3): 819–29.
Ceaser, James W., and Robert P. Saldin. 2005. “A New Measure of Party Strength.” Political
Research Quarterly 58(2): 245–256.
Colantuono, Michael G. 1987. “The Revision of American State Constitutions: Legislative Power,
Popular Sovereignty, and Constitutional Change.” California Law Review 75(4): 1473–512.
Damore, David F., and Shaun Bowler. 2011. “Governance Reform via Direct Democracy: Comparing the Referendum and Initiative Processes.” Paper presented at the Annual Meeting of the
Midwest Political Science Association, Chicago, IL. March 31-April 3.
Damore et al.
391
Damore, David F., and Stephen P. Nicholson. 2011. “The Etiology of Interest Mobilization in
Direct Democracy Elections.” Paper presented at the Annual Meeting of the Southwestern
Social Science Association, Las Vegas, NV. March 16-19.
Dyck, Joshua J. 2009. “Initiated Distrust: Direct Democracy and Trust in Government.” American
Politics Research 37(4): 539–68.
Figlio, David N., and Arthur O’Sullivan. 2001. “The Local Response to Tax Limitation Measures: Do Local Governments Manipulate Voters to Increase Revenues?” Journal of Law &
Economics 44(1): 233–57.
Fisch, William B. 2006. “Constitutional Referendum in the United States of America.” The
American Journal of Comparative Law 54(Fall): 485–504.
Fort, Rodney. 1988. “The Median Voters, Setters, and Non-Repeated Construction Bond Issue.”
Public Choice 56(1): 213–31.
Gazey, Penelope J. 1971. “Direct Democracy—A Study of the American Referendum.” Parliamentary Affairs 24(Spring): 123–39.
Gerber, Elisabeth. 1996. “Legislative Response to the Threat of Popular Initiatives.” American
Journal of Political Science 40(1): 99–128.
Gerber, Elisabeth. 1999. The Populist Paradox: Interest Group Influence and the Promise of
Direct Legislation. Princeton: Princeton University Press.
Gray, Virginia, and David Lowery. 2001. “The Expression of Density Dependence in State
Communities of Organized Interests.” American Politics Quarterly 29(4): 374–91.
Harden, Jeffrey J. 2011. “A Bootstrap Method for Conducting Statistical Inference with Clustered Data.” State Politics & Policy Quarterly 11(2): 223–46.
Key, V. O., and Winston W. Crouch. 1938. The Initiative and Referendum in California. Berkeley:
University of California Press.
Kiewiet D. Roderick. 1995. “Constitutional Limitations on Indebtedness: The Case of California.”
In Constitutional Reform in California, eds. Bruce Cain and Roger Noll. Berkeley: Institute of
Governmental Studies, 377-397.
Kiewiet, D. Roderick and Kristin Szakaty 1996. “Constitutional Limitations on Borrowing:
An Analysis of State Bonded Indebtedness.” Journal of Law, Economics, & Organization
12(1): 62–97.
King, James D. 2000. “Changes in Professionalism in U.S. State Legislatures.” Legislative
Studies Quarterly 25(2): 327–43.
Kousser, Thad. 2005. Term Limits and the Dismantling of State Legislative Professionalism.
New York: Cambridge University.
Krislov, Marvin, and Daniel M. Katz. 2008. “Taking State Constitutions Seriously.” Cornell
Journal of Law and Public Policy 17:295–342.
Lascher, Edward L. Jr., Michael G. Hagen, and Steven A. Rochlin. 1996. “Gun Behind the
Door? Ballot Initiatives, State Policies, and Public Opinion.” Journal of Politics 58(3):
760–75.
Lax, Jeffrey, and Justin Phillips. 2009. “Gay Rights in the States: Public Opinion and Policy
Responsiveness.” American Political Science Review 103(3): 367–86.
LeDuc, Lawrence. 2003. The Politics of Direct Democracy: Referendums in Global Perspective. Toronto, Ontario, Canada: University of Toronto Press.
392
State Politics & Policy Quarterly 12(4)
Magleby, David B. 1984. Direct Legislation: Voting on Ballot Propositions in the United States.
Baltimore: Johns Hopkins Press.
Markham, Jerry W. 2001. A Financial History of the United States, Volume 3: From The Age of
Derivatives into the New Millennium. (1970-2001). Armonk, NY: M.E. Sharpe.
Martinez-Vazquez, Jorge. 1981. “Selfishness versus Public “Regardingness” in Voting Behavior.” Journal of Public Economics 15(3): 349–61.
Matsusaka, John G. 1992. “Economics of Direct Legislation.” The Quarterly Journal of Economics 107(2): 541–71.
Matsusaka, John G. 2001. “Problems with a Methodology Used to Evaluate the Voter Initiative.” Journal of Politics 63(4): 250–1256.
Matsusaka, John G. 2004. For the Many or the Few. Chicago: University of Chicago.
Matsusaka, John G. and Nolan M. McCarty. 2001. “Political Resource Allocation: Benefits
and Costs of Voter Initiatives.” Journal of Law, Economics, & Organization 17(2): 413–48.
Moncrief, Gary F., Richard G. Niemi, and Lynda W. Powell. 2004. “Time, Term Limits, and
Turnover: Trends in Membership Stability in U.S. State Legislatures.” Legislative Studies
Quarterly 29(3): 357–81.
Nicholson, Stephen P. 2005. Voting the Agenda: Candidates, Elections, and Ballot Propositions. Princeton: Princeton University Press.
Price, Charles M. 1975. “The Initiative: A Comparative State Analysis and Reassessment of a
Western Phenomenon.” Western Political Quarterly 27(2): 243–62.
Priest, Thomas B., and Linette P. Fox. 2005. “Minority Support for School Bonds in CharlotteMecklenburg, A Cautionary Note.” Education and Urban Society 27(2): 192–203.
Romer, Thomas, and Howard Rosenthal. 1979. “Bureaucrats versus Voters: On the Political
Economy of Resource Allocation by Direct Democracy.” Quarterly Journal of Economics
93(4): 563–87.
Schmidt, David. 1989. Citizen Lawmakers: The Ballot Initiative Revolution. Philadelphia: Temple
University Press.
Sears, David O., and Jack Citrin. 1985. Tax Revolt: Something for Nothing in California
(Enlarged Edition). Cambridge: Harvard University.
Shabman and Kurt Stephenson. 1994. “A Critique of the Self-Interested Model: The Case of a
Local Single Issue Referendum.” Journal of Economic Issues 28(4): 1173–86.
Smith, Daniel A., and Dustin Fridkin. 2008. “Delegating Direct Democracy: Interparty Legislative Competition and the Adoption of the Initiative in the American States.” American
Political Science Review 102(3): 333–50.
Squire, Peverill. 1992. “Legislative Professionalization and Membership Diversity in State
Legislatures.” Legislative Studies Quarterly 17(1): 69–79.
Squire, Peverill. 2007. “Measuring Legislative Professionalism: The Squire Index Revisited.”
State Politics & Policy Quarterly 7(2): 211–27.
Tedin, Kent L., Richard E. Matland, and Gregory R. Weiher. 2001. “Age, Race, Self-Interest,
and Financing Schools through Referenda.” Journal of Politics 63(1): 270–94.
Tolbert, Caroline J. 1998. “Changing Rules for State Legislatures: Direct Democracy and Governance Policies.” In Citizens as Legislators: Direct Democracy in the United States, eds.
Shaun Bowler, Todd Donovan, and Caroline J. Tolbert. Columbus: Ohio State University,
171-190.
Damore et al.
393
Tolbert, Caroline J. 2003. “Direct Democracy and Institutional Realignment in the American
States.” Political Science Quarterly 118(3): 467–89.
Zimmerman, Joseph. 2001. Referendum: The People Decide Public Policy. Westport. CT:
Praeger.
Bios
David F. Damore is an associate professor of political science at the University of Nevada,
Las Vegas and a Nonresident Senior Fellow in Governance Studies at the Brookings Institution.
His primary research interests are the study of campaigns and elections and policy making.
Damore’s research has been published in the Journal of Politics, Political Behavior, Political
Research Quarterly, Journal of Political Marketing, and American Politics Quarterly.
Shaun Bowler is a professor of political science at the University of California, Riverside.
Professor Bowler’s research interests include comparative electoral systems and voting behavior.
His previous work on direct democracy include Demanding Choices: Opinion Voting and
Direct Democracy (with Todd Donovan) and Citizens as Legislators (with Todd Donovan and
Caroline Tolbert).
Stephen P. Nicholson is associate professor of political science and (by courtesy) cognitive
science at the University of California, Merced. Nicholson studies American politics, with a
focus on mass political behavior, political psychology, and direct democracy. His research has
been published in the American Journal of Political Science, Journal of Politics, Cognitive
Science, British Journal of Political Science, Political Behavior, Political Research Quarterly,
and PS: Political Science & Politics.