At the time this article was written Barry Kay was Professor of Political Science at Wilfrid Laurier University in Waterloo.

In its final report the Royal Commission on Electoral Reform and Party Financing proposed to restrict the publication of polls during the final 72 hours of election campaigns. This is one of the few recommendations to be acted upon yet it is supported by little documented evidence about the "contaminating" effect of polls. This article questions the assumption that the publication of public opinion polls have a "bandwagon" effect on election results.

American research experience on the bandwagon effect has been mixed. As has been observed by others "some researchers detect a bandwagon and others do not."¹ In the cases where evidence supporting a bandwagon effect is found, typically the data are uncontrolled for campaign effects, or else a controlled hypothetical experiment is created where individuals are asked to react to fictitious information.² The limitations of such research are an indication of the enormous obstacles that face any inquiry on this subject.

In order to be able to draw conclusions with confidence, ideally the research situation should allow for controlled samples during an electoral contest where one group has access to polling information and the other does not. Such a situation is not easily achieved with limited resources, and as a result adaptations frequently become necessary.

Apart from those works, there are other researchers who have been unable to corroborate a bandwagon effect in their inquiry.³ Others still have found evidence for an opposite underdog effect suggesting that unfavourable polling results for a party will disproportionately attract voters to them to mitigate against an apparent loss.⁴ There is much additional literature on related themes that refer to the conceptual implications of predictions affecting outcomes in the electoral arena.⁵ Indeed beyond the bandwagon effect there are a variety of other works concerning with other related hypotheses such as strategic voting. References to such research are not to attempt an exhaustive literature review, but rather to acknowledge that its cumulative impact has been anything but definitive, whether we are considering the less studied case of Canada or other comparable electoral systems.

The bandwagon effect is not the only illustration of public opinion polls influencing the electoral process, but it is a common enough example that it seems appropriate to explore the relationship of Canadian federal election results with a portrait of the public mood at the beginning of each of those campaigns. Table 1 reviews the last pre-election Gallup polls over the period since 1945 when public access to such polls has been available in Canada. The right-hand columns reveal the fortunes of the party that was ahead at the outset of the campaign. It shows that of sixteen elections during this time frame, on only five occasions did the early leader increase its support.

Indeed, if one excludes the earliest period when public access to polling figures was less available, there has been only one occasion (1974) over the past eleven elections when the party with the greatest support at the time an election was called, actually improved its standing by the end of the campaign. Mention might also be made of the 1988 contest, the only other instance when there was no net loss in support for the early front-runner. Polling data during that campaign shows a precipitous decline for the Conservative following the televised debate resulting in a slight Liberal lead, but a subsequent Liberal drop saw the Conservatives finish at the same figure they began. Caution is certainly advisable in drawing any conclusions from such a table, however if there was consistent evidence of a net bandwagon effect at the national level, one would expect to see a different pattern to the results. Chance alone would suggest that the pre-campaign leading party would not decline so regularly.

 

	Table 1 Pre-election Polls vs Election Results 1945 à 1993
	Last Pre-Campaign Poll			Election Result
	L	PC	CCF NDP	L	PC	CCF-NDP	% Change in Pre-election Leader
1945	36%	29%	20%	41%	28%	15%	+5%
1949	42	32	17	50	30	13	+8
1953	46	31	13		32	10	+4
1957	47	32	11	41	39	11	-6
1958	35	50	9	33	54	10	+4
1962	45	38	9	37	37	14	-8
1963	47	32	10	42	33	12	-5
1965	45	29	15	41	33	18	-4
1968	50	29	16	45	31	17	-5
1972	42	32	15	39	35	18	-3
1974	40	33	21	43	35	16	+3
1979	41	41	15	40	36	18	- 1
1980	47	27	23	44	32	20	-3
1984	48	39	11	28	50	19	-20
1988	33	43	22	32	43	20	0
1993	33	36	8 +11 +10	41	16	7 +19+14	-20

 

*The last two figures for 1993 refer to the Reform Party and the Bloc Québécois. Most of this information is drawn from the Toronto Star, September 1, 1984. It has been augmented with the comparable Gallup poll published on October 3, 1988. Since no pre-election Gallup poll was a available in 1993, the analogous figure for that year is drawn from the ComQuest poll in the Globe and Mail, September 16, 1993.

Alternate explanations are available including the underdog effect which may be more compatible with the data presented in Table 1, but there is little in the way of cogent argumentation that such a theory has widespread applicability in Canada. Even if there is little indication of nation-wide bandwagon voting over time, is it not possible that such a phenomenon could be occurring at sub-national or regional levels? This was the suggestion made by Johnston et al., pertaining to Quebec and it is quite plausible.⁶

 

	Table 2 Percentage of Defectors during 1988 and 1993 Campaign By Expectation of Party Performancei
	Lib		PC		NDP		Reform	B.Q.
Chance of Winning	National 1988	1993	National 1988	1993	National 1988	1993	National 1993	National 1993
0-49	30,5	4.2	49,0	37,8	46.4	18,8	12.1	0
50	13,1	5.1	35,0	33,3	47,6	23,8	18,9	0
51-74	19,8	8.4	26,1	36,6	45,5	56,3	15,9	3,9
75-100	15,3	8.1	18,1	42,9	32,5	33,3	20,0	5,1
C TAU	0,10*	-.02	O,22*	-0,05	0,10*	-0,18*	-0,05	-0,03
	* Indicates that the risk of results occuring by chance is less than 1/20 (This is known as the .0 level of significance test)

A related question arises however as to whether increased support for a party in a region, subsequent to its establishing a strong national popularity is necessarily an indication of that region’s voters jumping on a bandwagon. There can be a number of other reasons that account for a party gaining support in a region late in a campaign. If this pattern of late movement to a winning party recurred systematically, then the bandwagon explanation seems more valid. Some would question whether the 1988 example, even if joined by the 1958 and 1984 cases really constitutes a systematic recurring bandwagon effect. As already mentioned, there is an equally valid alternate explanation that in the modern Canadian experience, Quebec makes the winner rather than moves toward it. Quebec’s habitual widespread support for the Liberals prior to the Mulroney era is hard to attribute to a bandwagon phenomenon.

In probing this matter more fully, the most recent national election studies include question items that can shed light on the relationship between individuals’ expectations of electoral outcomes and how they vote. These items do not refer to polls, but do ask respondents about their expectations of each party winning the national election, as well as winning at the individual riding level. Although there is no linkage provided indicating whether they are aware of certain polls, the logic of the bandwagon thesis suggests that if voters expected their party to perform poorly, they would be more likely to defect.

The figures presented in Table 2 take advantage of the panel aspect of the election studies. The respondents expressing a voting preference during the campaign are monitored as to whether they switched away on election day, in conjunction with their mid-campaign assessment of the expected performance of their chosen party. For example, in the left-hand column of Table 2, during the 1988 campaign, of those declared Liberal voters who thought the party had less than a 50% chance of winning the election 30.5% defected during the campaign, whereas among Liberals who thought there was a 75% or better prospect of their party winning the election, only 15.3% defected. This is consistent with the bandwagon hypothesis. By contrast in the adjacent 1993 column when few Liberals abandoned their party during the campaign the defection rate was 4,2% among those who thought the Liberals’ prospects were under 50% nationally, but 8.1% among those most confident of a national Liberal victory. These figures contradict the bandwagon hypothesis and produce a negative correlation coefficient.

The bandwagon hypothesis anticipates voters deserting a sinking ship and moving toward a vessel with better prospects. The evidence in Table 2 is somewhat mixed, but there is little consistent pattern to support the bandwagon effect. Only among Conservative supporters in 1988 is there a strong trend to switch away among those with gloomier evaluations of party fortunes. Indeed the 1993 study shows a majority of cases with negative correlations, and the only strong coefficient being negative for the NDP nationally. It might be mentioned that a series of control variables were applied to these relationships, including province and the date of interview, but none had much impact upon the overall trend.

 

	Table 3 Percentage of voters switching toward pary during 1988 and 1993 Campaign by expectationm of party performance.
	Lib		PC		NPDp		Reform	B.Q.
Chance of winning	1988	1993	1988	1993	1988	1993	1993	1993
75-100%	33,0	16,5	17,3	3,0	14,4	0	0	1,6
51-74%	22,5	15,4	10,5	2,2	13,6	13,3	13,3	1,7
50%	18,8	15,6	12,9	0,3	8,0	5,1	5,1	1,9
0-49%	7,5	15,2	8,1	0,8	4,0	6,4	6,4	0,3
TAY C	0,16*	0,01	0,06	0,01	0,05	0	00	0,01
	Inicates that the risk of results occuring by chance is less tht 1/20. (This is known as the .05 level of significance test)

 

Shortcomings can be found with this particular research procedure. Interviews were conducted at different times during the campaigns, and accordingly the respondents by definition were subject to differing election stimuli. Moreover, individual assessments of party performance could well change between the original interview and election day.⁷ Another complicating factor is that for some voters, partisan support can come to cloud one’s judgement about electoral prospects, such that the wish becomes father to the perception. By contrast with 1988, some of the parties in 1993 have relatively smaller vote bases from which these figures are derived.

Table 3 is an analogue to Table 2 in that it correlates vote switching during the 1988 and 1993 elections by voter perceptions of how they expect the parties to perform. To illustrate with the upper left-hand column of Table 3, during the 1988 campaign, 33.0% of non-Liberals who thought the Liberal party had a 75% or better chance of winning the election nationally eventually switched toward the Liberals while only 7.5% among those who thought the Liberal prospects were under 50% switched, a result supporting the bandwagon hypothesis. In the adjacent 1993 column, the table shows a much more consistent pattern of switching toward the Liberals ranging from 16.5% among those non-Liberals most confident of a national Liberal victory to 15.2% among those most doubtful of a national Liberal victory. Hence the correlation coefficient is only .01 in this case.

The difference is that Table 3 examines the propensity of respondents to move toward (rather than away from) a party based upon their sense of its electoral prospects. To complete the analogy, if the bandwagon effect pertains, voters are more likely to switch to a party they feel has a better chance of winning. This table examines the proportion of respondents not supporting a party during their mid-campaign interview that reported switching to it on election day.

Again there is little patterned evidence of a bandwagon. Only in the case of the Liberals in 1988 is there clear support for the hypothesis that voters are attracted to a party they think has a better chance of winning.⁸ In 1993, apart from the local riding perceptions of those switching to the Liberals (not shown in the table) there were no other examples that were statistically significant. As with Table 2, the 1993 data are hindered by a small vote base in some categories for a number of parties.

It is not the intent of the paper to deny the existence of evidence that is consistent with the bandwagon phenomenon. Circumstances can occur where voters move to parties that they think may have a better chance of winning, quite possibly including the 1988 Quebec example. It is also entirely possible that some voters are motivated by the underdog effect, or a variant of it, in which they may vote against a party with a big lead or to which they wish to deny a parliamentary majority. There are numerous other conceptual adaptations of strategic voting which can generate a range of alternate hypotheses, that link some expectations of electoral outcome to an individual’s voting decision. In each of this host of possible examples, access to polling information might well influence how voters ultimately behave.

Does such influence constitute interference or even contamination of the Canadian electoral process? For this to be true, a case should be made that public opinion polls present a systematic pattern of intervention in the outcome of elections. Although the bandwagon effect has been the focus of this paper, it certainly is not the only type of intervention that could interfere with the process. In addition to the bandwagon hypothesis’ assumption that parties leading in the polls, will win by even larger margins as impressionable voters move toward them, there are other possible systematic effects. The underdog thesis suggests that parties leading in the polls will tend to lose support. One form of strategic voting suggests that third place parties will decline, while second place parties improve their position to more effectively challenge the first place party. Each of these possible theories could apply nation-wide or systematically among some demographic category or sub-region, as was hypothesised about Quebec. By this criterion, it would appear that there is no compelling evidence of polls having a systematic pattern of intervention in the electoral process.

If there is such insubstantial evidence to corroborate the supposition that polls interfere with Canadian elections, why is the view so widespread among elites? It might be an extension of a more generalized concern that polls have a negative impact upon the conduct of Canadian politics. This has been expressed in the view that social critics and activists condemn polls as tending to institutionalize a conservative anti-innovative mass impulse.⁹ The disdain suggests that pollsters have "replaced elected representatives ... as the major determinants of political action."¹⁰

Jeffrey Simpson of the Globe and Mail has stated that polls are responsible for timidity in politicians and incrementalism in policy-making. Worse still is their impact upon his own breed, the fourth estate. He has written that "just as love is wasted on the young, so polls are often wasted on the media. More specifically he is critical of the media’s obsession with a horserace perspective, and the inability of journalists to report polls responsibly. Among his comments are that "once the media rush to judgement following a given poll ... the politicians feel they must shift their behaviour accordingly." It has been suggested by others that horserace stories are more easily reported than substantive issues.

When hostility to the role of polls in the political process has reached such intensity among the journalistic elite, is it such a leap of faith that unsubstantiated assertions be attributed to the potential impact upon the public? The claims made about undue influence upon the electoral process then rest in part on the fact that the media does not trust its ability to present the material appropriately.

One wonders if the publication of polls was restricted as some media commentators wish, how they would perceive similar constraints upon their own ability to publicly speculate about election outcomes, and the possibility that might alter how some citizens vote.

It is difficult to draw definitive conclusions from the research reported in this paper. While one can affirm the absence of any prevailing general bandwagon effect, there is little basis to suggest that the thesis can be categorically dismissed in all circumstances. Evidence exists, idiosyncratic as it might be, that in some situations Canadian voters may be more likely to defect from a party if they expect it to perform poorly. There is also scattered evidence, that in a few cases Canadians may be less likely to switch away from a party they feel has little chance of success. Although these data were derived independently of any information about poll awareness by respondents, nothing found has precluded the possibility that a small minority of Canadians might be influenced by polling information. However if this is so, there is also little persistent pattern to the direction of the phenomenon. Rather it is more likely that the effect, if any, is diffuse with a number of cross-cutting manifestations that result in minimal net impact upon election results.

Another perspective on this matter suggests that concern with whether a bandwagon or comparable effect exists, finesses a more important point. Such an argument suggests that whether people are affected by polls is of less concern than why they may be affected. The gaining of information that permits a voter to make an informed strategic decision is not of the same order as someone who merely wants to be on the winning side as an end in itself. Those who would restrict public access to polls should then have an additional burden of proof to establish, beyond the question of influence upon the electorate. Namely, is the influence dysfunctional for the political system or constructive?

Notes

1. Richard Henshel and William Johnston, "The Emergence of Bandwagon Effects: A Theory", The Sociological Quarterly, 28, 1987, p. 493.

2. Beniger, "Winning the Presidential Nomination", Public Opinion Quarterly, 40, 1976, pp. 22-38; P. Straffin, "The Bandwagon Curve", American Journal of Political Science 21, 1977, pp. 695-710; J. Gush, "The Impact of Candidate Expenditures, Regionality and Prior Outcomes on the 1976 Democratic Presidential Primaries", Journal of Personality and Social Psychology 38, 1980, pp. 337-347; A. Skalaban, "Do Polls Affect Elections? Some 1980 Evidence", Political Behavior 10, 1988, pp. 136-149; R. Nadeau, E. Cloutier and J.H. Guay, "New Evidence About the Existence of a Bandwagon Effect in the Opinion Formation Process", International Political Science Review 14, 1993, pp. 203-213.

3. H. Dizney and R. Roskens, "An Investigation of the Bandwagon Effect in a College Straw Election", Journal of Educational Sociology 36, 1962, pp. 108-114; C. Roll and A. Cantril, Polls: Their Use and Misuse in Politics, Basic Books, New York, 1972, p. 28, R. Navizio, "An Experimental Approach to Bandwagon Research", Public Opinion Quarterly, 41, 1977, pp 217-225; J. Tyson and S. Kaplowitz, "Attitudinal Conformity and Anonymity", Public Opinion Quarterly 41, 1977, pp. 226-234; A.Cantril, Polling on the Issues, Seven Locks Press, Cabin John, Maryland, 1980; S. Herbst, Numbered Voices, University of Chicago Press, Chicago, 1993, pp. 166.

4. J. Laponce, "An Experimental Method to Measure the Tendency to Equibalance in a Political System", American Political Science Review, 60, 1966, pp. 982-993; D. Fleitas, "Bandwagon and Underdog Effects in Minimal Information Elections", American Political Science Review 65, 1971, pp. 434-438; G. Gaskell, "Polls and the Voters", New Society r, April 28, 1974, pp. 23-24; S. Ceci and E. Kain, "Jumping on the Bandwagon with the Underdog", Public Opinion Quarterly 46, 1982, pp. 228-242.

5. H. Simon, "Bandwagon and Underdog Effects and the Possibility of Election Predictions", Public Opinion Quarterly 18 (1954), pp. 245-253; M. Gartner, "Endogenous Bandwagon and Underdog Effects in a Rational Choice Model", Public Choice 25 (1976), pp. 83-89; S. Brams and J. Gerrigo-Pico, "Bandwagons in Coalition Formation", American Behavioral Scientist 18 (1975), pp. 472-496; C. Zech, "Leibenstein’s Bandwagon Effects as Applied to Voting", Public Choice 21 (1975), pp. 117-122; Catherine Marsh, "Back on the Bandwagon: The Effect of Opinion Polls on Public Opinion", British Journal of Political Science 15 (1985), pp. 51-74.

6. See Richard Johnston, André Blais, Henry Brady and Jean Crête, Letting the People Decide: Dynamics of a Canadian Election, McGill-Queen’s University Press, Montreal, 1992, pp. 199-200. Probably the most useful attempt to provide research data on the electoral impact of polls in Canada is contained in the National Election Studies of 1988 and 1993. Its treatment of this subject focuses upon expectations of electoral outcomes during the 1988 campaign by monitoring trends within the rolling samples. After acknowledging limitation with the bandwagon hypothesis applicability nationwide in 1988, the primary example offered is the Quebec subsample whose support for the Mulroney Tories is presented as a response to strong Conservative polls, part of an extended pattern of voting for the likely election winners, a perception that usually meant Liberal support historically. There is anecdotal evidence in 1984 at least, and probably 1958 as well to show that Quebec moved to Conservative support after a Tory victory was indicated elsewhere, apart from the 1988 case. The 1993 federal election result in Quebec, on the other hand, would lead to a different interpretation. Conceptually, Johnston presents two models to account for the Québécois tendency toward block voting. In one model "Quebecers go with winners" while in the other model they "make the winner". As is stated in the book, "only the first model is really about a kind of bandwagon." During modern times there are really only three elections 1958, 1984 and 1988 where there is evidence consistent with substantial numbers of Quebecers voting because of a bandwagon effect, and even then alternate models exist to explain the phenomenon. The 1993 success of the Bloc Québécois suggests anything but bandwagon voting, at least in the traditional sense.

7. As respondents in the rolling samples were interviewed over seven week periods, it is reasonable to expect a greater tendency of vote switching among those contacted earlier in the respective campaigns. This anticipated effect was monitored by dividing the samples into four categories based upon interview date, and this control variable was found to have little impact upon the relationships presented.

8. This finding, together with the observation in Table 2 that Conservatives were less likely to defect in 1988 if they were confident of the party’s national prospects, led to the use of a control on Conservative expectations for voters switching to the Liberals, and produced a .11 Tau C coefficient. Other applications of this cross-party control produced substantially weaker or negative correlations.

9. Hugh Whalen, "The Reward and Perils of Polling", in Paul Fox, ed., Politics: Canada fifth edition, McGraw-Hill Ryerson, Toronto, 1982, p. 250.

10. Claire Hoy, Margin of Error: Pollsters and the Manipulation of Canadian Politics, Key Porter Books, Toronto, 1989, pp. 39-40.