You are on page 1of 21

ARTICLE IN PRESS

Electoral Studies XX (2003) XXXXXX www.elsevier.com/locate/electstud

Time-of-voting decision and susceptibility to campaign eects


Blais a, Patrick Fournier a,, Richard Nadeau a, Andre Elisabeth Gidengil b, Neil Nevitte c
a

partement de science politique, Universite de Montre al, Montre al, Que bec, Canada H3C 3J7 De b al, Que bec, Canada H3A 2T7 Department of Political Science, McGill University, Montre c Department of Political Science, University of Toronto, Toronto, Ontario, Canada M5S 3G3

Abstract There is mounting evidence that election campaigns matter. There are also reasons to expect interpersonal heterogeneity in the susceptibility to campaign inuence. Time-of-voting decision has been suggested as a key mediating variable for campaign eects. However, there is no persuasive empirical evidence to substantiate the claim that people who decide during campaigns actually respond to campaign events or campaign-specic information. This study incorporates time of decision into dynamic models of campaign eects in order to test whether there is a signicant interaction eect between time of decision and campaign persuasion. In sum, the vote intentions of campaign deciders are indeed more volatile because they respond to actual campaign events and coverage, not because they uctuate haphazardly. People who say they decided before the campaign are, reassuringly, not inuenced by campaigns. # 2003 Elsevier Ltd. All rights reserved.
Keywords: Time of decision; Vote intentions; Campaign eects; Heterogeneity

There is mounting evidence that election campaigns matter. They aect individuals opinions about parties, candidates and issues, the determinants of their voting decisions, and their vote intentions (Bartels, 1988, 1992; Blais et al., 1999; Holbrook, 1996; Johnston et al., 1992; Johnston et al., 1996; Just et al., 1996; Mendelsohn, 1994, 1996). However, it is reasonable to assume that campaigns do not aect all citizens similarly. A substantial body of evidence indicates that there is

Corresponding author. E-mail address: patrick.fournier@umontreal.ca (P. Fournier).

0261-3794/$ - see front matter # 2003 Elsevier Ltd. All rights reserved. doi:10.1016/j.electstud.2003.09.001

ARTICLE IN PRESS
2 P. Fournier et al. / Electoral Studies XX (2003) XXXXXX

interpersonal heterogeneity in the determinants of political decisions (Rivers, 1988; Sniderman et al., 1991; Johnston et al., 1996; Fournier, 2000).1 There are reasons to expect similar interpersonal heterogeneity in the susceptibility to campaign inuence. Time-of-voting decision has been suggested as a key mediating variable for campaign eects (Lazarsfeld et al., 1944; Berelson et al., 1954; Chaee and Choe, 1980; Whitney and Goldman, 1985; Bowen, 1994; Gopoian and Hadjiharalambous, 1994; Chaee and Rimal, 1996; Box-Steensmeier and Kimball, 1999). But, there is hardly any empirical evidence to substantiate the claim that people who decide during campaigns actually do respond to campaign events and/or campaign-specic information. This study incorporates time of decision into dynamic models of campaign eects in order to assess the extent to which campaign deciders really do make up their mind as a result of the campaign itself. First, it looks at the sociodemographic and attitudinal characteristics of campaign deciders to ascertain whether there are reasons to believe that this group is more susceptible to campaign persuasion. Then, it examines the extent to which those who decide during campaigns are inuenced by two major types of campaign factors: leaders debates and media coverage. Such an investigation can help us better understand campaigns, voters and political party activity.

1. The context Many studies contend that the time-of-voting decision is a key predisposition which mediates campaign eects (Lazarsfeld et al., 1944; Berelson et al., 1954; Chaee and Choe, 1980; Whitney and Goldman, 1985; Bowen, 1994; Gopoian and Hadjiharalambous, 1994; Chaee and Rimal, 1996; Box-Steensmeier and Kimball, 1999). Individuals who make up their mind about which candidate or party to support before the campaign even begins should be less likely to switch their vote choice in response to events and messages occurring during a campaign. This is because their decisions tend to be determined by long-term factors such as partisan identication and ideology. In contrast, those who are not committed to a choice before the start of a campaign should be much more susceptible to campaign persuasion. Whether they are moving from indecision to decision, or from one choice to another, campaign deciders could be responding to campaign events and messages. However, the presence of campaign eects among campaign deciders can not be assumed. Individuals who decide for whom to vote during a campaign could simply recognize the need to make a decision, and side with their underlying predispositions without being inuenced by campaign events and messages. Campaign
1 Individuals do not reason the same way about political choices: they rely on dierent considerations, or they give dierent weight to similar considerations. Models which fail to take into account this variation provide incomplete and even erroneous explanations of these political decisions.

ARTICLE IN PRESS
P. Fournier et al. / Electoral Studies XX (2003) XXXXXX 3

time-of-voting decision is not, by itself, a sucient condition for greater susceptibility to campaigns. To be inuenced by a political message, a person has to be exposed to it, has to receive it, and has to accept it (McGuire, 1969; Zaller, 1992). Being undecided or uncertain about ones choice is not sucient for persuasion to occur. One must experience certain conditions which favor exposure to, reception of, and acceptance of the message. Political interest, media attention, political information, and partisan independence all favor persuasion. Interest and attention render an individual more likely to be aware that a campaign message is in the air. Information makes successful interpretation and understanding of the message and of its context more likely. Information also increases the ability to accept information in line with previous beliefs. Partisan independence defuses some of the lters which reject dissonant information, that is, information which is not in concordance with the persons prior beliefs. Do campaign deciders exhibit characteristics which favor exposure to, reception of, and acceptance of political stimuli? American data do reveal sociodemographic and attitudinal dierences between time-of-voting decision subgroups (Lazarsfeld et al., 1944; Berelson et al., 1954; Chaee and Choe, 1980; Whitney and Goldman, 1985; Gopoian and Hadjiharalambous, 1994; Chaee and Rimal, 1996). While all studies conclude that people who decide during campaigns are less attached to political parties than pre-campaign deciders, the evidence is conicting in most other regards. Two groups of campaign deciders appear to exhibit dierent characteristics inconsistently. Early campaign deciders often express high interest in the campaign, high attention to media coverage, and high levels of political information; while late campaign deciders are generally uninterested, inattentive, and uninformed (Chaee and Choe, 1980; Gopoian and Hadjiharalambous, 1994). However, it seems that races with a strong third candidate display the opposite pattern (Whitney and Goldman, 1985; Chaee and Rimal, 1996). Despite these mixed results, have campaign deciders actually been found to exhibit greater susceptibility to campaign inuence? While there is some evidence on the persuasibility of campaign deciders, it is rather scant and indirect. One strand of studies has shown that campaign deciders possess higher levels of attention to political ads and to campaign coverage in newspapers and on television (Chaee and Choe, 1980; Bowen, 1994; Chaee and Rimal, 1996). Another stand has demonstrated that there are dierences in vote choice between time-of-decision groups (Chaee and Rimal, 1996; Abramson et al., 1998; Box-Steensmeier and Kimball, 1999). Nevertheless, these results do not constitute persuasive evidence of the assertion that campaign deciders are inuenced by campaign messages or by campaigns events. Greater attention only taps openness to persuasion. Dierences in vote choice are not symptoms of persuasion if they are not linked to actual campaign events or coverage. Serious tests which use dynamic data are needed to provide direct evidence that campaign events and coverage actually inuence the vote intentions of campaign deciders. In light of the unresolved state of the debate on the interpersonal variability in the susceptibility to campaign inuence, it is important to further examine the issue. This study reassesses the notion that time of decision is an important

ARTICLE IN PRESS
4 P. Fournier et al. / Electoral Studies XX (2003) XXXXXX

mediator of campaign persuasion. It uses Canadian data which allows us to conduct more appropriate tests of the claim that campaign deciders are strongly aected by campaigns.

2. The study The analysis relies on two sources of data: the 1997 Canadian Election Study (CES) and a content analysis of televised news broadcasts during the 1997 Canadian election. The rolling cross-sectional component of the CES surveys is used to conduct analyses of the impact of the leaders debates and of daily media coverage on campaign vote intentions.2 The data on the tone of all daily stories about each political party are drawn from a content analysis of the nightly news broadcasts on three major television networks.3 To capture time-of-voting decision, we use answers to a recall question. In the post-election wave of the 1997 CES, respondents were asked: When did you decide that you were going to vote X? About half the respondents claimed to have made up their mind prior to the start of the ocial 36-day campaign.4 Because it is a post-election measure, people who decided before the campaign should not, by denition, have changed their mind in response to the campaign. However, when respondent recall is used to operationalize a behavior, the validity of the responses can be dubious. If there is misreporting of time-of-voting decision, for whatever reason, some of those who say they decided before the campaign could actually have been aected by the campaign (and vice versa). Since the quality of the measurement of the key mediating variable is important, the rst section of the analysis examines the validity of answers to the time-of-voting-decision question by comparing them to the behavioral pattern revealed by the rolling cross-sectional panel.
2 The 1997 Canadian Election Study was funded by the Social Sciences and Humanities Research Council of Canada. The surveys contain a rolling cross-sectional component and a panel component. A representative sample of 80140 respondents (110, on average) was interviewed each day of the 36-day campaign to track the dynamics of vote intentions. Respondents from the campaign wave were reinterviewed twice after the election, once by telephone, and once by mailback questionnaire. A total of 3949 eligible voters were surveyed during the campaign, 3170 with the post-election telephone interview, and 1851 with the mailback questionnaire. The response rate for the campaign survey was 59%, and the average daily sampling error is about 11 percentage points. The survey was conducted by the Institute for Social Research at York University. Copies of the questionnaires, technical documentation and data can be obtained at http://www.fas.umontreal/pol/ces-eec and through the Inter-university Consortium for Political and Social Research. See Nevitte et al. (2000). 3 The media content analysis dealt with late-night news bulletins on three networks (CBC, CTV, and SRC). The news stories were analysed by groups of coders made up of undergraduate university students. Coders were asked to rate each story on a three-point scale, depending on whether treatment of the party in question was favorable (+1), neutral (0), or unfavorable (1). This coding was done for each nights television news coverage. 4 For details on the construction of these and other variables, see the Appendix.

ARTICLE IN PRESS
P. Fournier et al. / Electoral Studies XX (2003) XXXXXX 5

The second section of the analysis identies the sociodemographic and attitudinal characteristics of pre-campaign and campaign deciders to determine whether a dierence in susceptibility to campaign persuasion could have been expected between the two groups on theoretical grounds. The third section of the analysis examines to what extent campaign deciders are indeed aected by campaign events (leaders debates) and by campaign information (media coverage). At the same time, we will ascertain whether there is truly an absence of campaign eects among those classied as pre-campaign deciders as a further validation of time-of-voting-decision recall.

3. The analysis 3.1. Validation of reported time of decision Is reported time of decision an accurate indicator of actual behavior? Recall questions have been found to be an unreliable measure of time-of-voting decision with American data (Plumb, 1986; Chaee and Rimal, 1996). The results of the 1997 CES reect a dierent pattern. Table 1 reports the stability in vote choice between pre- and post-election waves of the 1997 CES according to reported time of decision (before or during the campaign). Individuals are categorized as stable if they report having voted for the party they were intending to support in the campaign survey. Individuals are categorized as having changed their opinion if they report having voted for a party other than the one indicated by their campaign vote intention. Individuals are categorized as having formed their opinion if they were undecided during the campaign and report that they voted in the post-election interview.5 Since the pre-election interview occurred anywhere from 36 to one day prior to election day, campaign deciders could report having decided before or after their pre-election interview. To properly capture the dierences in stability between groups, we split respondents according to the moment of the pre-election interview. If responses to the question about time of decision are accurate, we should observe instability in vote choice before the reported time of decision, and stability afterwards. The data indicate that there is great stability (84%) among individuals who claim to have decided before the campaign. Few made up their mind (12%) during the campaign, and even fewer changed their vote intention (4%). Highly stable vote choices (77%) are also found among campaign deciders interviewed after their reported time of decision. By contrast, there is little stability (26%), and much formation (55%) and change (19%) of opinion among those interviewed before their reported time of decision. Table 2 shows that the link between time of decision and stability of vote choice holds when we control for strength of partisan identication, political sophistication, and various sociodemographic characteristics. Respondents who say they
5

Refusals and abstainers are excluded from the analysis.

ARTICLE IN PRESS
6 P. Fournier et al. / Electoral Studies XX (2003) XXXXXX Table 1 Stability of electoral choices according to time of decision and time of interview Time of interview Time of decision Before the campaign During the campaign Change of opinion (%) 19 15 5

Stability of Formation Change of Stability Formation of opinion opinion of opinion of opinion opinion (%) (%) (%) (%) (%) Before time of decision During time of decision After time of decision 84 12 4 26 46 77 55 39 18

Table 2 Logistic regression of electoral choice stability b (standard error) Time of decision After campaign interview (0/1) Before campaign interview (0/1) (Missing category during campaign interview period) Day of campaign interview (136) Strength of party identication (01) Campaign political information (01) General political information(01) Interest in the election (01) Interest in politics (01) Media attention (01) Political discussions with relatives, friends or other people (01) Most important source of election information Family and friends (0/1) Radio (0/1) Newspapers (0/1) Age (18101) Gender: women (0/1) Education (01) Income (01) Region Atlantic (0/1) British Columbia (0/1) Prairies (0/1) Quebec (0/1) Constant Percent predicted correctly Chi-square (degrees of freedom) Number of cases

0.89 (.20) 1.70 (.18) 0.01 (.01) 1.57 (.21) 0.41 (.26) 0.42 (.28) 0.30 (.34) 0.12 (.36) 0.80 (.37) 0.03 (.28) 0.70 (.32) 0.36 (.20) 0.18 (.17) 0.00 (.00) 0.09 (.14) 0.45 (.37) 0.17 (.24) 0.22 (.27) 0.08 (.26) 0.08 (.20) 0.01 (.18) 1.88 (.47) 78.7 % 547.4 (21) 1475

Signicant at the 0.10 level; signicant at the 0.05 level; signicant at the 0.01 level.

ARTICLE IN PRESS
P. Fournier et al. / Electoral Studies XX (2003) XXXXXX 7

decided before their pre-election interview exhibit greater stability, while those who were interviewed before they made their nal decision are considerably more likely to have changed their vote choice.6 Reported time of decision appears highly reliable within the context of a short parliamentary election campaign.7 Some people do respond erroneously. For instance, a small percentage of people who say they decided before the campaign began actually switched sides during the campaign (4%). But overall, the recall question provides a good approximation of the moment of decision. Most individuals truly move from indecision to decision, or from one choice to their nal decision, at the time which they report as their moment of decision.8 Beyond the problem of inaccurate recall, there is another measurement issue.9 What is the time of decision of individuals who have not quite made up their mind but who are leaning towards party A at the start of the campaign, whose initial preference is reinforced by campaign events, and who end up casting a vote for party A? Faced with the time-of-decision question, these respondents could answer that they decided either before or during the campaign. On one hand, they were aected by campaign events. On the other hand, their vote inclination remained unchanged. This ambiguity introduces two possible biases. If these individuals consider themselves pre-campaign deciders, then we will underestimate the number of campaign deciders. If they consider themselves campaign deciders, then we will underestimate the strength of campaign eects among campaign deciders. Since our analyses rely on attitude change, and typical measures of vote choice only document complete attitude change (i.e. switching preferences), the data does not capture evidence of campaign inuence among these individuals. To be able to uncover campaign eects without attitude reversalsuch as reinforcementwe would need more sensitive measures of vote choice, such as respondents probability of voting for each party. Is it possible to ascertain whether reinforced individuals tend to respond one way more than the other to the time-of-decision question and whether the above biases are large or small? Not completely. However, we can isolate empirically a group which can be construed as reinforced individuals: people who voted for the party they were initially only leaning towards. Of the 16% of respondents who are campaign leaners, the vast majority (77%) declare they decided during the campaign. Moreover, the proportion of campaign deciders among leaners is high
The reference group is made up of those who said they decided around the time they were interviewed. 7 The lower reliability of time-of-decision recall in the United States is likely attributable to the longer length of American campaigns and the heavier burden that an extended period imposes on memory. In fact, results comparable to those based on Canadian data are obtained if a period of similar length in American presidential campaigns is considered, namely the post-convention period (Fournier et al., 2001). 8 For a more extensive and detailed demonstration of the validity of reported time of decision in the 1997 CES, see Fournier et al. (2001). 9 This paragraph was inspired by questions raised by two anonymous referees.
6

ARTICLE IN PRESS
8 P. Fournier et al. / Electoral Studies XX (2003) XXXXXX

whether they switched sides (85%) or not (72%). These results suggest that, in all likelihood, the campaign eects uncovered below among campaign deciders represent conservative estimates of the true eects. 3.2. Who are the campaign deciders? The American evidence on the characteristics of early and late deciders is conicting. What do our respondents look like? Table 3 reports the average value of pre-campaign and campaign deciders on a variety of sociodemographic and attitudinal characteristics.10 Women and younger respondents were more likely to decide during the campaign. But pre-campaign and campaign deciders have almost identical levels of education and income. In terms of political attitudes, pre-campaign and campaign deciders are surprisingly similar. Campaign deciders are slightly less interested in the election and in politics in general, slightly less attentive to media coverage, and slightly less knowledgeable about campaign-specic and general political facts. Even though some of these dierences are statistically signicant, they are very small. The only important dierence between the two groups concerns party identication: campaign deciders are much less committed to a particular party. The average strength of partisan identication is much higher among pre-campaign deciders (0.60) than among campaign deciders (0.41).11 More than 72% of pre-campaign deciders identication with a party is fairly or very strong, and only 16% do not identify with any party. The corresponding numbers for campaign deciders are 46% and 34%, respectively. Even those who decided on election day resemble those who decided before the campaign began (results not reported here). Election day deciders are less attentive to media coverage and are less informed about the parties campaign promises for most of the campaign, but they make up for these shortfalls later on; the data indicate that they experienced extensive gains in attention and campaign information during the latter part of the campaign. They also exhibit levels of political interest and general political information comparable to those of other respondents.12 Pre-campaign and campaign deciders are similarly interested in politics, similarly attentive to media coverage, similarly informed about general and campaign political facts, but they dier signicantly in their level of partisan commitment. As a result of this last key discrepancy, the two groups should not react similarly to campaign stimuli. Campaign deciders constitute more promising subjects for
10 We dichotomize time of decision into two categories (those who decided before the campaign, and those who decided during the campaign), because more complex categorizations did not reveal major dierences among campaign deciders. The briefness of campaigns in Canada may explain why the distinction between early and late campaign deciders is not as important in the Canadian case as it is in the US. 11 On this scale, 1 represents strong partisan identication, 0.67 represents fairly strong identication, 0.33 represents not very strong identication, and 0 represents no identication. 12 Analysis of 1993 Canadian Election Study data revealed the same absence of dierences between time of decision groups.

ARTICLE IN PRESS
P. Fournier et al. / Electoral Studies XX (2003) XXXXXX 9 Table 3 Average sociodemographic and attitudinal characteristics by time of decision (means and standard deviations) Decided before the campaign Mean Woman (0/1) Year of birth (18961978) Education (01) Income (01) Interest in election (01) Interest in politics (01) Media attention (01) Campaign political information (01) General political information (01) Strength of party identication (01)

Decided during the campaign Mean 0.55 1955 0.56 0.47 0.63 0.57 0.41 0.33 0.65 0.41 SD 0.50 15.1 0.21 0.31 0.25 0.27 0.24 0.30 0.28 0.34

Dierence

SD 0.50 16.3 0.21 0.31 0.25 0.25 0.25 0.31 0.29 0.33

0.47 1951 0.54 0.46 0.65 0.61 0.45 0.36 0.67 0.60

0.08 4 0.02 0.01 0.02 0.04 0.04 0.03 0.02 0.19

Signicant at the 0.10 level; signicant at the 0.05 level; signicant at the 0.01 level.

campaign eects. They are more available to campaign messages, because they are equally reachable but more receptive than pre-campaign deciders. Time-of-voting decision thus appears to be a useful variable to distinguish the potential subjects of campaign eects, since it identies relatively interested, attentive, and informed individuals that are less hampered by the blinders of partisan commitment. Next, we test that claim empirically. 3.3. Time of decision and susceptibility to campaign eects The evidence that campaign deciders are strongly aected by campaigns is rst analysed, in a crude fashion, in Table 4. Table 4 shows the dierences in vote share between the rst and the last seven days of the 1997 campaign for both pre-campaign and campaign deciders. The evidence supports the expectation that campaign deciders exhibited a good deal of volatility in vote intentions. The absolute changes
Table 4 Parties vote share (%) during rst seven and last seven days of campaign, by time of decision Liberal Decided before the campaign First seven days Last seven days Dierence 37.3 35.4 1.9 Conservative NDP 16.4 16.1 0.3 24.6 30.0 5.4 11.6 9.5 2.1 5.6 8.4 2.8 Reform 17.2 24.8 7.6 11.2 19.2 8.0 Bloc 17.2 11.8 5.4 8.9 7.1 1.8

Decided during the campaign First seven days 48.0 Last seven days 33.3 Dierence 14.7

Signicant at the 0.10 level; signicant at the 0.05 level; signicant at the 0.01 level.

ARTICLE IN PRESS
10 P. Fournier et al. / Electoral Studies XX (2003) XXXXXX

in aggregate vote intentions between the beginning and the end of the campaign are statistically signicant for two parties among campaign deciders. These results also oer further validation of the time-of-decision measure. If their postelection recall is correct, pre-campaign deciders should have expressed the same vote choice before and after the election. In four of the ve cases, the absolute change in aggregate vote intentions is smaller for those who classied themselves as pre-campaign deciders, and only one of the changes is statistically signicant. That campaign deciders exhibit greater volatility in vote intentions is not surprising. By denition, campaign deciders are likely to have made up their mind during the campaign (whether they are moving from indecision to decision, or from one choice to another). This evidence, however, much like that of Chaee and Rimal (1996) and Abramson et al. (1998), does not show that campaign deciders are responding to specic campaign events or messages. To determine whether this is the case, we turn to two additional tests. The rst pertains to campaign events, and the second concerns campaign information. Electoral debates are sometimes responsible for substantial shifts in vote intentions (Johnston et al., 1992; Blais and Boyer, 1996). It has been shown that the leaders debates in the 1997 Canadian federal election campaign had notable eects on the vote intentions of the entire electorate (Blais et al., 1999): they had a positive but temporary impact on support for the Conservatives, the party whose leader was perceived to have emerged victorious from the debates.13 Fig. 1 suggests, however, that the impact of the debates, which were held on May 1213, was limited to campaign deciders. Conservative vote intentions increased substantially after May 13 only among that group. Table 5 presents the results of logistic regression analyses of the impact of the leaders debates on Conservative vote intentions (0/1) among then entire population, and among pre-campaign (for validation purposes) and campaign deciders separately. Following the form established by Blais and colleagues (1996, 1999), there are two key independent variables: the number of days since the debates occurred (020), ascertained for each respondent according to his/her date of interview, and the square of the previous variable (0400) to capture the presence of a quadratic function. This specication is applied to all respondents, not only those who saw the debates, to accommodate the possibility (Blais and Boyer, 1996; Blais et al., 1999) that people who do not witness the debates are aected indirectly by media coverage of the events and by word of mouth.14 The model also contains controls for partisan identication, political interest, and various sociodemographic aliations. Table 5 conrms that support for the Conservatives among campaign deciders was signicantly aected by the debates, and that those claimed to be pre-campaign
There were two debates: one in English, and one in French. The former took place on May 12, and the latter occurred the following day. 14 Further analyses, not reported here, indicate that viewers were more inuenced by the leaders debates than non-viewers, but that the later group was also aected by the debates.
13

ARTICLE IN PRESS
P. Fournier et al. / Electoral Studies XX (2003) XXXXXX 11

Fig. 1. Evolution of Conservative vote intentions by time of decision (5-day moving averages).

Table 5 Impact of debates on conservative vote intentions by time of decision (logistic regression) All respondents b (standard error) Debate Debate2 Party identication 21 other controls Constant Number of cases Percent correctly predicted

Decided before campaign b (standard error) 0.06 (0.08) 0.003 (0.004) 4.88 (0.40) 2.78 (0.68) 776 92.6

Decided during campaign b (standard error) 0.11 (0.06) 0.006 (0.003) 2.98 (0.29) 1.90 (0.51) 649 81.1

0.07 (0.03) 0.003 (0.002) 3.24 (0.16) 2.62 (0.31) 2249 86.5

Signicant at the 0.10 level; signicant at the 0.05 level; signicant at the 0.01 level. 21 other controls: interest in politics, region: Atlantic, region: Quebec, region: Prairies, region: British Columbia, gender, traditionally married, baby boomer, gen X, rural, high school dropout, university graduate, employed, public sector, union household, catholic, no religion, north European, non European, low income, high income.

deciders were not inuenced. The coecient for the events main eect is larger among campaign deciders than among pre-campaign deciders. In fact, it is almost twice as large. The dierence between the two groups is particularly evident when we simulate the predicted impact of the leaders debates on Conservative support15 According to our estimates, the Conservatives gained as much as ve
We used the regression equation to generate predicted values for each day of the campaign. We only manipulated the day of interview, all of the other characteristics of the sample remained intact.
15

ARTICLE IN PRESS
12 P. Fournier et al. / Electoral Studies XX (2003) XXXXXX

percentage points (nine days after the debates) among campaign deciders. In contrast, they only reached a maximum gain of one percentage point (eight days after the debates) among pre-campaign deciders. In both cases, however, the gain had dissipated by election day. Nevertheless, these results conrm that support for the Conservatives among respondents who classied themselves as campaign deciders was signicantly aected by the debates, and that pre-campaign deciders were not aected. Mass media coverage is generally believed to aect public opinion indirectly through phenomena such as agenda-setting (McCombs and Shaw, 1972; Iyengar et al., 1984; Iyengar and Kinder, 1987) and priming (Krosnick and Kinder, 1990; Krosnick and Brannon, 1993). But there is little evidence that media coverage has a direct persuasive impact on vote choice (Klapper, 1960; McGuire, 1969; Miller and Krosnick, 1996; Zaller, 1996). Yet, that is precisely what we consider here: that media coverage directly inuences the vote intentions of campaign deciders. Fig. 2 presents the most clear-cut illustration of this pattern in the 1997 campaign. Reform support among campaign deciders closely follows the ups and downs in the tone of the partys coverage in televised news. In contrast, the evolution of vote intentions among those who said they decided before the campaign only vaguely resembles the dynamics of coverage. Table 6 reports the ndings from a multivariate test of these relationships.16 Included in the model to explain individual Reform vote intentions (0/1) are a three-day moving average of the partys tone of coverage (11), ascertained for each respondent according to his/her date of interview, and controls for partisan identication, political interest, and sociodemographic aliations.17 The results of the logistic regressions show that the pattern is robust. Coverage is positively and signicantly related to the evolution of vote intentions among campaign deciders (third column), but not among those who indicated that they were pre-campaign deciders (second column). Any estimation which covers the entire electorate (rst column) would miss these dierent relationships and erroneously indicate that there is no link between coverage and vote choice. Simulations indicate that, compared to the average tone of coverage (0.28) received by the party, Reforms most positive (0.00, obtained several days) and most negative (0.67, also obtained several days) coverage brought/cost the party ve percentage
Since Reform did not present candidates in Quebec, resident of that province are excluded from the estimation. 17 The tone of each partys nightly network news coverage was ascertained by a group of untrained undergraduate students. These untrained coders were simply asked to give a positive rating to a news item they considered positive, a negative rating to one they felt is negative, and a neutral rating to one they deemed neither positive nor negative. The logic is that untrained coders provide assessments which are more representative of the general populations likely reactions. It has been shown that untrained coders evaluations are quite similar to those of trained coders, though slightly more positive (Nadeau et al., 1998). Further analyses, not reported here, reveal that trained and untrained coders evaluations of media coverage are not dierently correlated to vote choice.
16

ARTICLE IN PRESS
P. Fournier et al. / Electoral Studies XX (2003) XXXXXX 13

Fig. 2. Evolution of reform vote intentions by time of decision (5-day moving averages).

points among campaign deciders.18 Reform support among those identied as pre-campaign deciders was hardly aected by the tone of coverage, moving by a tenth of a percentage point in each direction. To provide a complete picture of the mediating role of time-of-voting decision on campaign persuasion, we performed stacked logistic regressions for all four parties outside Quebec (see Table 7).19 Four pairs of debates variables are included to allow the impact of the debates to vary across parties. Since, theoretically, tone of coverage should always be positively correlated to the vote intentions of a party, a single (stacked) media coverage variable is used in the model. Also present in the estimation are controls for partisan identication, political interest, sociodemographic aliations, and party dummies. Table 7 shows that the positive eect of the debates on Conservative vote intentions among campaign deciders holds when controlling for media coverage. In fact, the eect is even stronger. The results also conrm that the Conservative gains following the debates were made at the Liberals expense (Blais et al., 1999). They further reveal that the Liberal losses only occurred among campaign deciders. Support for the Liberals among people who said they decided before the campaign was not negatively aected by the debates. The debates did not, however, signicantly aect support for the New Democratic Party and the Reform Party among either time-of-voting decision subgroups.
18 We used the regression equation to generate predicted values for specic quantities of tone of coverage. We only manipulated the tone of coverage, all of the other characteristics of the sample remained intact. 19 Quebec and the rest of Canada are treated separately because they experienced dierent media coverage. A pooled analysis conducted on the Quebec sample produced similar results (not reported here).

ARTICLE IN PRESS
14 P. Fournier et al. / Electoral Studies XX (2003) XXXXXX Table 6 Impact of media coverage on vote intentions by time of decision (logistic regression) All respondents b (standard error) Tone of media coverage Party identication 20 other controls Constant Number of cases Percent correctly predicted 0.23 (0.31) 4.62 (0.36) 1.50 (0.33) 1605 86.4 Decided before campaign Decided during campaign b (standard error) 0.03 (0.60) 5.20 (0.63) 1.52 (0.61) 548 88.0 b (standard error) 1.37 (0.62) 4.21 (0.67) 1.78 (0.68) 440 86.2

Signicant at the 0.10 level; signicant at the 0.05 level; signicant at the 0.01 level. 20 other controls: interest in politics, region: Atlantic, region: Prairies, region: British Columbia, gender, traditionally married, baby boomer, gen X, rural, high school dropout, university graduate, employed, public sector, union household, catholic, no religion, north European, non European, low income, high income.

Table 7 Impact of debates and media coverage on vote intentions by time of decision (stacked logistic regression) All respondents b (standard error) Liberal debate Liberal debate2 Conservative debate Conservative debate2 NDP debate NDP debate2 Reform debate Reform debate2 Tone of media coverage Party identication Party dummies Conserv. NDP Reform 20 other controls Constant Number of cases Percent correctly predicted 0.08 (0.04) 0.004 (0.002) 0.09 (0.04) 0.004 (0.003) 0.02 (0.05) 0.003 (0.003) 0.01 (0.04) 0.002 (0.002) 0.22 (0.22) 3.52 (0.11) 0.87 (0.49) 0.94 (0.52) 0.29 (0.46) 1.40 (0.31) 6420 86.5

Decided before campaign b (standard error) 0.10 (0.08) 0.005 (0.005) 0.04 (0.11) 0.002 (0.006) 0.16 (0.12) 0.007 (0.007) 0.06 (0.08) 0.004 (0.004) 0.00 (0.48) 5.45 (0.28) 0.83 (1.10) 1.60 (1.20) 0.54 (0.96) 2.15 (0.71) 2192 91.5

Decided during campaign b (standard error) 0.13 (0.07) 0.005 (0.004) 0.15 (0.07) 0.009 (0.004) 0.11 (0.10) 0.008 (0.006) 0.01 (0.08) 0.003 (0.004) 0.76 (0.42) 2.81 (0.19) 0.45 (0.85) 1.41 (1.03) 1.14 (0.90) 1.40 (0.58) 1760 83.9

Signicant at the 0.10 level; signicant at the 0.05 level; signicant at the 0.01 level. 20 other controls: interest in politics, region: Atlantic, region: Prairies, region: British Columbia, gender, traditionally married, baby boomer, gen X, rural, high school dropout, university graduate, employed, public sector, union household, catholic, no religion, north European, non European, low income, high income.

ARTICLE IN PRESS
P. Fournier et al. / Electoral Studies XX (2003) XXXXXX 15

The relationship between tone of media coverage and vote intentions among campaign deciders, though weaker, remains positive and statistically signicant when calculated across all parties. It should be noted, however, that the relationship is signicant only at the 0.10 level, and that more detailed analyses (not reported here) indicate a positive signicant coecient for only two of the parties (NDP and Reform). Tone of coverage is never signicantly related to the vote intentions of those classied as pre-campaign deciders. In summary, not only are the vote intentions of campaign deciders volatile, but they are volatile because they respond to actual campaign events and coverage. They do not uctuate haphazardly. People who say they decided before the campaign, as they should be if the measure is valid, were not induced by campaigns to change their vote choice.

4. Conclusions People are not equally aected by election campaigns. Some individuals are not suciently interested in or attentive to political matters to be reached by campaign coverage and rhetoric. Others are aware of campaign messages, but they are already strongly committed to a particular party or candidate, and so they are not easily persuaded to change their mind. Such are the obstacles for all forms of mass communication (Zaller, 1992). This study shows, however, that election campaigns may contain a large contingent of individuals who are susceptible to persuasion. In the Canadian case, over half of citizens say they made their voting decision during the course of the campaign, and they form a relatively interested, attentive, informed, and less committed group which is more likely to be reached by, to be receptive to, and to be responsive to campaign stimuli. The evidence shows that campaign deciders vote intentions are more strongly aected by campaign events such as leaders debates and media coverage. We thus grossly underestimate the strength of campaign eects by estimating them across the entire electorate, since they actually only concern a specic portion of the public. The existence of a link between time of decision and susceptibility to campaign persuasion has important implications. Campaign deciders represent roughly half of the Canadian electorate. This proportion surpasses comparable statistics from Great Britain and the United States (Flanigan and Zingale, 1994; Norris et al., 1999). This fact may explain the apparent contrast in the size and the quantity of campaign eects found among these countries, particularly comparing national elections in Canada and the United States. Additionally, there is evidence that the proportion of campaign deciders has been increasing over the last few decades in several countries (see Tables 73 in Flanigan & Zingale, 1994; Archer & Kanji, 1998; Norris et al., 1999; Dalton et al., 2000). Our results oer an explanation for this pattern. We nd that the only characteristic distinguishing pre-campaign from campaign deciders is the level and strength of partisanship. Individuals who decide during and are aected by campaigns are less attached to political parties. One can thus speculate that the

ARTICLE IN PRESS
16 P. Fournier et al. / Electoral Studies XX (2003) XXXXXX

observed decline in partisanship (Dalton, 2000) is responsible for the swelling of the ranks of campaign deciders and the increasing number of studies that nd campaign eects. One can further speculate that if the decline in partisanship persists in the future, then the proportion of campaign deciders and the frequency of campaign eects could also continue to increase. Acknowledgements The authors gratefully acknowledge nancial assistance from the Social Sciences be cois and Humanities Research Council of Canada (SSHRC) and the Fonds que te et la culture (FQRSC). de la recherche sur la socie Appendix. Description of variables and 1997 CES question numbers (listed in order of appearance) Note: questions from the campaign wave start with a c, while those from the post-election wave start with a p. Stability of vote choice (0/1) [cpsa4, pesa4]: 0 respondent moved from indecision to a choice or from one party to another between the two waves. 1 respondent specied same party choice in pre- and post-election interviews. (Abstentions, refusals and non-voters were excluded.) Reported time of decision, validation [pesa4c, pesa4d, cpsdate]: After campaign interview (0/1), 1 decided after time of pre-election wave, 0 decided at other time. Before campaign interview (0/1), 1 decided before time of pre-election wave, 0 decided at other time. (Missing dummy: During campaign interview (0/1), 1 decided during time of pre-election wave, 0 decided at other time.) Day of interview (136) [cpsdate]: 1 first day of campaign eldwork, (...) 36 last day of campaign eldwork (day before election). Strength of party identication (01) [cpsk1, cpsk2]: 0 no identication, .33 not very strong, .67 fairly strong, 1 very strong identication. Campaign political information (01) [cpsf13, cpsf14, cpsf15]: Knowledge of the source (political parties) of three main campaign promises (lower income taxes by 10%, cut unemployment in half by 2001, against distinct society status for Quebec), 0 three incorrect or Dont Know responses, (...) 1 three accurate answers.

ARTICLE IN PRESS
P. Fournier et al. / Electoral Studies XX (2003) XXXXXX 17

General political information (01) [cpsl6, cpsl11, cpsl12, cpsl13]: Knowledge of four political actors (current American President, current Federal Minister of Finance, current Provincial Premier, and rst woman Canadian Prime Minister), 0 four incorrect or Dont Know responses, (...) 1 four accurate answers. Interest in election (01) [cpsb4]: 0 no interest at all in this election, (...) 1 a great deal of interest in this election. Interest in politics (01) [cpsb5]: 0 no interest at all in politics generally, (...) 1 a great deal of interest in politics generally. Media attention (01) [cpsb1, cpsb2, cpsb3]: 0 no attention at all paid to election news on television, in newspapers, and on the radio, (...) 1 a great deal of attention paid to election news on television, in newspapers, and on the radio. Talked about election with relatives, friends and other people (01) [cpsb8, cpsb8a]: 0 did not talk at all about the election with relatives, friends, or other people, (...) 1 often talked about the election with relatives, friends, and other people. Most important source of election information, three dichotomies [cpsl4]: Family and friends (0/1), 1 family and friends were most important source of election information, 0 other. Radio (0/1), 1 radio was most important source of election information, 0 other. Newspapers (0/1), 1 newspapers were most important source of election information, 0 other. (Missing dummy: Television (0/1), 1 television was most important source of election information, 0 other.) Age in years (18101) [cpsage]. Gender, woman (0/1) [cpsrgen]: 0 male, 1 female Education, highest level completed (01) [cpsm3]: 0 no schooling, 0.1 some elementary, 0:2 complete elementary, 0:3 some secondary/high school, 0:4 complete secondary/high school, 0:5 some

ARTICLE IN PRESS
18 P. Fournier et al. / Electoral Studies XX (2003) XXXXXX

technical college, 0:6 complete technical college, 0:7 some university, 0:8 B:A:; 0:9 M:A:; 1 Professional=Ph:D: Income, total household (01) [cpsm16, cpsm16]: 0 < $20; 000, 0:11 $20; 000 $29; 999, 0:33 $40; 000 $49; 999, 0:44 $50; 000 $59; 999, 0:67 $70; 000 $79; 999, 0:78 $80:000 $89:999, 1 > $100; 000.

0:22 $30; 000 $39; 999, 0:55 $60; 000 $69; 999, 0:89 $90; 000 $99; 999,

Region, four dichotomies [province]: Atlantic (0/1), 1 resident of Newfoundland, Nova Scotia, Prince Edward Island, or New Brunswick, 0 resident of other province. British Columbia (0/1), 1 resident of British Columbia, 0 resident of other province. Prairies (0/1), 1 resident of Alberta, Saskatchewan, or Manitoba, 0=resident of other province. Quebec (0/1), 1 resident of Quebec, 0 resident of other province. (Missing dummy: Ontario (0/1), 1 resident of Ontario, 0 resident of other province.) Reported time of decision, campaign eects [pesa4c, pesa4d]: Before campaign (0/1), 0 decided during campaign, 1 decided before start of campaign. During campaign (0/1), 0 decided before start of campaign, 1 decided during campaign. Vote intentions, various party dichotomies (0/1) [cpsa4]: 0 think will vote for other party than X, 1 think will vote for party X. (Leaners included.) Debate (020) [cpsdate]: Number of days since the debates occurred. Debate2 (0400) [cpsdate]: Squared number of days since the debates occurred. Party identication, various party dichotomies (0/1) [cpsk1, cpsk2]: 0 does not identify very or fairly strongly with party X, 1 identifies very or fairly strongly with party X. Tone of coverage (11): Average daily tone of coverage, televised nightly news content analysis, student coding, 1 negative, 0 neutral, 1 positive.

ARTICLE IN PRESS
P. Fournier et al. / Electoral Studies XX (2003) XXXXXX 19

Traditionally married (0/1) [cpsm2]: 0 not married, 1 married. Baby boomer (0/1) [cpsage]: 0 born during other period, 1 born between 1946 and 1962. Gen X (0/1) [cpsage]: 0 born before 1963, 1 born in 1963 or later. Rural (0/1) [postcode]: 0 urban place of residence, 1 rural place of residence. High school dropout (0/1) [cpsm3]: 0 completed high school, 1 did not complete high school. University graduate (0/1) [cpsm3]: 0 did not complete a university degree, 1 completed a university degree. Employed (0/1) [cpsm4]: 0 not presently working, 1 presently working. Public sector (0/1) [cpsm7]: 0 working for private company or not presently working, 1 working for government, public sector, or government owned company. Union household (0/1) [cpsm9]: 0 no household member belongs to a union, 1 household member belongs to a union. Catholic (0/1) [cpsm10]: 0 other or no religious aliation, 1 Catholic religious aliation. No religion (0/1) [cpsm10]: 0 possesses a religious aliation, 1 no religious aliation. North European (0/1) [cpsm13]: 0 other ethnic or cultural aliation, 1 Scandinavian, Baltic, German, Dutch, or Belgian ethnic or cultural aliation. Non European (0/1) [cpsm13]: 0 European ethnic or cultural aliation, 1 non European ethnic or cultural aliation.

ARTICLE IN PRESS
20 P. Fournier et al. / Electoral Studies XX (2003) XXXXXX

Low income (0/1) [cpsm16, cpsm16a]: 0 not lowest quartile, 1 lowest quartile of total household income. High income (0/1) [cpsm16, cpsm16a]: 0 not highest quartile, 1 highest quartile of total household income.

References
Abramson, P., Aldrich, J., Rohde, D., 1998. Continuity and Change in the 1996 Election. CQ Press, Washington. Archer, K., Kanji, M., 1998. When and how Canadian voters decide: searching for systematic trends in canadian election campaigns. Paper presented at the Annual Meeting of the Canadian Political Science Association, Ottawa. Bartels, L., 1988. Presidential Primaries and the Dynamics of Vote Choice. Princeton University Press, Princeton, NJ, pp. 244277. Bartels, L., 1992. The impact of electioneering in the United States. In: Butler, D., Ranney, A. (Eds.), Electioneering: A Comparative Study of Continuity and Change. Clarendon Press, Oxford. Berelson, B., Lazarsfeld, P., McPhee, W., 1954. Voting. University of Chicago Press, Chicago, IL. Blais, A., Boyer, M., 1996. Assessing the impact of televised debates. British Journal of Political Science 26, 143164. Blais, A., Nadeau, R., Gidengil, E., Nevitte, N., 1999. Campaign dynamics in the 1997 Canadian election. Canadian Public Policy 25, 197205. Box-Steensmeier, J., Kimball, D., 1999. The timing of voting decisions in presidential campaigns. Paper presented at the Annual Meeting of the Midwest Political Science Association, Chicago. Bowen, L., 1994. Time of voting decision and use of political advertising: The Slade Gorton-Brock Adams senatorial campaign. Journalism Quarterly 71, 665675. Chaee, S., Choe, S.Y., 1980. Time of decision and media use in the FordCarter campaign. Public Opinion Quarterly 44, 5369. Chaee, S., Rimal, R.N., 1996. Time of vote decision and openness to persuasion. In: Mutz, D., Sniderman, P., Brody, R. (Eds.), Political Persuasion and Attitude Change. University of Michigan Press, Ann Arbor, MI, pp. 267291. Dalton, R.J., 2000. The decline of party identications. In: Dalton, R.J., Wattenberg, M.P. (Eds.), Parties without Partisans: Political Change in Advanced Democracies. Oxford University Press, Oxford, pp. 1936. Dalton, R.J., McAllister, I., Wattenberg, M.P., 2000. The consequences of partisan dealignment. In: Dalton, R.J., Wattenberg, M.P. (Eds.), Parties without Partisans: Political Change in Advanced Democracies. Oxford University Press, Oxford, pp. 3763. Flanigan, W., Zingale, N., 1994. Political Behavior of the American Electorate. Congressional Quarterly Press, Washington. Fournier, P., 2000. Heterogeneity in political decision-making: the nature, extent, sources, dynamics, and consequences of interpersonal dierences in coecient strength. Unpublished doctoral dissertation, University of British Columbia. Fournier, P., Nadeau, R., Blais, A., Gidengil, E., Nevitte, N., 2001. Validation of time of voting decision recall. Public Opinion Quarterly, 65, 95107. Gopoian, D., Hadjiharalambous, S., 1994. Late deciding voters in presidential elections. Political Behavior 16, 5578. Holbrook, T., 1996. Do Campaigns Matter?. Sage, Thousand Oaks, CA. Iyengar, S., Kinder, D.R., 1987. News that Matters. The University of Chicago Press, Chicago, IL. Iyengar, S., Kinder, D., Peters, M.D., Krosnick, J.A., 1984. The evening news and presidential evaluations. Journal of Personality and Social Psychology 46, 778787.

ARTICLE IN PRESS
P. Fournier et al. / Electoral Studies XX (2003) XXXXXX 21 te, J., 1992. Letting the People Decide: Dynamics of a Canadian Johnston, R., Blais, A., Brady, H., Cre Election. McGill-Queens University Press, Kingston. Johnston, R., Blais, A., Gidengil, E., Nevitte, N., 1996. The Challenge of Direct Democracy: The 1992 Canadian Referendum. McGill-Queens University Press, Montreal. Just, M.R., Crigler, A.N., Alger, D.E., Cook, T.E., Kern, M., West, D.M., 1996. Crosstalk: Citizens, Candidates, and the Media in a Presidential Campaign. The University of Chicago Press, Chicago, IL. Klapper, J., 1960. The Eects of Mass Communication. Free Press, New York. Krosnick, J.A., Brannon, L.A., 1993. The impact of the Gulf War on the ingredients of presidential evaluations: multidimensional eects of political involvement. American Political Science Review 87, 963975. Krosnick, J.A., Kinder, D.R., 1990. Altering the foundations of popular support for the president through priming. American Political Science Review 84, 497512. Lazarsfeld, P., Berelson, B., Gaudet, H., 1944. The Peoples Choice. Columbia University Press New York. McCombs, M., Shaw, D., 1972. The agenda-setting function of the mass media. Public Opinion Quarterly 36, 176187. McGuire, W., 1969. The nature of attitudes and attitude change. In: Lindzey, G., Aronson, E. (Eds.), Handbook of Social Psychology, vol. 3. Addison-Wesley, Reading, MA, pp. 136314. Mendelsohn, M., 1994. The medias persuasive eect: the priming of leadership in the 1988 Canadian election. Canadian Journal of Political Science 27, 8197. Mendelsohn, M., 1996. The media and interpersonal communications: the priming of leaders, issues, and party identication. Journal of Politics 58, 112125. Miller, J.M., Krosnick, J.A., 1996. News media impact on the ingredients of presidential evaluations: a program of research on the priming hypothesis. In: Mutz, D.C., Sniderman, P.M., Brody, R.A. (Eds.), Political Persuasion and Attitude Change. The University of Michigan Press, Ann Arbor, MI, pp. 7999. Nadeau, R., Gidengil, E., Nevitte, N., Blais, A., 1998. Do trained and untrained coders perceive electoral coverage dierently. Paper presented at the Annual Meeting of the American Political Science Association, Boston. Nevitte, N., Blais, A., Gidengil, E., Nadeau, R., 2000. Unsteady State: The 1997 Canadian Federal Election. Oxford University Press, Toronto. Norris, P., Curtice, J., Sanders, D., Scammel, M., Semetko, H., 1999. On Message: Communicating the Campaign. Sage, London. Plumb, E., 1986. Validation of voter recall: time of electoral decision making. Political Behavior 8, 301312. Rivers, D., 1988. Heterogeneity in models of electoral choice. American Journal of Political Science 32, 737757. Sniderman, P.M., Brody, R.A., Tetlock, P.E., 1991. Reasoning and Choice: Explorations in Political Psychology. Cambridge University Press, Cambridge. Whitney, D.C., Goldman, S., 1985. Media use and time of vote decision: a study of the 1980 presidential election. Communication Research 12, 511529. Zaller, J., 1992. The Nature and Origins of Mass Opinion. Cambridge University Press, New York. Zaller, J., 1996. The myth of mass media impact revived: new support for a discredited idea. In: Mutz, D.C., Sniderman, P.M., Brody, R.A. (Eds.), Political Persuasion and Attitude Change. The University of Michigan Press, Ann Arbor, MI, pp. 1778.

You might also like