Content uploaded by Raul Gomez
Author content
All content in this area was uploaded by Raul Gomez on Jun 19, 2014
Content may be subject to copyright.
This is an Accepted Manuscript of an article published by Taylor & Francis Group in Journal of Elections, Public
Opinion and Parties on 19th March 2013, available online at: http://www.tandfonline.com/
10.1080/17457289.2013.776056."
All that you can(not) leave behind.
Habituation and vote loyalty in the Netherlands.
Raul Gomez1
Abstract
Scholarly work has pointed out that party loyalty develops with age.
However, most of the literature has focused on two-party systems and
employed party identification as dependent variable in order to investigate
this phenomenon. This article sheds more light on how this process takes
place, employing a measure of party utility that is available for all the parties,
and not only for the one that voters feel identified with (if they do at all).
Findings suggest that preferences become more skewed toward the chosen
party as choices are repeated, supporting the presence of some kind of
habituation process in voting behavior even in contexts where the presence
of multiple parties should in principle work against party loyalty. Moreover,
repeated voting for a party is shown to partly explain the effect of age on
loyalty. Habituation is, thus, argued to be essential in order to understand
why volatility is strongly and significantly higher among young voters.
Keywords: habituation, party loyalty, age, voting behavior, party utility, multi-party systems.
1 Introduction
Much theorizing in political science expects choices to be made afresh at each election (eg.
Fiorina, 1981; Riker, 1986; Fearon, 1999; Manin, Przeworski and Stokes, 1999). Yet the first
thing that one notices is that the number of loyal voters tends to always outweigh the number of
switchers. Nearly forty years ago, Butler and Stokes (1974) realized that vote loyalty in the UK
increased dramatically among individuals that had repeatedly chosen the same party on several
Correspondence Address: Dr. Raul Gomez, Department of Politics, University of Liverpool,
Liverpool L69 3BX, United Kingdom. Email: rgomez@march.es. This article makes reference
to supplementary material available on the publisher’s website at <www.doiXXX.org>.
2
occasions. Following McPhee and Ferguson (1962), they called this process ‘immunization’,
suggesting that the repeated experience of older voters with party politics made them less
susceptible to the contagion of political change. But how does this type of ‘immunization’ work?
And to what extent does this concept serve us to understand the development of party loyalty over
the life course?
To date, scholars have presented evidence of the crystallization of party attachments in two-
party systems employing measures of party identification for this purpose. This article is an
attempt to shed more light on this by investigating the impact of inertia on party support and on
subsequent choices in the context of Dutch elections.
1
It employs a measure of party utility that,
unlike party identification, is available for all the parties. Findings are consistent with the presence
of some kind of habituation process in voting behavior. Party support increases after voting for a
party, an effect that is particularly strong when the same choice has been made for at least two
elections. Party loyalty is, therefore, expected to become stronger and crystallize after a number of
elections, and this should explain why older voters are less likely to switch parties. In this regard,
it is shown that repeated voting for a party explains part of the effect of chronological age on vote
stability and is therefore essential in order to understand why young voters are much more
volatile.
In what follows, I first introduce the different approaches to the concept of party loyalty and
the way in which it has been usually dealt with in the political science literature (Section 2).
Theoretical expectations and hypotheses are then introduced in Section 3, while Section 4
elaborates on the data and methods employed in the analyses. Findings are shown and commented
on in Section 5. Finally, Section 6 concludes with a summary of findings and a brief discussion.
2 Dealing with (in)stability: the political science literature.
At first sight, it may not seem striking that voters tend to repeat the same choice over time. After
all, parties that match up to voters’ preferences most of the time should regularly get their support
(Adams, 2001). Yet, the fact that older adults are more likely to repeat their vote (cf. McPhee and
3
Ferguson, 1962; Butler and Stokes, 1974) suggests that there is something more to be explained.
Party choice is not made anew at each election. Rather, citizens carry memories of their past
choices into the voting booth and use prior behavior as a standard against which to evaluate
current choices (Alwin and Krosnick, 1991; Nadeau and Mendelsohn, 1994; Zuckerman, 1989).
Thus, with the passage of time, voting behavior seems to be impacted by some kind of inertia that
leads voters to stick with the same party at most elections.
The fact that support for the chosen party becomes stronger and crystallizes over time may be
explained by the presence of a habituation process in voting behavior. Now, how and why this
process takes place is a different question. In the literature, inertia (that is, the tendency to repeat
one’s behavior) tends to be framed in the context of bounded rationality. Under conditions of
limited time and information, it is a cost-efficient strategy to stick with an alternative that one
knows works reasonably well instead of trying to find the best performing option in each case
(Simon, 1978; Shugan,1980; Payne, Bettman and Johnson, 1993). According to this view,
individuals should try to maximize the accuracy of their choices and minimize the effort put into
the decision-making process. Prior knowledge obtained by experience determines which strategies
are available to a decision maker in her memory, and so, when making new choices, individuals
are likely to repeat the same strategy that they used in the past unless new information is received
that discourages its use (Payne, Bettman and Johnson, 1993). Similar arguments have been used to
explain the presence of habituation in voting behavior, which is often claimed to be a mechanism
that makes the decision-making process of voting less demanding (Aldrich, Montgomery and
Wood, 2011).
For some scholars, however, inertia need not be at odds with full rationality. Individuals may
tend to stick with the choice that they know best rather than switch to another whose performance
is uncertain (Chorus and Dellaert, 2012). The mechanisms may not be very different when it
comes to choosing a party. Individuals choose alternatives based on their positive expectations
about them, but they are uncertain as to how satisfied they would be with their choice had they
chosen any other alternative. And since alternatives that are chosen more than once in a row are
4
precisely those that individuals have had a good experience with, it makes sense to think that
voters should develop stronger preferences for those parties.
In studying the habituation process of voting, most political scientists have focused on turnout
as a dependent variable (eg. Campbell et al., 1960; Green and Shachar, 2000; Plutzer, 2002;
Gerber, Green and Shachar, 2003; Franklin, Lyons and Marsh, 2004; Denny and Doyle, 2009;
Aldrich, Montgomery and Wood, 2011). But inertia, as Campbell (1960) himself pointed out, may
be related not only with the probability to turn out as such, but also with the party that voters
decide to choose. In this regard, much scholarly work has paid attention to the development of
party identification and its reinforcement over time. Drawing on Converse’s (1964; 1969) robust
findings that relate age to the stability of party identification, scholars have provided explanations
for this phenomenon from very different perspectives. Some claim vote stability to be the
byproduct of a psychological process by which people generate attachments to parties which are
then reinforced over the life course (Campbell et al., 1960; Miller and Shanks, 1996). Others,
though, consider party identification to be a ‘running tally’ of voters’ assessments of the
performance of parties (Key Jr., 1952, 1966; Achen, 2002).
From a purely rational point of view, inertia – or habituation – has often been disregarded
because party attachments are expected to change as a consequence of political events and other
short-time factors (Grynaviski, 2006). Achen (1992, 2002), however, justifies the presence of
inertia in that the marginal effect of new information decreases with age, which would explain
why older individuals do not change their partisanship so often and are, therefore, expected to
keep stable preferences. Green2002, Green, Palmquist and Schickler (2002), on the other hand,
put this rational-revisionist explanation into question. In their opinion, identification concerns the
way in which people think of themselves rather than just their evaluations about parties. They
show that assessments of parties’ performance are more likely to change over time than is party
identification. Moreover, they argue that there is no reason to think that older people learn less
from new events than younger adults. In their view, as soon as there is some degree of political
change, assuming that older people do not learn from new events would imply that they prefer to
give more weight to obsolete information. And that, they argue, is irrational.
5
More recently, scholars have argued that the reason why preferences becomes stable over time
may also be related to some sort of strengthening effect of voting behavior. In their opinion, the
very act of voting impacts on party support by strengthening the beliefs that voters have about the
parties and leaders that they have voted for (Meredith, 2009; Mullainathan and Washington, 2009;
Dinas, 2010, but see contradicting findings in Elinder 2012). Their argument is that voters adapt
their preferences to their own behavior and end up developing stronger attachments to the chosen
party. Several alternative mechanisms may, in principle, cause this phenomenon (for a more
detailed account, see Dinas, 2010). For one thing, elections may operate by increasing the
information of voters about political parties, reinforcing the strength of their preferences toward
the chosen alternative. For another, based on Festinger’s (1957) theory of cognitive dissonance,
elections might in turn trigger a psychological mechanism that makes young voters adapt their
preferences to their final choice. An alternative explanation, though, is that voters simply infer
their attitudes from their past behavior (Bem, 1972).
The treatment given to inertia in the literature has, nevertheless, certain limitations. First of all,
employing party identification as a proxy for a person’s preferences may be problematic on
several grounds. Party identification is often measured in a unidimensional way. Either
identification is asked solely for the party that voters feel identified with (if any), or these are
asked to place themselves using a unidimensional scale with one party at each extreme. By
definition, and regardless of the way in which it is measured, party identification is an ipsative
concept (Converse, 1964, 1969). This means that identification with one party excludes
identification with another, and so one unit change in voters’ preferences for one of the parties is
expected to produce a change by the same amount against its competitors. Downs (1957),
however, argued that voters get different utilities from each of the parties and decide to vote for
the one that they prefer the most. As utilities are not exclusive, voters may show particularly high
or low preferences for more than one party, as well as they may even have more than one first
preference. As useful as the concept of party identification may be in order to study vote choice, it
gives us very little information about party utilities of different alternatives, which is of especial
concern in multi-party systems. An additional limitation of previous research is precisely the
6
overwhelming focus on two-party systems of studies aiming to investigate the presence of
habituation processes in voting behavior. Skeptics may, nevertheless, argue that if there is one
particular system where habituation is likely to be found that is one where the number of
alternatives is small and, therefore, the chances to feel attracted to another party are lower. On the
contrary, the presence of more than two parties makes it easier for voters to find other attractive
alternatives and increases electoral volatility (Bartolini and Mair, 1990). This means that multi-
party systems can provide us with a more critical test of the presence of habituation processes.
Lastly, the literature has provided very little evidence of the extent to which inertia is a useful
concept for understanding the effect of age on party loyalty. In this paper, I intend to overcome
these limitations employing data from a multi-party system (the Netherlands) and a measure of
party support that covers all the alternatives in the system. But before elaborating on this, let me
first turn to my theoretical expectations.
3 Theoretical expectations.
What inertia implies is that individuals increasingly favor the party that they have repeatedly
chosen over any other alternative. Arguably, there are two complementary mechanisms behind
this sort of habituation process. The first mechanism is reinforcement. Voters choose parties that
they strongly prefer, but their preferences are reinforced with every choice they make. Thus, as
choices are repeated, support for the chosen party will be strengthen, which at the same time
decreases the probability of choosing other alternatives. The second mechanism is crystallization.
It implies that reinforcement is not a never-ending process. On the contrary, after voting for the
same party on several occasions, support for that party should reach a maximum level and stop
growing (or do it more smoothly). This may well be what is behind Butler and Stokes’ (1974)
‘immunization’ process, according to which young adults’ voting behavior becomes very stable
after repeating the same choice for three consecutive elections.
2
Unfortunately, the data employed
here does not enable us to follow voters for more than a couple of elections, which makes the
crystallization mechanism difficult to test. But it is still possible to investigate whether support for
7
a certain party is reinforced by the act of voting for it. So, the first hypotheses may be formulated
in these terms:
H1: Support for chosen parties will increase more than support for parties that were not
chosen.
H1a : Repeated voting for a party will increase support for that party further, at least up until a
certain number of repetitions.
Now, in order to fully assess the relevance of the habituation process, it is important to
investigate to what extent it helps us to understand why loyalty is higher among older voters. In
principle, the effect of chronological age on the likelihood to switch might be due to other
processes different from inertia that are also associated with the passage of time. Scholarly
literature has pointed out that, as voters grow older, their social situation stabilizes and clearer
attitudes and partisan identities emerge (Dalton 2000, p.30-31; Evans 2004, p.176-177; Schmitt-
Beck, Weick and Christoph 2006). We may therefore expect the aging process to be correlated
with other variables such as, for example, a person’s education, political knowledge and interest.
To what extent is thus inertia necessary for understanding vote stability? Answering this question
is tremendously important because if the processes of reinforcement and crystallization operate
over time in such a way as to lead more experienced voters to stick with a party, then most of the
volatility that we observe in election results must be caused by the action of younger voters.
In the context of turnout research, Strate et al. (1989, p.443-5) argue that, if all the variables
that are usually employed as a surrogate for age did really provide the reason why age affects the
likelihood of voting, then age would lose significance in all those models that control for them.
But the fact that this does not happen is a signal that there is something else about age that is not
being taken account of. The same argument may be used here. It is obvious that if all those
variables mentioned earlier do not render age insignificant when explaining vote loyalty (and they
never do), there must be something else that does. This ‘something else’ might well be the
habituation process, as the reason why older voters tend to be more loyal than the rest may be
8
strongly related to the fact that, if only for probabilistic reasons, they are more likely to have
repeated the same choice several times. If inertia is essential in order to understand the tendency
for older voters to remain loyal to a party, then including this variable together with other factors
associated with the aging process should render age redundant in a model of party loyalty. I
therefore hypothesize that:
H2 : Young voters are significantly more likely to switch their vote.
H3 : The effect of age is explained to a large extent by the accumulation of inertia. So, adding
inertia as measured by the accumulation of identical choices to a model of vote loyalty should
greatly diminish the effect of age.
4 Data and methodology.
In order to test these hypotheses, I will use the four panel surveys that are publicly available to
date as part of the Dutch Parliamentary Election Studies (DPES). The elections with available
panel studies are 1981-1984-1986, 1986-1989, 1989-1994 and 2002-2003 (van der Eijk, Irwin and
Niemöller, 1997b,a; Anker and Oppenhuis, 1997; Irwin, van Holsteyn and den Ridder, 2005).
Most of these panel studies have three waves. The first wave is a pre-election survey. Respondents
were then re-interviewed after that election (second wave) and once again after the following
election (third wave). The only exception to this rule is the 1981-1984-1986 study, which covers
five different waves, as respondents were also interviewed before and after the 1981 election.
3
Studies provide information on vote choice at three consecutive elections: the most recent one (t),
the previous one (t-1) and the one before that (t-2).
4
But, why the focus on the Dutch case? The reasons for doing so are two. First of all, Dutch
studies are the only panels that systematically provide a non-ipsative measure of electoral utility
for at least all the parties represented in the Parliament.
5
In addition, the Netherlands as a country
presents some particularities that makes it suitable as a case for study. Let me focus on this second
reason first and come back to the other one later. Dutch levels of electoral volatility –in the terms
9
defined by Pedersen (1983) – are clearly above the average of other established democracies since
the 1990s, as the Netherlands is the country where electoral volatility has increased the most in
recent decades. The 1994 general election marked an inflection point in the recent history of
Dutch politics. The two parties in the coalition government (the Labor Party and the Christian
Democratic Appeal) experienced a landslide defeat, losing 8% and 13.1% of the votes
respectively. For the Christian Democrats this loss meant much more than a simple defeat, as for
the first time since 1918 they were not pivotal and were excluded from the new coalition
government that was formed after the elections. The consequences for the party system went even
further. The average index of fragmentation from 1977 to 1989 in the Netherlands was 3.9
parliamentary parties
6
In the period between 1994 and 2003, the average went up to 5.3.
Moreover, in the 1994 election the effective number of parties rose by 1.7 points from 3.75 in
1989. Thus, the day after that election voters found themselves in a different scenario, with a more
fragmented parliament and a somewhat different party system. About 30% of the respondents in
the post-election survey of 1994 declared themselves to have voted for a party that they had not
voted for in at least one of the previous two elections. Arguably, many of them had to re-evaluate
whether they would stay with the newly chosen party or whether they would rather switch back to
their former choice. Indeed, it looks like political enterpreneurs took advantage of this new
scenario, judging from the sudden changes in the vote share that many parties would continue to
experience, together with the emergence and disappearance of new parties that marked the
beginning of the 2000s. It is not the intention of this paper to look into the intricacies of these
changes, but to focus on the power of inertia even in a changing scenario such as this one. With
most institutions remaining constant, the Netherlands provides an excellent opportunity to test the
habituation process of voting under very different scenarios of volatility. In fact, if habituation is
still found to operate during the years where instability is pervasive and most people seem to have
had incentives to change their vote, that would provide a very robust test in favor of the
mechanisms that are proposed here.
The other reason that justifies the selection of Dutch data is even more important for the aim
of this study. Since 1982, Dutch surveys contain a question that has more recently been asked in
10
many other survey studies in Europe
7
and employed as both dependent and independent variable
in much published work (see van der Eijk et al., 2006, for some examples): the propensity to vote
for a given party, which, in its English translation, looks like this:
“Some people are quite certain that they will always vote for the same party. Others
reconsider in each case to which party they will give their vote. I shall mention a number of
parties. Would you indicate for each party how probable it is that you will ever vote for that
party? ”
A card is presented, where Category 1 is labeled ‘I will certainly never vote for this party’ and
category 10 is labeled ‘I will certainly vote for this party at some time’
The question intends to be an empirical observation of the (Downsean) electoral utility that
each voter would obtain from voting for each political party, and it has several advantages that are
essential for the purpose of this study. First, the measure is available for every alternative,
providing information with regard to all the parliamentary parties (and not only with regard to the
one that voters feel more attached to, as is often the case with standard measures of party
identification). Secondly, these utility measures are non-ipsative. That is, they do not sum to a
fixed total. Respondents may, therefore, declare themselves very attracted to several parties, or to
be not at all attracted to any. A third important advantage of this measure is that it performs very
well in predicting vote choice. Van der Eijk et al. (2006) have shown that the percentage of
correctly predicted cases derived from using actual choice as the dependent variable is not very
different from the one obtained when declared utility is employed instead. Moreover, the
percentage of respondents that give the highest score to the party that they actually chose is 93%
and above in surveys conducted in the Netherlands and elsewhere (Tillie, 1995; van der Eijk and
Franklin, 1996; van der Eijk, Franklin and van der Brug, 1999). Electoral utilities measured this
way do also perform much better in terms of matching between choice and score than other non-
ipsative measures such as thermometer scores commonly used in analyses of directional voting
(eg. Rabinowitz and Macdonald, 1989; Macdonald, Listhaug and Rabinowitz, 1991). Moreover,
11
they have been demonstrated not to be an artifact of a person’s recent choice and to meet specific
requirements that are necessary for their validity as measures, such as an internal level of
measurement and no loss of information because of the limited resolution of categories (Tillie,
1995).
Party utilities overcome the limitations of previous research by letting us use information on
voters’ declared support for all the parties. Here, they will be used in several ways. First, utilities
will be employed to test whether parties that have been voted for yield more utility by the time
that the next election takes place.
8
Second, utilities will then be used to investigate whether this
effect is reinforced for voters that had already made the same choice in the past. After this, they
will also be employed as controls in yet another model of vote loyalty aimed at showing that
inertia does not become redundant when other variables are introduced into a regression equation
predicting vote choice, but that, by contrast, the effect of age does when an indicator of repeated
choice is added.
In those models where party support (as measured by party utilities) will be used as dependent
variable, it will be necessary to transform the dataset so that the unit of analysis is not individuals
but the utility that each of the parties yield for each individual (van der Eijk et al., 2006). In
practice, this simply implies a further disaggregation of the data within the individual that
indicates how much each of the possible choices is preferred. The procedure is the same as
required for conditional logistic regression (McFadden, 1974) and consists in turning the original
dataset, organized as in Table 1, into another one that has as many rows as available alternatives
(see Table 2). As this results in a multiplication of cases (as many as parties * individuals),
standard errors must be corrected in some way. One way is to weight the cases in the stacked
dataset to the original number of respondents. A second solution is to cluster standard errors by
respondent. Finally, it is also possible to use a random-effects hierarchical model where
respondents are used as a higher level of aggregation. Here, results with clustered standard errors
are presented, but findings were substantially the same regardless of the method employed.
It might be argued that, given the multi-level structure of the data, a hierarchical model using
the party as a level of aggregation should be employed instead. This logic is, however, not very
12
reasonable in this case. As Van der Brug, Franklin and Toka (2008) argue, not only do the parties
contained in the data not represent a random sample from the population of parties but also their
number is not large enough to employ a hierarchical model without raising the probability to
obtain a type-II error. Instead, the parties are treated in the same way as are the choices in a
discrete choice model. It is worth mentioning, though, that models introducing fixed effects by
party did not produce substantial differences.
Table 1: Original dataset.
Case ID
Dependent
variable
Individual-specific
Party-specific
Party-specific
Party-specific
Variable
variable
variable
variable
(party 1)
(party 2)
(party 3)
1
1
53
0
0.26
0.38
2
3
25
0.4
0
0.06
Table 2: Stacked dataset.
Case ID
Dependent
variable
Individual-specific
variable
Party-specific
variable
1 (party 1)
1
53
0
1 (party 2)
0
53
0.26
1 (party 3)
0
53
0.38
2 (party 1)
0
25
0.4
2 (party 2)
0
25
0
2 (party 3)
1
25
0.06
13
5 Findings.
Regarding hypotheses H1 and H1b, it will be recalled that, rather than assuming that voters are
loyal to a party, vote for that party most of the time and keep being approximately as loyal as they
always used to be, I do not expect voters to retain the same level of support after choosing a
particular party. On the contrary, support for the party voted for is argued not to remain still but to
become stronger with time. If this is the case, results will indeed be consistent with the presence of
some kind of habituation process in voting behavior. Habituation was hypothesized to work in two
steps. When voters choose a particular party, their preferences for that party increase. So, by the
next election, we should find support for that party to have increased more than support for other
parties (H1). Moreover, repeated vote for the same party should produce further increases in
support (H1b). Hypotheses 1 and 1b set no limit in the number of elections during which this
phenomenon can happen. However, the reinforcement process is expected to stop at some point,
preferences becoming more stable after a number of repetitions. But, as already mentioned, longer
panel data would be needed in order to falsify this second part of the theory.
So, in order to test Hypotheses 1 and 1b I proceeded in the following way. First, the dataset
was stacked in the way explained in the previous section so that the dependent variable (party
utility) is measured at the party level. Second, I performed multiple imputation in order to avoid
loss of efficiency and possible bias from missing cases (Rubin, 1987, 1996).
9
Third, cases were
weighted in order to prevent results from being influenced by the different number of respondents
in each of the surveys. Lastly, I introduced a lagged version of the dependent variable. This makes
the model dynamic and allows us to interpret coefficients in terms of changes rather than levels of
the dependent variable. Now, given that it is more difficult for party utility to decrease when its
value approaches 0 and to increase when it approaches 10, a cubic form of the lagged dependent
variable will be employed. Comparisons between parties with different values may be misleading
unless we control for this kind of boundary effect.
10
Having done this, I tested H1 by regressing current party utility on a binary variable that
accounts for whether the voter had voted for a given party in the previous election (t-1) or not. If
preference updating did not occur in a systematic way after choosing a party, the average change
14
in support for a party that has been previously voted for should be zero. That would imply that
voters choose parties that they have strong preferences for, but preferences only change randomly
after voting for them. In contrast, results (shown in Model 1, Table 3) tell us a completely
different story. As can be seen, the average change in party utilities (ie. the coefficient of the
binary variable ‘Party chosen at t-1’) is 1.97 points higher for those parties that had been voted for
in the previous election as compared with any other party. This is quite a strong effect if we take
into account that the standard deviation of the dependent variable is 3.6. So, far from remaining
constant over time, there is an important systematic reinforcement of voters’ preferences that
affects previously chosen parties, which is consistent with H1.
Table 3: Models of changes in party support. Dependent variable: propensity to vote for a party.
(Model 1)
(Model 2)
Party chosen at t-1
1.968***
1.267***
(0.103)
(0.108)
Repeated vote (vote at t-1 = vote at t-2)
-0.192***
(0.038)
Repeated vote * Party chosen at t-1
1.234***
(0.100)
Lagged utility
0.477***
0.437***
(0.061)
(0.066)
Lagged utility2
0.060***
0.071***
(0.014)
(0.016)
Lagged utility3
-0.005***
-0.006***
(0.001)
(0.001)
Constant
1.249***
1.411***
(0.068)
(0.077)
Year dummies
YES
YES
Observations
34,445
34,445
Individuals
3,467
3,467
Adjusted R²
0.536
0.539
Coefficients are unstandardized
Robust standard errors (clustered by individual) in parentheses
*** p<0.01, ** p<0.05, * p<0.1; two-tailed tests.
15
As a counterfactual, I ran 1,000 simulations of the very same model with artificial data in which
utilities were assumed to equal previous utilities plus a random component (results are shown in
the appendix available on the publisher’s website together with the simulation code employed).
11
In plain words, those data correspond to a scenario of no systematic updating of party utilities. As
expected, a test on the coefficients of the simulations revealed that, under such an assumption, the
average effect of having voted for a party in the previous election is not significantly different
from zero. This greatly differs from what survey data show, providing stronger support for the
presence of habituation.
As mentioned, the length of the panel studies used in this paper does not permit the tracing of
individuals back in time for more than a couple of elections. But it is worth testing whether the
effect of voting for a party is reinforced with repeated voting, as stated in H1b. With that aim,
utility change was regressed on several variables: 1) a dichotomous variable (‘Party chosen at t-1’)
that indicates whether a party was chosen at the previous election; 2) another dichotomous
variable (‘repeated vote’) accounting for whether the voter had also chosen the same party before
(at t-2); and 3) an interaction between the two. This model enables us to distinguish the effect of a)
choosing a party once, b) choosing a party for at least two consecutive times, and c) not choosing
a party for at least two consecutive times.
12
The expectation is that preferences for parties that
were voted for at t-1 will be more likely to increase than preferences for any other party. Whether
this increase takes also place for those who chose the same party at least two times in a row is
another question, although, as mentioned, findings in the literature point to a number of elections
higher that 2 in order for crystallization to take place.
Results are presented in Model 2. As expected, party utilities increase more strongly for parties
that have been voted for in the previous election. This is true for those who had only voted for that
party at t-1 (this is the coefficient of the main term ‘party chosen at t-1’), but also for those who
had voted for the same party on more than one occasion, which is consistent with H1b. In fact,
support for the chosen party becomes even higher by the next election when the voter had
16
previously voted for that party (the interaction term). Regarding the size of the effects, utilities for
parties that had only been voted for in the previous election (t-1) increased by 1.27 points. When
parties had been voted for both at t-1 and at t-2, their utility increased even more: 2.3 points.
Additionally, support for those parties that had not been chosen for the last two elections tends to
grow significantly less than those of any other party (as shown by the principal term ‘repeated
vote’, whose coefficient is -0.192).
13
So, support for the chosen party increases by the next
election, the effect being even stronger for parties that had been voted for on at least two
consecutive occasions. Regarding the latter effect, an alternative mechanism must, however, be
acknowledged. It is possible that coefficients are not reflecting an effect of repetition itself but
only that loyal voters are different from those who changed their vote, the latter being somewhat
more reluctant to give their recently chosen party a higher score. Be that as it may, the fact that
preferences for the chosen party tend to increase by the next election for both types of voter
provides support for the presence of a habituation process in voting behavior.
Now, in order to provide a counterfactual I followed the same logic employed earlier and
generated data under the assumption that the process of preference updating is not systematic – ie.
that preferences are the sum of their past values plus a random disturbance – and ran 1,000
simulations of Model 2.
14
This is just a test to check that coefficients are different from what
would be found in a model with different assumptions. Indeed, results (Table 4) present a radically
different scenario from the one that we find in the surveys. Under the new assumptions, the
coefficient for ‘Party chosen at t-1’ (which refers to parties that have only been voted for at t-1 but
not at t-2) is negative instead of positive. Moreover, for voters that had repeated the same choice
the effect of voting for a party (which is the sum of the interaction and the two main terms) is
negative in the simulated models (-.407).
15
More importantly, all the coefficients in the simulation
are significantly different from those in the model of observed data (p<0.01). It is then clear that,
under a scenario of no systematic updating of preferences, effects would not correspond to those
that are found with actual data.
17
Table 4: Counterfactual simulations of Model 2 under a scenario of non-systematic utility change.
N of replications: 1,000.
Independent variable
Mean
coefficient
Mean
standard error
Party chosen at t-1
-1.447***
.108
Repeated vote (vote at t-1 = vote at t-2)
-1.042***
.090
Repeated vote * Party chosen at t-1
2.081***
.125
Total effect of the interaction
-.407***
.096
Other coefficients from Model 2 are omitted.
*** p<0.01, ** p<0.05, * p<0.1; two-tailed tests.
Let us now move on to the remaining hypotheses. According to H2, the effect of inertia should
accumulate over time. If only because the chances to have repeated one’s choice several times are
higher as individuals grow older, we should expect older voters to present much more stable vote
patterns compared to young adults. Model 3.1 in Table 5 introduces the effect of age.
16
Not only is
this effect positive and significant, but also very strong indeed. The probability of remaining loyal
to the party chosen at the previous election is .55 for voters between 21 and 25 years old and
increases dramatically with age, reaching .76 points among those over 81 years-old. The question
is to what extent inertia helps us to understand this effect.
18
Table 5: Logit model of vote loyalty. DV: Voting for the same party as in the previous election
(vote t = vote t-1) or not.
(Model 3.1)
(Model 3.2)
(Model 3.3)
Age (13 categories)
0.086***
0.045**
0.029
(0.014)
(0.019)
(0.020)
Previously loyal voter
0.662***
(0.114)
Utility of previously chosen party
0.699***
0.644***
(0.059)
(0.060)
Utility of next most preferred party
-0.683***
-0.644***
(0.057)
(0.057)
Political interest (0-1=max)
0.697***
0.722***
(0.174)
(0.177)
Church attendance (0-1=max)
-0.334**
-0.258*
(0.140)
(0.143)
Education (0-1=max)
0.268
0.184
(0.192)
(0.197)
Gender (1=female)
0.174*
0.153
(0.101)
(0.103)
Civil status (1=married/cohabiting)
0.054
0.009
(0.111)
(0.113)
Urbanization (0-1=max)
0.237
0.247
(0.155)
(0.159)
Constant
0.559***
-0.125
-0.304
(0.123)
(0.530)
(0.529)
Year dummies
YES
YES
YES
Observations
3,053
3,053
3,053
Pseudo R²
.022
.344
.354
Unstandardized logistic coefficients
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1; two-tailed.
19
Hypothesis H3 argues that in order to render the effect of age on vote loyalty redundant, it is
necessary to add the effect of inertia into a model containing conventional variables that are
related to the passage of time. Thus, Model 3.2 includes several indicators of political
sophistication that have been used in published work on vote loyalty and are likely to run parallel
to the aging process: political interest (eg. Zuckerman, Kotler-Berkowitz and Swaine, 1998) and
level of education (eg. Söderlund, 2008). Together with these, other variables related to the
stability of social networks were introduced (namely civil status, which is also expected to change
with age, religious attendance and urbanization of the place of residence). Other controls, such as
utility of the previously chosen party and of the first other alternative were also introduced.
17
As
expected, the effect of age shrinks with the introduction of these other controls, even if it remains
significant at p<0.05. Model 3.3, however, adds a measure of previous loyalty that is codified as 1
if the voter had chosen the same party in the last two elections and 0 if she had not. As can be
seen, doing so further reduces the effect of age, making it statistically indistinguishable from
zero.
18
In turn, having repeated one’s choice at the previous election increases the probability of
doing so at the next election too, the effect being significant at p<0.01.
19
This suggests that using
previous loyalty is essential in order to understand why the probability of changing one’s vote
increases with age.
20
Moreover, when interactions of year with age and with former loyalty were
introduced, none of them came out significant at p<0.05.
21
Thus, results support the hypothesis
that inertia accumulates over time and that this accounts for an important part of the effect of age
on switching.
6 Conclusion.
That voting behavior is hard to change for most individuals is often taken for granted, but attempts
to explain this mechanism are very rare. Political science has usually focused on the development
of party identification in two-party systems and concluded that these tend to crystallize with age.
But instead of leading to a better understanding of the way in which the habituation process
operates, discussions about the paradigm that best serves to explain the presence of this kind of
inertia have populated the literature.
20
This article has attempted to shed more light on the reasons why the vote is so difficult to
change employing panel data from Dutch national election surveys. It was hypothesized that the
habituation process of voting takes place in two stages: reinforcement and crystallization. First,
repeatedly voting for a particular party affects voters’ distribution of preferences, as support for
the chosen party tends to increase more than support for other parties. Stability does, therefore, not
necessarily imply absolute immobility. Results show that support for parties that have been
recently voted for tend to increase more by the next election than does support for any other party.
This suggests that the possibility of ever voting for a different party is increasingly ruled out as
choices are repeated. Moreover, it was shown that a scenario of no systematic updating of
preferences would have resulted in very different effects. In an additional test, voters that chose
the same party on at least two occasions were compared with those who did not. Support for both
groups of voters tend to grow by the next election, although those of voters that repeated their vote
seem to increase more strongly. Arguably, this is evidence that the reinforcement effect lasts for
more than one election, although the possibility of alternative mechanisms should be
acknowledged, as voters that change their vote may simply be more reluctant to become
‘enamored’ of their new choice. Be it as it may, the reinforcement mechanism does probably not
continue to operate ad infinitum . As other studies have shown, preferences should tend to remain
stable after a number of repetitions (eg. Butler and Stokes, 1974; Dinas, 2010). Unfortunately, the
characteristics of the data employed here did not enable testing the crystallization effect, as it was
not possible to trace voters back for more than a couple of elections. Finally, it was argued that the
effect of inertia tends to accumulate over time. In other words, the more elections that a voter
experiences, the higher are the chances that the same choice will be repeated. In this regard,
repeated voting for a party was shown to be essential in order to understand higher levels of party
loyalty among older voters. In fact, introducing previous loyalty and other variables that change
with the life course, and especially political interest, rendered insignificant the effect of age on the
probability to stay loyal to a party. The rendering of age insignificant when inertia is taken into
account can be regarded as a critical test of the habituation hypothesis.
21
This article has contributed to the current literature by using a non-ipsative measure of party
utility in order to test how preferences evolve for both the chosen party and the other alternatives.
The measure is, of course, not free from error, but it seems more appropriate to test reinforcement,
especially in the context of a multi-party system, than the unidimensional measures of party
identification that are usually employed for this purpose. Moreover, the use of Dutch data has
provided the opportunity to check that the mechanisms explained earlier work under very different
conditions: elections with low and high levels of volatility (1986 and 1989; and 1994 and 2003,
respectively), and one exceptional raise in the levels of instability (in 1994). Even if the data did
not allow taking account of former repetition of choices beyond the previous two elections, the
effect of inertia was present in the four scenarios. In general, results do not seem to differ much
across elections, and even when they do, they do not contradict the theory. Thus, the effect of
repeated vote for a party seems to work in a similar fashion regardless of how volatile elections
turn out to be – at least in scenarios such as the Dutch case and most other advanced democracies
where volatility does not tend to come together with severe party system meltdown between
elections. Moreover, the fact that effects consistent with a certain habituation process were found
in the Netherlands, where the multi-party system should make it harder for voters to keep long-
term allegiances, adds support to the body of literature that has focused on the effect of repeated
voting in a different context. Of course, there is always the possibility that habituation is produced
by mechanisms different from the act of voting itself. It is certainly possible that voting for a party
only reflects an underlying process of becoming more fond of that particular party. This paper did
not test the particular underlying mechanisms leading to habituation, which remains for further
research.
The evidence presented here point to the existence of habitual voting and is aimed at
encouraging researchers to pay attention to this important phenomenon when trying to explain
electoral change. Taking into account the impact of inertia is essential to understanding the
mechanisms of vote loyalty. Moreover, the accumulation of inertia over time has very important
consequences because it implies that the greatest part of the change observed in elections is likely
due to the effect of young switchers. Of course, the presence of alternative mechanisms must be
22
acknowledged and investigated, but electoral studies should definitely be more aware of the
implications of the habituation process and, therefore, take account of the responses of young
voters when modeling volatility.
Acknowledgments
I am indebted to Wouter van der Brug, Mark Franklin, Jose Ramon Montero and three anonymous
reviewers for their very useful comments. I would also like to express my gratitude to the editor of
this article, Edward Fieldhouse, for his careful work and very good suggestions. Part of this article
was made possible thanks to the funding of the European University Institute and the Spanish
Ministry of Education. Any mistakes or inaccuracies are my responsibility alone.
References
Achen, Christopher H. 1992. “Social Psychology, Demographic Variables, and Linear
Regression: Breaking the Iron Triangle in Voting Research.” Political Behavior 14(3):195–212.
Achen, Christopher H. 2002. “Parental Socialization and Rational Party Identification.” Political
Behavior 24(2):151–170.
Adams, James. 2001. Party Competition and Responsible Party Government: a Theory of Spatial
Competition Based on Insights from Behavioural Voting Research. Ann Arbor: University of
Michigan Press.
Aldrich, John H., Jacob M. Montgomery and Wendy Wood. 2011. “Turnout as a Habit.” Political
Behavior 33(4):535–563.
Allison, P.D. 2000. “Multiple imputation for missing data.” Sociological methods and
Research 28(3):301–309.
Alwin, D.F. and J.A. Krosnick. 1991. “Aging, cohorts, and the stability of sociopolitical
orientations over the life span.” American Journal of Sociology pp. 169–195.
Anker, H. and E.V. Oppenhuis. 1997. “Dutch Parliamentary Election Panel Study, 1989-1994”
[Computer file].
23
Armingeon, Klaus, Sarah Engler, Panajotis Potolidis, Marl`ene Gerber and Philipp Leimgruber.
2010. Comparative Political Data Set 1960-2008. Institute of Political Science, University of
Berne.
Bartolini, Stefano and Peter Mair. 1990. Identity, Competition and Electoral Availability. The
Stabilisation of European Electorates, 1885-1985. Cambridge: Cambridge University Press.
Bem, D.J. 1972. Self-Perception Theory. In Advances in Experimental Social Psychology, Vol. 6,
ed. L. Berkowitz. New York, NY: Academic Press pp. 1–62.
Butler, David and Donald Stokes. 1974. Political change in Britain: the evolution of electoral
choice. London: Macmillan.
Campbell, Angus, Philip E. Converse, Warren E. Miller and Donald E. Stokes. 1960. The
American Voter. Chicago: University Of Chicago Press.
Chorus, C.G and B.G.C Dellaert. 2012. “Travel choice inertia: The joint role of risk aversion and
learning.” Journal of Transport Economics and Policy 46(1):139–155.
Converse, P E. 1969. “Of Time and Partisan Stability.” Comparative Political Studies 2(2):139.
Converse, Philip E. 1964. The Nature of Belief Systems in Mass Publics. New York: Free Press.
Dalton, Russel J. 2000. The Decline of Party Identifications. In Parties without Partisans:
Political Change in Advanced Industrial Democracies, ed. R Dalton and M Wattenberg. Oxford:
Oxford University Press.
Denny, K and O Doyle. 2009. “Does Voting History Matter? Analyzing Persistence in Turnout
Source.” American Journal of Political Science 53:1–17.
Dinas, Elias. 2010. The impressionable years: the formative role of family, vote and political
events during early adulthood. Ph. D. Thesis. Florence: European University Institute.
Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper and Row.
Elinder, Mikael. 2012. “Correcting mistakes: cognitive dissonance and political attitudes in
Sweden and the United States.” Public Choice 153(1-2): 235–249.
Evans, Geoffrey. 2004. Voters and Voting: An Introduction. London: Sage Publications.
Fearon, James D. 1999. “Electoral Accountability and the Control of Politicians: Selecting Good
Types versus Sanctioning Poor Performance.”
24
Festinger, L. 1957. A theory of cognitive dissonance. Evanston, IL: Row & Peterson.
Fiorina, M P. 1981. Retrospective Voting in American National Elections. New Haven: Yale
University Press.
Franklin, M, P Lyons and M Marsh. 2004. “Generational basis of turnou decline in established
democracies.” Acta Politica 39(2):115–151.
Gerber, A, Donald P Green and R Shachar. 2003. “Voting may be habit-forming: Evidence from
a randomized field experiment.” American Journal of Political Science 47(3):540–550.
Green, Donald P, Bradley Palmquist and Eric Schickler. 2002. Partisan Hearts and Minds:
Political Parties and the Social Identities of Voters. Yale ISPS series New Haven, CT: Yale
University Press.
Green, Donald P and Ron Shachar. 2000. “Habit Formation and Political Behaviour: Evidence of
Consuetude in Voter Turnout.” British Journal of Political Science 30(4):561–573.
Grynaviski, Jeffrey D. 2006. “A Bayesian Learning Model with Applications to Party
Identification.” Journal of Theoretical Politics 18(3):323–346.
Harel, O. 2009. “The estimation of R 2 and adjusted R 2 in incomplete data sets using multiple
imputation.” Journal of Applied Statistics 36(10):1109–1118.
Irwin, G.A., J.J.M. van Holsteyn and J.M. den Ridder. 2005. “Dutch Parliamentary Election
Study 2002-2003 [Computer file].”
Key Jr., V. O. 1952. Politics, Parties, and Pressure Groups. New York, NY: Thomas Y. Crowell
Company.
Key Jr., V. O. 1966. The Responsible Electorate: Rationality in Presidential Voting, 1936-1960.
Cambridge: Belknap Press of Harvard University Press.
Laakso, Markku and Rein Taagepera. 1979. “Effective number of Parties: A measure with
Application to West Europe.” Comparative Political Studies 12(1):3–27.
Macdonald, Stuart Elaine, Ola Listhaug and George Rabinowitz. 1991. “Issues and party support
in multiparty systems.” The American Political Science Review 85(4):1107–1131.
Manin, Bernard, Adam Przeworski and Susan Stokes. 1999. Elections and Representation.
Cambridge: Cambridge University Press.
25
McFadden, Daniel. 1974. Conditional logit analysis of qualitative choice behavior. In Frontiers
in Econometrics, ed. P Zarembka. Vol. 1 New York: Academic Press chapter 4, pp. 105–142.
McPhee, William N and Jack Ferguson. 1962. Political Immunization. In Public Opinion and
Congressinoal Electinos, ed. William N McPhee. Glencoe: The Free Press.
Meredith, Marc. 2009. “Persistence in Political Participation.” Quarterly Journal of Political
Science 4(3):187–209.
Miller, W E and J Merrill Shanks. 1996. The New American Voter. Cambridge: Cambridge
University Press.
Mood, Carina. 2010. “Logistic Regression: Why We Cannot Do What We Think We Can Do,
and What We Can Do About It.” European Sociological Review 26(1):67–82.
Mullainathan, Sendhil and Ebonya Washington. 2009. “Sticking with Your Vote: Cognitive
Dissonance and Political Attitudes.” American Economic Journal: Applied Economics 1(1):86 –
111.
Nadeau, R. and M. Mendelsohn. 1994. “Short-term Popularity Boost Following Leadership
Change in Great Britain.” Electoral Studies 13:222–228.
Payne, J.W., J.R Bettman and E.J Johnson. 1993. The Adaptive Decision Maker. Cambridge:
Cambridge University Press.
Pedersen, M. 1983. Changing patterns of electoral volatility in European party systems, 1948-
1977: Explorations in explanation. London: Sage.
Plutzer, E. 2002. “Becoming a Habitual Voter: Intertia, Resources, and Growth in Young
Adulthood.” American Political Science Review 96:41–56.
Rabinowitz, G. and S.E. Macdonald. 1989. “A directional theory of issue voting.” The American
Political Science Review pp. 93–121.
Riker, William H. 1986. The Art of Political Manipulation. New Haven, CT: Yale University
Press.
Rubin, D.B. 1987. Multiple Imputation for Nonresponse in Surveys. New York: J. Wiley &
Sons.
Rubin, D.B. 1996. “Multiple imputation after 18+ years (with discussion).” Journal of the
American Statistical Association 91:473–489.
26
Schmitt-Beck, Rüdiger, S Weick and B Christoph. 2006. “Shaky Attachments: Individual-
level Stability and Change of Partisanship among West German Voters, 1984-2001.” European
Journal of Political Research 45:581–608.
Shugan, S.M. 1980. “The Cost of Thinking.” Journal of Consumer Research 7:99–111.
Simon, H.A. 1978. “Rationality as Process and as Product of Thought.” American Economic
Review 68(2):1–116.
Söderlund, Peter. 2008. “Retrospective Voting and Electoral Volatility: A Nordic Perspective.”
Scandinavian Political Studies 31(2).
Strate, J.M., C.J. Parrish, C.D. Elder and C. Ford. 1989. “Life span civic development and voting
participation.” The American Political Science Review pp. 443–464.
Tillie, Jean. 1995. Party Utility and Voting Behavior. Amsterdam, the Netherlands: Het
Spinhuis Publishers.
van der Brug, Wouter, M Franklin and G Toka. 2008. “One electorate or many? Differences in
party preference formation between new and established European democracies.” Electoral
Studies 27(4):589–600.
van der Eijk, C., G.A. Irwin and B. Niemöller. 1997a. “Dutch Paliamentary Election Panel Study,
1986-1989” [Computer file]. ICPSR version.
van der Eijk, C., M. Franklin and W. van der Brug. 1999. Policy preferences and party choice. In
Schmitt, Hermann and Jacques Thomassen (eds.), Political Representation and Legitimacy in the
European Union. Oxford: Oxford University Press.
van der Eijk, C. and Mark N. Franklin. 1996. Choosing Europe?: the European electorate and
national politics in the face of union. Ann Arbor: University of Michigan Press.
van der Eijk, Cees, G.A. Irwin and B. Niem¨oller. 1997b. “Dutch Parliamentary Election Panel
Study, 1981-1986” [Computer file].
van der Eijk, Cees, Wouter van der Brug, Martin Kroh and Mark N Franklin. 2006. “Rethinking
the dependent variable in voting behavior: On the measurement and analysis of electoral utilities.”
Electoral Studies 25(3):424–447.
Zuckerman, Alan S, Laurence A Kotler-Berkowitz and Lucas A Swaine. 1998. “Anchoring
Political Preferences: The Structural Bases of Stable Electoral Decisions and Political Attitudes in
Britain.” European Journal of Political Research 33(3):285–321.
27
Zuckerman, A.S. 1989. “The bases of political cohesion: Applying and reconstructing crumbling
theories.” Comparative Politics 21(4):473–495.
28
Notes
1
The process that this paper studies is usually referred to as either inertia or habituation in the
literature. These two terms will be employed interchangeably here although perhaps inertia
should be understood as the mechanism that produces the habituation process.
2
In this regard, Dinas (2010) has shown party choice in the US to follow a rather stable pattern
after the third and especially the fourth election at which an individual has chosen the same
party.
3
Note, however, that only the 1984-1986 part of the study contains the propensity to vote for
each of the parties which, as will be explained, is one of the key variables of this study. This
is why information from the 1981 study will only be employed in order to build a variable
accounting for the number of times that individuals chose the same party.
4
In the case of the 1981-1984-1986 study, questions concerning vote choice were asked after
the corresponding election. This is true for most respondents in the other three studies too, as
most of them had participated in a previous survey (a common practice in Dutch panel
studies). Thus, information on party choice at the election previous to the study (which
corresponds to t-2 here) was, too, given by these respondents right after that election.
However, each study also incorporates an additional number of respondents that had not been
previously interviewed. For these respondents, party choice at t-2 was asked during the first
wave of the study – that is before the t-1 election. In principle, asking the question before the
election should avoid contamination from their most recent vote in voters’ recall. I compared
results between the 1981-1984-1986 study and the rest and no significant differences arose.
5
Other parties that may have chances to obtain representation according to the pre-election
surveys are usually included as well.
6
As calculated by Laakso and Taagepera (1979). Data from the Comparative Political Dataset
I: 1960-2006 (Armingeon et al., 2010).
7
Including the European Election Studies from 1989 onwards and national election studies in
countries such as Britain, Germany, Ireland and Spain among others (van der Eijk et al.,
2006, p.432).
8
Calculating changes in utilities was possible because information on these was retrieved two
times in the panel studies (at t-1 and at t).
9
The proportion of missing cases was not very high for any single variable – it was never
above 5%. For this reason, the standard number of 5 imputations was used. Results did
however not substantially differ before and after imputation. Multivariate normal regression,
as implemented by mi impute in Stata 11, was the method used for imputing the data,
following Allison (2000) to arrange categorical variables. The reported adjusted R²’s were
calculated using Fisher’s z transformation in order to apply Rubin’s rules (for more details,
see Harel, 2009). Besides all the variables included in these models, other variables such as
unionization, social class, religion, opinion on different issues, coalition preferences, province
of residence and so on were used in the imputation process.
29
10
This was also checked against simulated data in which preferences were set to depend on
their past values plus a random disturbance but limited between 0 and 10. The direction and
significance of coefficients showed that the cubic form is appropiate.
11
I am very thankful to an anonymous reviewer for suggesting these simulations as a way to
generate the counterfactuals and providing part of the code.
12
Parties in the reference category are those that were not chosen at the previous election (t-1)
by voters who had changed their vote.
13
It is worth noting that these effects are robust to the introduction of fixed effects by party and
are similar for elections with very different levels of volatility. Interactions with year
dummies were introduced and, even if two of them were statistically significant, they did not
contradict Hypothesis H1. First, having voted for a party once had an even stronger effect in
the 2003 election (in other words, parties that were voted for in 2002 but not in 1998 yielded
particularly high utilities by the 2003 election). Second, having been repeatedly chosen had a
positive but significantly smaller effect in 1994 than in the rest of the years. This suggests
that the same mechanisms keep operating even in elections with distinctly high levels of vote-
switching.
14
Further details are provided in the appendix available on the publisher’s website.
15
This makes sense. If deviations from past preferences were random and voters always picked
the party that they prefer the most, switching is likely to be produced by exceptional shocks
that temporarily decrease voters’ preferences for the chosen party and/or increase their
preferences for an alternative party. As deviations from past preferences are random, utilities
will tend to converge over time around their original levels. In the simulations, random
shocks are normally distributed. So, utilities that have experienced an exceptional increase
should, on the average, present lower levels by the next election, while those that experienced
exceptional decreases should tend to increase. Remember that the reference category in
Model 2 corresponds to parties that were not voted for when the voter had switched. Now, the
only parties whose utilities are expected to increase by the next election are those that
experienced exceptional decreases, which are likely to be found among the parties that were
defected from. And those are in the reference category. So, preferences for any other party are
expected to increase less than for the reference category. On the other hand, parties that were
voted for by a switching voter are likely to be among those who suffered from extreme
increases in their utilities. Therefore, stronger negative changes should be expected for this
category, which corresponds to the variable ‘Party chosen at t-1’ in the Model.
16
Note that in the 1994 study age is measured in 13 categories corresponding to 5-year
intervals. For the sake of coherence, age was recoded this way in the rest of the surveys too.
17
Because voters are expected to switch when their formerly chosen party does not (or does no
longer) yield a high utility, but also when other alternative parties do yield a high utility.
Regarding the number of alternatives that should be included, only the coefficients for the
chosen party and for the alternative party that yields the highest utility have a significant
effect on loyalty (results are available on request), suggesting that an average voter only has
30
two parties in mind when considering whether to switch. Thus, utilities for the rest of the
parties were excluded from the final model.
18
Mood (2010) shows that comparing coefficients across logistic regression models with
different independent variables can sometimes be misleading. In turn, she suggests employing
other measures, such as population-averaged conditional effects, that are not affected by
unobserved heterogeneity and are, therefore, more appropriate for comparisons across
models. Computing population-averaged probabilities in this case did not change the overall
conclusions. In the first model (Model 3.1), the probability to remain loyal to a party
increases by .018 points with every additional point of age (ie. every 5 years), the effect being
statistically significant at p<0.01. In the second model (Model 3.2), though, the probability
shrinks to .006 points, but it is still significant at p<0.05. Lastly, in the third model (Model
3.3), the population-averaged probability is .0038 and is not significantly different from zero
at p<0.1.
19
The population-averaged probability of remaining loyal to a party increases by 9% for voters
that had chosen the same party in the previous two elections.
20
Actually, it is only necessary to introduce inertia (measured as previous choice repetition)
and political interest without any other control in order to render the effect of age
insignificant.
21
Only the interaction with the 1994 election came out significant at p=0.09. The coefficient
indicates that the effect of previous loyalty was somewhat smaller at that exceptional election.
Note that conclusions do not change when probabilities deriving from these interactions are
computed.
