Electoral Studies, Vol. 15, No. 1, pp.

39-52, 1966
Pergamon Copyright 0 1996 Elsevier Science Ltd
Printed in Great Britain. AU rights reserved
0261-3794(94)00014-X 0261-3794/96 $15.00+0.00

Measuring Strategic Voting: A Two-Step

Procedure
ANDRE Bws AND RICHARD NADEAU

Dbpartement de Science Politique, Universite’de Montrkal, CP 6128, Succursale

Centre-Ville,Montreal, Canada H3C 3J7

The paper proposes a two-step procedure to measure strategic voting in

legislative elections. The approach uses voters’ rank-order of preferences
as well as their perceptions of the parties’ chances of winning the
election. It proceeds in two steps. The first step consists of identifying the
pool of potential strategic voters, those who are faced with a strategic
choice because their most preferred party (candidate) is perceived to have
smaller chances of winning than their second choice. The second stage is
confined to that group for whom strategic voting is a real option. We
examine the actual vote decisions of that group, and relate them to the
intensity of preferences and perceptions of the race. The methodology is
applied to the 1988 Canadian election, We show that at least 6 per cent
of the total electorate can be said to have voted strategically. We also show
that the propensity to vote strategically increases when the intensity of
preference for the first choice over the second decreases, when the
second choice is perceived to have much better chances of winning than
the first choice, and when the race between the second and third choices
gets closer.

For many years, political scientists have examined the existence and extent of strate-
gic voting in elections held under the plurality rule. The basic logic of strategic
voting has been aptly described by Downs (1957, 48):

A rational voter first decides what party he [sic] believes will benefit him
most; then he tries to estimate whether his party has any chance of
winning. He does this because his vote should be expended as part of a
selection process, not as an expression of preference. Hence even if he
prefers party A, he is wasting his vote on A if it has no chance of winning
because very few other voters prefer it to B or C. The relevant choice in
this case is between B and C. Since a vote for A is not useful in the actual
process of selection, casting it is irrational.

Strategic voting, then, implies the following. First, voters do not vote for the party
they like the best. Second, the voting decision is taken on the basis of the combi-
nation of preferences and perceptions of the various parties’ or candidates’ chances
40 Measuring Strategic Voting: A Two-StepProcedure

of winning the election. In the simple case of three parties, this entails that voters,
when realizing that their first choice cannot win, vote for their second preference
in order to prevent the party/candidate they like the least from winning. We may
thus define a strategic vote as a vote for a second-preferred party (candidate)
rather than for the frost-preferred one, motivated by the perception that
the former has a better chance of winning the election.
Whether strategic voting is rational is debatable since any individual voter’s chance
of infhtencing the outcome is negligible (see Riker and Ordeshok, 1968; Ferejohn
and Fiorina, 1974, 1975; Meehl, 1977). We leave aside this important theoretical
question and address the empirical issue of measuring the extent and import of
strategic voting in elections. We propose a procedure with two steps. The first step
consists of distinguishing two groups of voters, those who are and those who are
not faced with the decision to vote strategically or not. In the second step, the analy-
sis focuses on the group facing a strategic choice, and assesses the relative propen-
sity to vote for the first or second-preferred party (candidate). We show this
procedure provides a more rigorous test of the presence of strategic voting.
That test is applied to a context potentially conducive to strategic voting, that of
the Canadian 1988 election. That election was entirely fought over one issue, a free
trade agreement (FTA) between Canada and the United States. The FTA was
supported by the incumbent Conservative government but opposed by the two
main opposition parties, the Liberals and the New Democratic Party (NDP). Those
opposed to the FTA, in particular, were faced with the question, raised explicitly
in the campaign, of whether they should not support the party that was most likely
to defeat the Conservatives in their own riding.

1. How to Measure Strategic Voting

Various approaches have been used to measure strategic voting. The frost, most
indirect, approach consists in deducing the presence of strategic voting from aggre-
gate electoral outcomes: the lower support for minor parties in plurality systems,
compared to that observed in PR systems (Fisher, 1973; Blais and Catty, 1991) and
the pattern of inter-election vote shifts in different types of constituencies within
plurality systems (Spafford, 1972; Curtice and Steed, 1988; Galbraith and Rae, 1989;
Johnston and Pattie, 1991) have been interpreted as indirect evidence of strategic
voting.
The second approach is based on respondents’ own characterization of their
vote: a strategic voter is someone who indicates strategic or tactical motivations
as underlying his/her decision to vote for a given party (Heath et al., 1991; Niemi
et al., 1992; Evans and Heath, 1993, the ‘construct validity’ of this definition being
tested by inquiring whether those who are so identified as voting strategically have
the individual and contextual (with respect to constituencies) characteristics one
would expect from strategic voters.
The third approach goes directly to the heart of the matter, as it measures strate-
gic voting on the basis of a hiatus between an individual’s vote and his(her) prefer-
ences, a hiatus that is linked to the objective (Black, 1978; Cain, 1978) or subjective
(Brady and Johnston, 1987; Bartels, 1988; Abramson et al., 1992) parties’ or candi-
dates’ chances of winning the election.
We believe this third approach offers the most appropriate and direct test of
strategic voting. As the opening quote from Downs indicates, strategic voting entails
 .kNDd BLABAND RKHARD NADFAU 41

that voters come to a decision on how to vote on the basis of their preferences
among the parties and of the various parties’ probabilities of winning the election.
As the gist of the third approach is precisely to connect the vote decision to a
combination of preferences and probabilities, this seems to be the most logical way
to proceed.
This approach will of course display its full leverage to the extent that prefer-
ences and probabilities are adequately measured. With respect to probabilities, one
may either derive the various parties’ chances of winning from electoral outcomes
in the constituencies or use individuals’ perceptions. Scholars agree, however, that
derived probabilities are a second-best option and that “ideally one would use
individual pre-election expectations”. (Niemi et al., 1992, 233. See also Black, 1978,
616; Cain, 1978, 647.)
Of special interest from this perspective is the recent paper by Abramson and
his colleagues about ‘sophisticated’ voting in the American 1988 presidential
primaries. Looking at respondents’ rank orders of preferences and expectations in
the Democratic and Republican primaries, these authors show that the vote in the
primary season reflected not only voters’ preferences but also their evaluations of
the candidates’ viability.
We believe the general avenue followed by Abramson, Aldrich, Paolino and
Rohde, that is of looking at the rank-order of preferences, of determining who votes
for a candidate other than the most preferred one, and of ascertaining whether
such choice is related to the perceived viability of the candidates, is basically appro-
priate. This is the approach that will be used in the study to be presented below.
At the same time, however, our study departs from their analysis in some impor-
tant respects. These departures, we contend, enable us to offer new insights about
strategic voting and its measurement.
First, Abramson and his colleagues’ work deals with primaries, where there are
many candidates, many of whom are relatively unknown. This makes a very partic-
ular context, much different from the most standard election in which the contest
is primarily between parties, that have a long-standing existence.’ The study of a
parliamentary election, the 1988 Canadian federal election, offers the opportunity
to see if the approach can be applied outside the peculiar context of primaries.
Second, methodological improvements are incorporated in the present study.
Voters’ preferences are assessed through a more encompassing methodology, and
a more satisfactory measure of the closeness of the race, a crucial piece of infor-
mation for strategic voters, is presented.
Finally, and more fundamentally, we adopt a two-step procedure, identifying first
the pool of voters whose combination of preferences and expectations puts them
in a strategic choice situation, and distinguishing in a second step those, within
that pool, who actually vote strategically. Only if one’s preferred party (candidate)
is perceived to be less likely to win than one’s second choice does a voter have
the option of voting strategically or not (see Heath et al., 1991, 55-56). For those
whose first choice is perceived to be more likely to win than their second choice
strategic voting is not an option. This flows directly from the definition of strate-
gic voting given above. As a logical consequence, it follows that the analysis must
proceed in two steps, the first step screening out those for whom strategic voting
is not an option, and the second step examining the vote decisions of those who
have to choose whether to support their first-preferred party (candidate) even
though it is less likely to win than their second choice or their second-preferred
42 Measuring Strategic Voting: A Two-Step Procedure

party (candidate) because it is more likely to win. As we shall see, this procedure
considerably simplifies the measurement of strategic voting and offers an appealing
solution to the problem of obtaining measures of expectations ‘purged’ of projec-
tion effects.

2. Facing a Strategic Choice

We use the 1988 Canadian Election Study (CES).* We focus on the choice among
the three major parties, the Conservatives, who got 43 per cent of the votes and
won 169 seats to form a majority government, the Liberals, who had 83 of their
candidates elected with 32 per cent of the votes, and the NDP, who finished third,
with 20 per cent of the votes and 43 seats. Other parties obtained only 5 per cent
of the vote, altogether. Only one, the Reform party received more than 1 per cent.
And no minor party candidate was elected. In the final section of the paper, though,
we return to the question of minor party support.
The first step in our analysis consists of determining who belongs to the pool of
potential strategic voters. As this requires knowing individuals’ preferences and
expectations, our first task consists of sorting out respondents according to their
preferences among the three major parties, the Conservatives, the Liberals and the
NDP. Previous studies have relied on one single indicator to assess voters’ prefer-
ences (or utilities), feeling thermometer questions about the parties in the case of
elections (Black, 1978; Cain, 1978) or about the candidates in the case of primaries
(Abramson et al., 1992; Bartels, 1988). We believe that this is an inadequate proce-
dure. In the case of elections, there are three objects in the voting decision: parties,
leaders, and local candidates. Voters cast their vote for a local candidate, associated
with a given party. As a result of the election, the leader of the party that wins the
most seats becomes Prime Minister. The vote must thus reflect a combination of
preferences among parties as such but also among local candidates and party
leaders.
We thus use three components of voters’ preferences: feelings about parties,
leaders, and local candidates. In order to give each of these components its proper
weight, the voting choice was regressed against the three components. The logis-
tic regression thus obtained indicated the relative propensity to vote for each party
for various combinations of preferences, that is if the vote were entirely based on
preferences. This gives us, in fact, a combined preference scale for each respon-
dent with respect to each party. For each respondent, we use the three preference
scales for the Conservatives, the Liberals, and the NDP, to sort out his/her rank-
order of preferences among the three parties.j This yields six groups of voters,
distinguished according to their overall order of preferences: (1) CON>LIB>NDP,
(2) CON>NDP>LIB, (3) LIB>CON>NDP, (4) LIB>NDP>CON, (5) NDP>CON>LIB,
(6) NDP>LIBXON. Table 1 shows the proportion of voters in each group.
To determine which voters were faced with a strategic choice we have to
connect preferences with expectations about the outcome of the election. Our
criterion to include an individual in the pool of potential strategic voters is straight-
forward: only those who perceive their most preferred party as being less likely to
win than their second preferred party found themselves in a strategic choice situa-
tion (Cain, 1978, 642; Heath et al., 1991, 56).
In order to proceed, we need data on how voters perceived the race among the
parties. There are in fact two distinct though overlapping races in a parliamentary
 AND&Bws ANDRICHARD
NADEAU 43

TABLE1. Groups of voters according to their rank ordered

preferences

Group %

1. CON>LIB>NDP 30
2. CON>NDP>LIB I9
3. LIB>CON>NDP 20
4. LIB>NDP>CON 14
5. NDP>CON>LIB 6
6. NDP>LIB>CON 11
N=2689

election such as the Canadian one. There is the local race, at the constituency level;
different parties may have different probabilities of getting their candidate elected
in a given riding. There is the national race; different parties may have different
probabilities of forming the government. There is some evidence that Canadians
take both kinds of probabilities into account in their voting decision (see Johnston
et al., 1980; Blais, Johnston et al., 1990; Johnston et al., 1992: Chapters 7 and 8).
We focus on perceptions of the local race. If there is any possibility that an individ-
ual voter may make a difference in the outcome of an election, it is at the level of
the local election, and not at the national level.*
We use respondents’ perceptions of the three major parties’ chances of winning
the election in their constituencies to sort out those who were faced with a strate-
gic choice. Since respondents could perceive two parties as equally likely to win,
we must include ties as a possibility. This gives us 13 different sets of rank-ordered
expectations (see Table 2). The most frequent situation is one where voters
perceive the Conservatives to be ahead and the NDP to be trailing (group 1). The
three other situations with frequencies greater than 10 per cent are those where
the Liberals are seen to have better chances than the Conservatives, the NDP still

TABLE2. Groups of voters according to their rank ordered

expectations

Group %

1. CON>LIB>NDP 27
2. CON>NDP>LIB 8
3. LIB>CON>NDP 14
4. LIB>NDP>CON 4
NDP>CON>LIB 6
z: NDP>LIB>CON 4
7. CON>LIB=NDP 10
8. LIB>CON=NDP 3
9. NDP>CON=LIB 3
10. CON=LIB>NDP 10
11. CON=NDP>LIB 3
12. LIB=NDP>CON
13. PC=LIB=NDP :
N=2675
44 Measuring Strategic Voting: A Two-Step Procedure

being last (group 3), where the Liberals and the NDP are viewed to be tied in
second place (group 7) and where there appears to be a tie in first place between
the Conservatives and the Liberals, with the NDP trailing (group 10).
It is of course the interconnection between preferences and expectations that
determines whether or not a voter is faced with a strategic choice. What matters
is that a voter’s most preferred party be perceived to have little chance of winning
the election, thus leading him/her to consider the possibility of voting for his/her
second choice because it is better positioned to defeat the most disliked party.
For a given set of preferences, five different sets of expectations correspond to
a situation of strategic choice. Let us take, for instance, the group of voters whose
preferences run from the Liberals to the NDP to the Conservatives, group 4 in Table
1. Under what circumstances would those voters have to consider the option of
voting strategically? Of the 13 situations described in Table 2, five meet the condi-
tions we have just specified. In scenarios two and five and 11, voters who prefer
the Liberals perceive them to have the smallest chances of winning the election;
they would have to decide whether they should vote for the NDP, their second
choice, whose chances of defeating the Conservatives, their most disliked party,
look more promising. Likewise, in scenario 9, voters perceive the NDP to be ahead
with the Liberals and Conservatives tied for second place; some of these voters
whose first preference is for the Liberals might reason that they should vote for the
NDP, their second choice, so as to minimize the likelihood of a Conservative
victory. Finally, in scenario 6, voters perceive the NDP to be ahead of the Liberals;
some of them may decide to vote NDP rather than Liberal, because that party is
better positioned to defeat the Conservatives, the most disliked party.
We can, following the same logic, identify the five sets of expectations that lead
to a strategic choice situation for each set of preferences5 This gives us a total of
30 combinations of preferences and expectations, out of a grand total of 78, under
which strategic voting is a real option.
How many voters find themselves in a situation where they have to decide
whether they should vote strategically? In our sample, 535 respondents out of 2689,
or 20 per cent were faced with a strategic choice.6 This is of course a non-negli-
gible group.’ If a majority of them came out voting strategically, that could have
had substantial consequences on the outcome of the election. At the same time,
this figure puts the issue of strategic voting into its proper context. It is only a
relatively small fraction of the electorate that finds itself in the situation of think-
ing about strategic voting.8 For the majority of voters, this is simply not an issue.
As a consequence, the proportion of voters who actually vote strategically has to
be rather small. Previous studies have failed to establish the size of potential strate-
gic voters. Our methodology, which allows us to estimate the size of that pool of
voters who are actually faced with the choice of voting strategically, fills an impor-
tant gap.

3. To Vote or not to Vote Strategically

A strategic vote is a vote for a second-preferred party rather than for the most-
preferred one, motivated by the perception that the former has a better chance of
winning the election. As already indicated, we had 535 respondents in such a situa-
tion in our sample. Among these 535 individuals, 68 per cent voted for the party
they most preferred, 28 per cent voted strategically for their second choice, and 4
 ANDRE Bws ANDRICHARD NADEAU 45

per cent voted for their third choice.9 This yields an estimate of 6 per cent (28 per
cent of the potential pool of 20 per cent) of the electorate voting strategically.
Having estimated the overall magnitude of strategic voting, we may proceed to
examine the factors that increase or decrease the propensity to vote strategically.
Our analysis is still confined to those respondents who were in a situation where
they could potentially cast a strategic vote. We wish to determine what led some
of them to vote for their first choice and others to vote strategically for their second
choice.10
The assumption behind the idea of strategic voting is that voters decide how to
vote on the basis of expected utility maximization, and that that calculus centers
on two major criteria: preferences and evaluations of the various parties’ chances
of winning the election. First, preferences. The propensity to vote strategically
should be higher, the lower the intensity of preference for the fmt choice
over the second and the greater the intensity of preference for the second
choice over the third. The rationale is straightforward. Voters should be more
willing to vote for their second choice, the more acceptable that second choice
appears to them, the closer (in relative preferences) that second choice is to the
first. At the limit, if one likes the second almost as much as the first, he/she is
almost indifferent and should be more prone to switch to the second party in order
not to waste his/her vote. At the opposite, those who see a great difference
between their first two choices should be more reluctant to support a party which
they do not like very much. The logic is reversed when it comes to the compari-
son of second and third choices. The more one dislikes his/her third choice (relative
to the second), the more he/she will be willing to consider the option of voting
for the party that has a better chance of defeating it.
Second, expectations about the election outcome. We should keep in mind that
the group we will be looking at is in a strategic choice situation, that is the party
that is most preferred is perceived to be less likely to win than the second-preferred
party in the constituency. The propensity to vote strategically should be higher,
the greater the differences in the perceived chances of winning of the fmt
and second choices and the greater the perceived closeness of the race
between the second and third choices. Voters should be more prone to vote
for their second choice when it is clear that the first choice is extremely unlikely
to win and that the second choice has much better chances. At the opposite, those
who perceive their first choice’s chances as almost as good as those of their second
choice should be more reluctant to abandon their first choice. Finally, the closer
the race for first place between the second and third choices, the more likely is
one’s vote to make a difference and the more one should consider probabilities of
winning as well as preferences in the vote decision. If the race is not close, one’s
vote is extremely unlikely to matter and one should thus vote on the basis of prefer-
ences alone. The closer the race gets, the more appealing strategic voting becomes.
We now proceed to test these hypotheses. The hypotheses refer to differences in
preferences and perceived chances, among first, second and third choices. Relative
preferences for each of the three parties correspond, as already explained, to the
predicted probability of a respondent voting for a given party, on the basis of his/her
party identification and feelings about parties, leaders and local candidates. These
preference scores range between zero and one. What matters of course is differ-
ences in the scores of the three parties. As in Abramson et al. (1992) we rely on
normalized differences. A score of one was assigned to the difference between the
46 Measuring Strategic Voting: A Two-Step Procedure

first and third choice. What is critical from that perspective is where the second
choice stands. Our first independent variable is thus:

u 2, = u2 - Ul, (1)

where u, = (u, - us)/(ul - US) = 1

U, = (U, - U,)/(U, - u,> = u2 - 4.

U, indicates the preference score for the most preferred party, U, the score for the
second-ranked party and U, the score for the most disliked party. u2, takes into
account the intensity of preferences for each of the three parties and will thus allow
us to test the two hypotheses about the relationship between preferences and
strategic voting.”
Perceived chances of winning for the three parties were measured by the score
given to a party, on a scale from 0 to 100, of its likelihood of winning in the
constituency. Again, what matters is differences in the scores given to the three
parties. As in Abramson et al. (1992) we rely on normalized expectation scores,
where the score for the most preferred party is:

e, = E&E, + E2 + E& (2)

We have two expectation differential variables. First, e,,, which equals e2 - e,.
The variable is always positive, as a voter in a strategic situation always finds
him(_her)self in a situation where his/her first choice is perceived to be trailing in
the race. Second, e(abs),,, which is the difference, in absolute terms, between e3
and e,. The smaller that difference, the closer the race between the second and
third choices, and the greater the predicted propensity to vote strategically.
The dependent variable is a dichotomous variable taking the value of one when
the respondent votes for his/her second choice and zero when he/she votes for
his/her first choice. That dependent variable is regressed on our three basic
independent variables: u,,, e21 and e(ubs),,. Our procedure is similar to the one
followed by Abramson et al. (1992) except for the two following amendments.
First, and foremost, our analysis is confined to those who are faced with a strate-
gic choice situation. As already indicated, there is a compelling logical reason for
doing so. Voting strategically does not merely entail taking into account the
parties/candidates’ chances of winning the election. It entails, more substantially,
not voting for the most preferred party because it is perceived to have less chance
of winning the election than the second choice. It directly flows from this that the
option of voting strategically or not is irrelevant for those whose most preferred
party is not perceived as trailing the second choice.
This has an additional advantage. Because we are looking at those who see their
first choice as trailing in the race, we need not be concerned with projection
effects. These voters are not engaged in ‘wish fulfilment’ since they see their most
preferred party as having little chance of winning. When the analysis includes the
whole electorate, finding that expectations about the race affect the vote may signal
the existence of wish fulhlment or bandwagon effects as well as strategic voting
Abramson et al. (1992, 66) acknowledge the problem and resort to three-stage least
squares estimation to derive ‘purge’ measures of expectations. This introduces, in
our view, unnecessarily complex substantive and technical problems. First, to
 i‘iNDR6Bws ANDhXARD NADEXU 47

identify their model Abramson et al. (1992) had to assume that candidate traits
have no effect on viability ratings (p. 65). Second, the ‘full information’ three-stage
least squares technique they used entails considerable risks as any misspecification
in one equation will be transmitted throughout the system via a biased and incon-
sistent estimate of the residual variance which in turn invalidates statistical tests
(Hausman, 1978, 1264-1265). These projection effects can be more simply and
directly ruled out when we focus on those who find themselves in a situation where
their most preferred party is trailing in the race.‘*
Second, we use the absolute difference in the perceived probabilities of winning
of the second and third choices, whereas Abramson et al. (1992) use relative differ-
ences. Logically, it is the closeness of the race between the second and third
choices that matters and not which one is ahead.‘3 As we have pointed out, a close
race enhances the relative import of each vote and should induce voters to make
strategic calculations.

4. The Results

The results of the logit regression are presented in Table 3.‘* They neatly confirm
our predictions. The propensity to vote strategically increases as the intensity of
preference for the first choice over the second decreases,15 as the perceived
distance in the race between the first and second choices increases,16 and as the
race between the second and third choices gets closer.17 The coefficients obtained
from our two-step procedure are more readily interpretable than ‘purged’ estimates,
and, as expected given our focus on potential strategic voters, the fit of our model
is superior to that of Abramson et al. (1992). Our results also reveal the existence
of a hierarchy of factors in the calculus of strategic voting. Our findings indicate
that expectations lag behind utilities in explanatory power, a fact consistent with
Heath et d’s (1991, 60) observation that “whatever the strict logic of rational
choice may be the main factor [for strategic voting] is the utility difference
between the voter’s two preferred parties.” We believe this hierarchy gives an
accurate picture of the typical reasoning behind strategic voting, leaving unsettled
the question of whether rational choice theorists should take comfort or not in this
characterization.

TABLE 3. The determinants of strategic voting among voters

in a strategic choice situation: LOGIT estimates

Variables

%I 3.06 (0.37)
ezl 2.26 (0.57)
e(aW,, -1.70 (0.53):
Constant 0.80 (0.27)*
-2LLR 528.0
CHF for Covariates 90.5 @=0.0000)
Corrected Pseudo R2 0.27
N 512

Standard errors in parentheses. On the computation of the

corrected pseudo RZ, see Hagle and Mitchell, 1992.
*p 2 0.01 (one-tailed test)
48 Measuring Strategic Voting: A Two-StepProcedure

The coefficient estimates are quite large, and it can be shown that the likelihood
of voting strategically can substantially increase or decrease depending on voters’
sets of preferences and expectations. 29 per cent18 of our sample of potential strate-
gic voters ended up voting strategically. This is in part because most voters much
preferred their first choice over the second; the average distance between the first
two choices was -0.63, indicating that in terms of preferences the second choice
was viewed as only a little better than the third. This is also in part because the
first choice’s chances of winning the election were not perceived to be nil (the
average probability ascribed to the first choice was 24 per cent) and the race (for
fast place) between the second and third choice was often not viewed as close
(the average absolute difference is 0.33).
The typical voter in a strategic choice situation had, as we have seen, a proba-
bility of 0.29 of voting strategically. Let us assume, however, a voter who is typical
in terms of preferences (uzl = -0.63) but who perceives his/her first choice to have
no chance at all of winning and the second and third choices as equally likely to
win; the probability of a strategic vote then increases to 0.50. If we take another
voter with the same expectations but who likes the second choice only a little less
than the first preference (u,, = -0.10) the probability of a strategic choice jumps
to 0234.19
We should end our analysis with a caveat. This has to do with the exclusion of
minor parties in our study. We have restricted our analysis to the three major
parties. In doing so, we have left aside those whose first preference went to a party
other than the Conservatives, the Liberals or the NDP. As the survey on which this
research is based did not tap evaluations of the minor parties, of their leaders, or
of their local candidates, nor perceptions of their chances of winning the election,
it is impossible for us to estimate in any precise way the proportion of voters who
did not vote for a minor party because of strategic considerations. We know,
however, that 5 per cent of voters actually voted for minor parties. It is probably
fair to suppose that some voters had a minor party as their first preference and
voted for one of the major parties because of strategic considerations. As a conse-
quence, the overall percentage of the electorate voting strategically is likely to be
more than 6 per cent.“O

5. Conclusion

We have proposed a procedure for measuring strategic voting in plurality elections.

Our basic approach has been to determine which voters choose to vote for their
second-preferred party (candidate) rather than the one they most prefer because
the former was perceived to have better chances of winning the election. This
approach entails tapping voters’ rank order of preferences as well as their percep-
tions of the various parties’ chances of winning the election. It also entails a two-
step procedure, whereby the pool of potential strategic voters, that is those who
perceive their first-preferred party (candidate) to have smaller chances of winning
than their second option, is first identified, and then the analysis proceeds with an
examination of the actual vote decisions of those who find themselves in a strate-
gic choice situation.
We have shown that in the 1988 Canadian election, at least 6 per cent of the
total electorate can be said to have voted strategically, a figure close to the standard
estimates arrived at for most elections in Britain. More importantly, however, we
 ANDRE Bws ANDRICHARD NADF.AU 49

have shown why strategic voting is not more common. The simple answer is that
the great majority of voters are not faced with the decision to vote strategically or
not. This highlights the importance of proceeding in two stages and of clearly
identifying the pool of potential strategic voters. In the 1988 Canadian election and,
we suspect, in most plurality elections, that pool is limited (see Heath et al., 1991,
58).
Our findings establish that among those who are faced with a strategic choice
situation, a substantial proportion, nearly 30 per cent, does vote strategically.
Furthermore, the propensity to vote strategically increases when the intensity of
preference for the first choice over the second decreases, when the second choice
is perceived to have much better chances of winning than the first choice, and
when the race between second and third choices gets closer, exactly as the logic
of strategic voting would predict. And indeed, our simulations indicate that when
all the conditions are met, most voters do vote strategically.

Notes
1. Furthermore, primaries, contrary to legislative elections, are an ongoing process, and
therefore casting a vote for a given candidate increases that candidate’s chances of
continuing.
2. The principal investigators were Richard Johnston, Andrk Blais, Henry Brady and Jean
C&e. Fieldwork for the study was carried out by the Institute for Social Research, York
University. The CES had three components: a rolling cross-section campaign-period
telephone survey with a representative sample of 3609 Canadians; a post-election
telephone survey with 2922 of the campaign survey respondents; and a self-administered
mailback questionnaire with 2115 of the post-election survey respondents. All inter-
viewing was completed from centralized telephone facilities using Computer Assisted
Telephone Interviewing (CATI) techniques. We use here only the campaign-period
survey, which contains the full set of questions about preferences and expectations
about the outcome of the election. The response rate for that survey was 57 per cent.
Further detail on the sample and on other technical features of the CES can be found
in Johnston et al. (1992).
3. This methodology provides us with a measure of the rank-order and intensity of overall
preferences of one party over another. It is of course a constructed measure. We believe,
however, that this measure, which combines reactions to the parties, leaders and candi-
dates, is less imperfect than those that rely solely on evaluations of the parties. We
should note that our intent here is simply to tap preferences as such and not to indicate
the sources of these preferences, such as positions on issues or evaluations of govern-
ment performance. The results of the logistic regressions used to measure the rank-order
and intensity of partisan preferences are shown in the Appendix.
4. Justifying the use of local probabilities, Black (1978, 616) convincingly made the point
that “given the calculus center-stages the citizen’s impact on the outcome, logically this
impact can only be directly registered at the constituency level”.
5. Here is the full set of combinations that correspond to a strategic choice context (we
identify first the group according to preferences, as listed in Table 1, and second the
situation according to expectations, as listed in Table 2):
- for preference group 1, situations 3, 4, 6, 8, 12;
- for preference group 2, situations 4, 5, 6, 9, 12;
- for preference group 3, situations 1, 2, 5, 7, 11;
- for preference group 4, situations 2, 5, 6, 9, 11;
- for preference group 5, situations 1, 2, 3, 9, 10;
- for preference group 6, situations 1, 3, 4, 8, 10.
6. The situation hardly changed during the course of the campaign. More specifically, 19
per cent of those interviewed in the second half of the campaign were faced with a
strategic choice, compared to 21 per cent in the second half.
50 Measuring Strategic Voting: A Two-StepProcedure

7. Using objective probabilities (derived from electoral results) would inflate that pool since
these probabilities are purged of projection effects whereby individuals overestimate the
winning chances of their preferred party (candidate). But, as mentioned above, it is
subjective expectations, and not objective probabilities, that matter. More specifically,
an individual whose first-choice party (candidate) has little objective chance of winning
but who perceives that party (candidate) to be much stronger than it is really does not
engage in strategic calculus. For this reason, Abramson et al.‘s ‘purged’ estimates of the
proportions of straightforward, strategic, and sincere voters (p. 66) appear somewhat
artificial.
8. It should be kept in mind, however, that the fraction gets larger among some subgroups
of voters. For instance, whereas only 12 per cent of those whose most preferred party
was the Conservatives found themselves in a strategic choice situation, the proportions
reached 24 per cent and 34 per cent for those whose first preference went to the
Liberals and to the NDP respectively.
9. These proportions did not shift during the course of the campaign.
10. We leave aside the few respondents who voted for their third choice.
11, Since we rely on normalized scores, we cannot test the independent impact of prefer-
ence differentials between first and second choice and between second and third choice.
uj2 is by definition equal to 1 - uzl. uzl subsumes the two different evaluations. We also
performed analyses with non-normalized scores. Both non-normalized preference differ-
entials came out significant.
12. It could be argued that those who vote for the second preferred party because it is
perceived to be more likely to win than the most preferred party are not necessar-
ily immune from projection effects, as they may tend to inflate the chances of the
party they support. We have looked, however, at those who perceive the second
preferred party not to be the frontrunner in the race. The proportion voting strate-
gically in that subgroup is the same as the proportion in the total pool of voters in
a strategic choice situation, and the factors inducing a strategic vote turned out to
be the same.
13. Suppose, for instance, that a voter perceives his(her) second and third choices to have
equal chances of winning. This should, everything else being equal, induce that voter
to vote strategically since his(her) vote can make the difference. Suppose, however, that
the second choice is perceived to have much better chances of winning than the third
choice. Under such a circumstance, the voter would be less prone to vote strategically
since his(her) vote is unlikely to make a difference. But the same pattern applies if it is
the third choice that seems much more likely to win. The consequence is the same, the
voter’s vote is unlikely to make a difference and strategic voting is less appealing. It is
thus the closeness of the race as such that matters and not whether it is the second or
the third choice that is ahead.
14. The choice of an additive specification, rather than a multiplicative one, has been made
on empirical grounds. Details on the estimation of alternative models are available upon
request.
15. It must be remembered that u2, is always negative. The positive coefficient in Table 3
indicates that the less negative is u,,, that is the lower the intensity of preference for
the first choice over the second, the greater is the probability of casting a strategic vote.
16. Recall that the regression pertains only to those whose first choice is perceived to be
trailing in the constituency. The more that first choice is perceived to be trailing, relative
to the second choice, the more inclined one is to vote strategically for the second
choice.
17. The greater the absolute difference between the second and third choices (the smaller
the probability of casting the decisive vote), the smaller the probability of voting strate-
gically. The relatively high correlation between e,, and e(a6s)szr 0.65, makes the finding
about the latter variable more tentative.
18. This is different from the 28 per cent reported at the beginning of the section because
we have dropped the 4 per cent who voted for their third choice.
19. On the method to calculate predicted probabilities, see DeMaris, 1992, 47-48.
20. Our estimate of the amount of strategic voting does not uncover all the second-prefer-
ence voting in our sample. Among those who were not in a strategic choice situation,
we find 11 per cent voting for their second-preferred party. Second-preference, however,
 fiNDI& Bws ANDRICHARD NAIXAU 51

is almost three times higher among those facing a strategic choice situation, a fact that
buttresses our confidence that the two-step procedure provides a sound measure of strate-
gic voting in its standard understanding by political scientists.

References

Abramson, Paul R., Aldrich, John H., Paolino, Phil and Rohde, David. (1992) ‘Sophisticated’
Voting in the 1988 Presidential Primaries, American Political Science Review, 86 55-70.
Bartels, Larry M. (1988) Presidential Primaries and the Dynamics of Public Choice.
Princeton University Press, Princeton.
Black, Jerome H. (1978) The Multicandidate Calculus of Voting: Application to Canadian
Federal Elections, American Journal of Political Science, 22 609-638.
Blais, Andre and Catty, Ken (1991) The Psychological Impact of Electoral Laws: Measuring
Duverger’s Elusive Factor, British Journal of Political Science 21 79-93.
Blais, Andre, Johnston, Richard, Brady, Henry E. and Crete, Jean (1990) The Dynamics of
Horse Race Expectations in the 1988 Canadian Election. Paper presented at the Annual
Meeting of the Canadian Political Science Association, Victoria, BC.
Brady, Henry E. and Johnston, Richard (1987) What’s the Primary Message: Horse Race or
Issue Journalism? In Media and Momentum, Gary R. Orren and Nelson W. Polsby (Eds),
Chatham House, Chatham.
Cain, Bruce E. (1978) Strategic Voting in Britain, American Journal of Political Science, 22
639-655.
Curtice, John and Steed, Michael (1988). Appendix 2, in David Butler and Dennis Kavanagh,
me British General Election of 1987. St-Martin’s Press, New York.
DeMaris, Alfred (1992) Logit Modeling. Sage, Beverly Hills.
Downs, Anthony (1957) An Economic meory of Democracy. Harper and Row, New York.
Evans, Goeffrey and Heath, Anthony (1993) A Tactical Error in the Analysis of Tactical Voting:
A Response to Niemi, Whitten and Franklin, British Journal of Political Science, 23 13 1- 137
Ferejohn, John A. and Fiorina, Morris P. (1974) The Paradox of Not Voting: A Decision
Theoretic Analysis, American Political Science Review, 69 920-936.
Ferejohn, John A. and Fiorina, Morris P. (1975) Closeness Counts Only in Horseshoes and
Dancing, American Political Science Review, 68 920-925.
Fisher, S.L. (1973) The Wasted Vote Thesis: West German Evidence, Comparative Politics,
6 295-299.
Galbraith, John W. and Rae, Nicol C. (1989) A Test of the Importance of Tactical Voting:
Great Britain, 1987, British Journal of Political Science, 19 126-136.
Hagle, Timothy M. and Mitchell, Glenn E. (1992) Goodness-of-fit Measures for Probit and
Logit, American Journal of Political Science, 36 762-784.
Hausman, Jerry A. (1978) Specification Tests in Econometrics, Econometrica, 46 1251-1271.
Heath, Anthony, Jewell, Roger, Curtice, John, Evans, Goeffrey, Field, Julia and Witherspoon,
Sharon (1991) Understanding Political Change: 7&e British Voter 1964-1987. Pergamon
Press, Oxford.
Johnston, Richard J., Blais, Andre, Brady, Henry E. and Crete, Jean (1992) Letting the PeopZe
Decide: Dynamics of a Canadian Election. McGill-Queen’s University Press, Montreal.
Johnston, R.J. and Pattie, C.J. (1991) Tactical Voting in Great Britain in 1983 and 1987: An
Alternative Approach, British Journal of Political Science, 21 95-128.
Meehl, Paul E. (1977) The Selfish Voter Paradox and the Thrown-Away Vote Argument,
American Political Science Review, 71 11-30.
Niemi, Richard G., Whitten, Guy and Franklin, Mark N. (1992) Constituency Characteristics,
Individual Characteristics, and Tactical Voting in the 1987 British General Election, British
Journal of Political Science, 22 229-254.
Riker, William H. and Ordeshook, Peter C. (1968) A Theory of the Calculus of Voting,
American Political Science Review, 62 23-43.
Spafford, Duff (1972) Electoral Systems and Voter’s Behavior, Comparative Politics, 5
129-134.
52 Measuring Strategic Voting: A Two-Step Procedure

Appendix

Logistic regressions of vote intentions on thermometers of leaders, parties and local

candidates

Variables Conservative Liberal Party New Democratic

Party Party

Conservative Party 12.4 (0.78)t -3.4 (0.63)t -12.6 (0.87)t

Liberal Party 0.1 (0.65) 12.5 (0.79)t -15.1 (0.91)t
Conservative leader 8.3 (0.94.t 1.4 (0.70)* -4.9 (0.82)t
Liberal leader -1.9 (0.83) 7.4 (0.84)t -3.5 W3O>t
Conservative candidate 7.0 (1.49)t 4.3 c1.m -12.2 (1.6)t
Liberal candidate -1.3 (1.60) 12.9 (l.G>t -14.7 (1.8)t
Constant -9.3 (1.02)t -12.7 (0.93)t 19.3 (1.2)1-
Corrected Pseudo RZ 0.65 0.57 0.62
N 2765 2765 2765

Entries are logit coefficients and standard errors. All the independent variables are normal-
ized thermometer scores.
*p 5 0.05
tp I 0.01