One of the notable features of the current presidential nominating process is
its volatility. Frontrunners stumble unexpectedly, relative unknowns enjoy
sudden success, and seemingly strong contenders fail to gain traction during
the campaign. Such twists and turns in candidate fortunes are inevitable
given voters' lack of information at the start of the primary season (Bartels 1988; Geer 1989;
Popkin 1991), and they consequently offer
scholars an excellent opportunity to study campaign effects. Recently, we
have begun to use the New Hampshire primary as a laboratory to study the
impact campaigns have on voters (Vavreck et al.
2002; Fowler et al. forthcoming). Our
current work focuses on changes in voter attitudes over the course of a
campaign using a four-wave panel survey of voters in the 2000 primary
election. For this paper, we examine the impact of New Hampshire's open
primary rules to compare the behavior of registered partisans and undeclared
voters. Despite conventional wisdom that undeclared voters make primaries
more volatile, we find few differences in the way the two groups of voters
responded to campaign stimuli. In New Hampshire, at least, undeclared voters
were merely sheep in wolves' clothing.
The Trend Toward Open Primaries
In recent years, states have changed electoral rules to admit voters who
do not register with a party. Among the 44 states with some sort of
primary in at least one party, 29 now have either open or modified open
contests.1
This paper is a condensed
version of a paper presented at the Midwest Political Science
Association annual meeting in April 2002. The authors would like
to thank Tami Buhr for her role in designing and administering
the surveys on which this study is based. They would also like
to thank Jeffrey B. Lewis for advice on organizing the dataset.
Dartmouth students Jason Rubenstein, James Pfadenhauer, Timothy
Waligore, Robert Gienko, Christopher Smith, Alice Gomstyn,
Joshua Lozman, Kathleen Reeder, and Rohin Dhar provided valuable
research assistance.
More important is the fact that
the primary schedule has shifted to concentrate more open and modified
open primaries in the early weeks of the campaign. In 2000, two-thirds
of the open primaries and five-sixths of the modified open primaries
took place in weeks 1–7, while half of the closed primaries occurred in
weeks 9–20.
Many political observers disagree about the merits of allowing undeclared
voters into party primaries. Critics of open primaries con tend that
they prevent parties from framing a coherent message and invite meddling
from the opposition. They argue further that voters who lack a long-term
attachment to a party will be less knowledgeable about its potential
nominees and less committed to its viability. In contrast, proponents of
open primaries cite democratic norms that all voters should participate
in the selection of presidential nominees. Some advocates further
contend that open primaries enable parties to expand their base by
bringing new voters into the fold.
This debate gained significance in the 2000 primary as John McCain posted
early successes against George Bush in the Republican primaries,
seemingly with the aid of undeclared voters. In addition, New
Hampshire's contest was often described as a battle between McCain and
Democrat Bill Bradley for the undeclared vote. Paolino and Shaw (2001) have demonstrated that McCain's
campaign could not succeed over the long run, but the question remains
whether the introduction of undeclared voters added a new element of
uncertainty to an already unpredictable process.
Do Undeclared Voters Make Primaries More Volatile?
At the heart of the debate over open primaries is a presumption among
political observers that partisan registrants and undeclared voters have
different attitudes and information about politics and behave
differently with respect to elections. Findings from early research on
voting behavior support this view (c.f. Campbell et
al. 1960). Yet, subsequent research indicates that primary
voters and general election voters are quite similar (Geer 1989; Norrander 1992), and
that independent voters behave very much like weak partisans (Dennis 2002; Keith et al.
1992). There is little systematic evidence about the
differences between registered partisans and undeclared voters in
primaries, however. Instead, studies of primary elections have used
party identification rather than registration to analyze voters (Bartels 1988; Geer
1989; Norrander 1992). This
approach typically omits pure independents and assumes that weak
partisans and independent leaners vote in the party with which they
identify.
Table 1 indicates that 20% of undeclared New
Hampshire voters self-identified as pure independents. In addition, more
than 50% identified as partisan leaners. These voters may feel the pull
of partisanship less keenly at the beginning of the election season,
during the New Hampshire primary, for example, than at its ending.
Furthermore, undeclared voters may differ in their receptivity to
campaign stimuli. If they pay less attention to politics or have less
information about candidates, they may be more sensitive to campaign
effects when exposed to them, or may ignore the campaign altogether and
show no effects.
Most important, unlike registered partisans, undeclared voters must
select a primary before casting their ballot. To vote for John McCain,
they must first request a Republican ballot. The order of these
decisions has an important methodological implication because it
ultimately truncates the vote choice. Conventional regression techniques
consequently overestimate the likelihood that any voter chooses Bush
when Republican primary voters are the sole basis for calculating the
probabilities. Yet, the political science literature reveals very little
about how the sequencing of voter decisions works. Do voters first pick
the primary of the party to which they most closely identify and then
select their preferred candidate? Does their candidate preference
determine the choice of primary? Does this ordering of choices affect
the outcome of the election? To answer these questions, we employ a
selection model using data from our four-wave panel survey from the 2000
New Hampshire presidential primary.2
Compiled from Ragsdale (1998,
41–43); Federal Election Commission
(2002). Closed primaries limit voting to registered
partisans. Modified open primaries restrict registered partisans
to voting in their party's primary, but permit undeclared voters
to choose a party ballot or register for a party immediately
prior to entering the voting booth. Open primaries involve a
public declaration of party affiliation on election day or allow
voters to choose a party ballot in the voting booth.
In
this model, voters first decide in which primary to vote and then for
whom to vote.
3 These data are uniquely
suited to explore such questions because they track changes in
voters' attitudes during the campaign and enable us to link
voters' preferences about the candidates to their eventual
decisions. The dataset consists of a four-wave panel survey
conducted via telephone, with new respondents added in each wave
but the last, which was a post-election wave. The data were
collected by the Nelson A. Rockefeller Center at Dartmouth
College, in collaboration with the Associated Press, and include
2,540 likely New Hampshire voters. Of these respondents, 952 are
in the first wave (October 31–November 3, 1999), 1,055 are in
the second wave (January 9–12, 2000), 985 are in the third wave
(January 23–26, 2000), and 1,589 are in the fourth wave
(February 6–9, 2000). Thus, 1,589 respondents have at least one
pre-election wave interview and post-election interview. We use
this pre-post data set in the analysis that follows. The survey
contained a screen for likely voters that included all
Republicans, Democrats, Libertarians, or undeclared citizens
that stated their intention of voting in any primary. The data
set contains roughly a three-way split among registered
Democrats, Republicans, and undeclared likely
voters.
Measures and Methods
New Hampshire is a good setting for investigating the influence of
undeclared voters in the primary process because citi- zens who
registered as “undeclared” for the 2000 election cycle slightly
outnumbered registered partisans in either party. In addition, the New
Hampshire primary is unusual in its high salience among voters. Turnout
is consistently high; well over 50% in 2000. Finally, the information
environment is extremely rich as candidates, groups, and media lavish
resources on the contest (Vavreck et al. 2002;
Fowler et al. forthcoming). In short,
undeclared voters have exceptional opportunities to learn about the
candidates in both parties as the primary contest unfolds.
Drawing from the work of Patterson (1980),
Bartels (1988), Geer
(1989), Popkin (1991), Abramson et al. (1992), Norrander (1992), and Vavreck et al.
(2002), we focus on the key variables that appear to
influence voters' decisions in primary elections.4
Since both stages of this system are dichotomous, we use
full-information maximum likelihood to estimate the correlation
between the errors in both equations. We run the
full-information probit selection model once for each party
primary and interact substantively important variables with
whether the respondent was an undeclared registrant.
These include: demographic characteristics, such as age, sex, education
and income; party identification using the National Election Study (NES)
standard seven-point scale; respondents' self-reported party
registration; contact with specific candidates;
5 Questions about issue positions and ideological
placements were not included in the third wave of the survey, so
we do not test for the effect of issues on the choice of primary
and candidate.
and expectations about candidates'
viability and electability.
6 We employ two
dichotomous measures of candidate contact: 1) whether
respondents met a candidate either at a rally or at another type
of event; 2) whether respondents received telephone calls or
mail from a candidate. We have these data broken out
specifically by candidate.
In order to measure
information levels during the primary campaign, we use a respondent's
selfdeclared level of attention to the campaign, including four
categories from “a lot of attention” to “no attention.” We measure party
identification with a branching format of the traditional seven-point
NES self-placement question. We also rely on respondents to indicate
their party registration (Democratic, Republican, Libertarian, or
Undeclared). For voters' expectations about election outcomes, we use
measures of candidate viability and electability. In the model,
substantively interesting variables are interacted with whether the
respondent is an undeclared voter.
7 Viability represents a respondent's dichotomous judgment
about whether a candidate will win the primary election, and
electability is that same judgment for the general
election.
Results8 We initially employed
several other variables in the models. We eliminated variables
measuring uncertainty, the timing of vote decisions, and general
political attention due to very small effects and lack of
significance. We removed our measures of affinity for
candidates, such as candidate traits and thermometer ratings,
for other reasons. The traits questions, which tapped voter
opinions about trustworthiness and leadership characteristics,
were all uniformly high (70 degrees) and showed little variation
across candidates or over time. The thermometer variables
appeared to serve as voters' summary judgments about the
candidates and the voters' intention to support them, rather
than as exogenously determined dispositions toward the
candidates. Although both types of measures proved useful in
assessing the overall levels of support for the candidates, they
masked the impact of other important causes of the vote
choice.
The results of the sample selection model suggest that registering as an
undeclared voter had no direct effect on either the choice of primary
election or the candidate, controlling for many factors.9
We present the results in abbreviated form in
Table 2. Given the fully
interactive form of this model, the large number of
coefficients, and the difficulty with substantive interpretation
of probit selection coefficients, this is the simplest way of
conveying the results. Variables for which declared and
undeclared registrants have different effects are highlighted. A
complete table of results is available from the authors upon
request.
As
Table 2
indicates, there do not appear to be systematic differences between
partisan registrants and undeclared registrants in terms of their
probability of voting in a party primary or of voting for a particular
candidate.
10 The results indicate
that although sample selection effects are present, there is no
bias in the estimation.
Among undeclared voters, 60%
chose to vote in the Republican primary, but among all registered
partisans, a very similar 62% chose the Republican primary. In sum, the
Republican primary was a popular choice for all types of voters.
Direct candidate contact had an effect solely on the vote, as it did in
the 1996 New Hampshire primary (Vavreck et al.
2002).11
The vote choice
analysis is limited to choices over Bush or McCain in the
Republican primary, and Gore or Bradley in the Democratic
primary.
However, there were no differences between
undeclared and registered voters in their response to candidates.
Meeting McCain at a rally or in person increased voters' chances of
voting for him by 9%, while meeting Bush increased voters' chances of
voting for him by 29%. Despite the favorable influence of contact with
Bush, the overall probability of voting for McCain was still higher
because more people supported him overall.
Results from the Democratic choice model were similar. Meeting Gore
raised the probability of voting for him by 27%, and meeting Bradley
raised chances of voting for him by 18%. Interestingly, Bradley did not
make as many visits to New Hampshire as McCain, especially in the late
weeks of the campaign. This may explain why McCain was able to leverage
the increasing probability of voting for him among people who met him
while Bradley was not.
Candidate contact in the form of telephone calls did have a differential
effect on the choice of primary among undeclared voters, however, at
least among Republicans. Registered partisans were 16% more likely to
choose the Republican primary after receiving a call, but undeclared
voters were 9% less likely to select that primary after receiving a call
from a GOP candidate. The ineffectiveness of telephone calls in
mobilizing voters is consistent with our earlier work on the 1996
primary (Vavreck et al. 2002) and Gerber and Green's (2000) study of turnout in
New Haven. Direct mail did produce mobilization effects that were
similar for partisans and undeclared registrants.
Being an undeclared voter, however, did condition the effects of other
important variables. For example, party identification clearly motivated
self-declared partisans to select their party's primary, although it had
no effect over their candidate choice. For registered partisans,
shifting self-identification from Independent to Weak Democrat increased
the probability of voting in the Democratic primary by 24%, holding all
else equal with other variables at their means. In contrast, a similar
shift in party identification among undeclared voters was about 5%
less.
Status as an undeclared voter also conditioned the effect of gender on
the choice of both party primary and candidate. Although women generally
had a higher probability of voting in the Democratic primary, undeclared
women were 10% less likely to vote in the Democratic primary than their
partisan counterparts, all else equal and holding all other variables at
their mean. Undeclared women were not only less likely to choose the
Democratic primary, but were also 19% more likely to select Bush than
their registered counterparts. Whatever mechanism drew undeclared women
away from the Democratic primary in greater numbers than registered
women also seems to have induced them to vote for Bush instead of
McCain.
Although undeclared voters were less constrained by party identification,
our results indicate that McCain would have won the primary if it had
been restricted to Republicans. Among people who self-identified as
Republicans, whether registered or undeclared, McCain won 66% of the
vote.12
Previous work on the 1996
primary indicates that contact and vote intention may be
endogenous, but we have controlled for that here. See Vavreck et al. (2002).
Overall,
McCain derived 31% percent of his support from people who described
themselves as either Independent or some type of Democrat, while only
10% of Bush's support came from this group. Finally, McCain attracted
52% of the undeclared voters, while Bush drew only 19%. The basic story
of the 2000 New Hampshire primary was McCain's high popularity among
many types of voters.
13 We are sensitive
to issues of endogeneity here.
The most intriguing aspect of the New Hampshire primary, and one that we
intend to explore more fully in subsequent work, is the high degree of
volatility among all likely voters in terms of their choice of primary
and candidate. As Figure 1 demonstrates,
between October and January, both Bush and McCain gained voters who
previously thought they might vote for someone else.14
The numbers in boxes are raw numbers of voters. The
first two rows represent people who said they were going to vote
for Bush or McCain. The third row contains people who were
undecided about which Republican they would vote for or who knew
they would vote for a Republican other than Bush or McCain. The
final row contains people who said they would vote in the
Democratic primary or were unsure in which primary they would
vote. The sizes of the arrows are in direct proportion to the
magnitude of the changes in voter support from wave to wave. A
dark arrow represents a change across candidates or primaries of
more than 10 people. Thus, arrows should be read with attention
both to width and color.
What was unusual is that Bush
lost heavily to McCain in this period and that the defections were
mainly from registered Republicans rather than undeclared voters.
15 Seventy-five percent of the people who
move from Bush to McCain between October and January were
registered as Republicans in October. Twenty-five percent were
undeclared.
Some results not presented
in Table 2 warrant a passing mention.
For example, paying attention to the campaign had no effect on
primary choice, but did increase the chances of voting for Gore
among people who selected the Democratic contest for both
registered partisans and undeclared voters. Candidate viability
also showed no effect on primary election or vote choice, but
our measure of electability in November did increase the chances
of voting for that particular candidate in the party primary by
a substantial amount. As in the case of candidate contact,
endogeneity may be at work here.
McCain also gained
votes from people who initially thought they were going to vote in the
Democratic primary or were unsure about which primary they would
choose.
Then, between January and the election in February, both Bush and McCain
surprisingly lost votes as people decided to vote in the Democratic
primary. All of the voters who moved from Bush to the Democratic primary
were undeclared voters; while 83% of those voters who abandoned McCain
also were undeclared. By examining Figure 1
with an eye toward the heavy dark lines, a pattern emerges of arrows
moving away from Bush and McCain. This is exactly the opposite of what
political observers argued happened in New Hampshire. From January to
February, McCain did pick up votes from people who thought they were
going to vote in the Democratic primary, but 72% were registered
Democrats and only 18% were undeclared voters. Again, the conventional
wisdom had McCain and Bradley locked in a battle for undeclared voters,
when it appears that partisans were also receptive targets.
Conclusion
Taken as a whole, there is very little evidence that undeclared voters in
New Hampshire behaved much differently than registered party voters—both
groups gave a victory to John McCain. Although the open primary system
did allow for a lot of movement across candidates and parties, in the
end, most sub-groups of the electorate favored McCain over Bush. In sum,
both critics and advocates of open primaries appear to have exaggerated
the impact of admitting undeclared voters to the primary process.
Primaries are volatile because voters and candidates are operating in a
low information environment in which new facts and new impressions count
heavily. Undeclared voters are subject to all of these uncertainties, to
be sure, but so are registered partisans. Allowing undeclared voters to
vote in nominating elections expands the size of the primary electorate,
but does not seem to make an already unstable situation decidedly
worse.
