Public Opinion and Policy Making in Canada 1994–2001
Published online by Cambridge University Press: 03 March 2005
Abstract
Abstract. This study examines the consistency between public opinion and public policy during the period 1994–2001 by matching responses to national survey questions on 230 issues with enacted policy proposals on the same issues. Policy outcomes were consistent with majority opinion 49 per cent of the time. This represents a significant drop from 69 per cent during the Mulroney years (1985–1993). Low opinion-policy consistency since 1994 is primarily attributable to divergences between public majorities that are increasingly supportive of a change toward the right and the policies of Jean Chrétien that are more leftist and status quo oriented than those of his predecessor. We argue that these divergences go largely unnoticed by the public because they tend to occur on low-profile issues. On the other hand, the evidence suggests a much tighter correlation between opinion and policy on a small number of high-profile issues of which the public is much more aware, thereby creating the appearance of attentiveness to Canadian public opinion.
Résumé. En comparant les décisions sur 230 enjeux de politiques publiques avec les résultats de sondages nationaux sur ces mêmes enjeux, cet article cherche à quantifier le degré d'adéquation entre l'opinion publique et la politique gouvernementale entre 1994 et 2001. Les calculs révèlent que seulement 49 pour cent des décisions du gouvernement de Jean Chrétien sont allées dans le même sens que l'opinion publique, en nette diminution par rapport aux 69 pour cent observés pendant la période Mulroney (1985–1993). La baisse de corrélation depuis 1994 est principalement attribuable à la divergence entre une opinion publique de plus en plus favorable au changement et idéologiquement orientée à droite et la politique du gouvernement de Jean Chrétien sensiblement plus résistante au changement et idéologiquement plus à gauche que celle de son prédécesseur. Le public a tendance à ignorer le manque de corrélation entre l'opinion et les politiques gouvernementales parce que les enjeux en question sont relativement peu importants. Par contre, il semble que la corrélation entre l'opinion et les politiques soit beaucoup plus forte dans un petit nombre d'enjeux importants que le public reconnaît, créant ainsi l'apparence d'un gouvernement attentif aux souhaits de l'opinion publique canadienne.
- Type
- Research Article
- Information
- Canadian Journal of Political Science/Revue canadienne de science politique , Volume 37 , Issue 3 , September 2004 , pp. 505 - 529
- Copyright
- © 2004 Cambridge University Press
Politicians and policy makers become tightly bound to the unreflective whims of constituents mobilized by special interests…; decisions on highly technical matters of public policy are … made by leaders glued to polling results.
Thomas Homer-Dixon, 2000
There is a firmly embedded assumption in the popular folklore of politics that politicians are highly attentive to public opinion. This folklore takes two contrary forms, one which evinces that political leaders “pander” to the public by heeding their every whim, another which contends that political leaders use polls to help craft their messages and present their programs in ways that are acceptable to the public. In both the government responsiveness and public support scenarios, the resulting relationship between public opinion and public policy should be the same: a tight correlation between the expressed wishes of the public as measured by polls and the decisions of public officials. Yet this presumed relationship stands in stark contrast to the large number of specific cases in which public opinion and public policy are inconsistent. Recent examples of opinion-policy inconsistency are available in both the United States (the impeachment of President Clinton, tobacco legislation, campaign finance reform, US involvement in peacekeeping operations) and Canada (capital punishment, the Canada-US and then the North American Free Trade Agreements, the Goods and Services Tax, compensation to hepatitis C victims, gun control, privatized health care, euthanasia and the Young Offenders Act).
This article attempts to resolve the apparent dilemma of presumed attentiveness to public opinion and the frequent lack of a correlation between opinion and policy. We do so through an exhaustive search, summary and quantitative analysis of survey results and policy outputs over the period 1994–2001. By using data on how individual survey respondents voted at the last election, we are able to show that the low correlation is in large part the result of ideological divergences between conservative public majorities supporting the right side of low-profile issues and more liberal government policies that are on the left side of the same issues. We argue that due to their low profile nature, these divergences go largely unnoticed by the public. On the other hand, the evidence suggests a much tighter correlation between opinion and policy on a small number of high-profile issues of which the public is much more aware, thereby creating the appearance of attentiveness to public opinion.
Theoretical Expectations Regarding the Correlation Between Opinion and Policy
Research on the relationship between public opinion and policy, particularly in the United States, often presumes that public opinion is an autonomous and rational force responding in sensible ways to changes in the political and economic environment, that policy makers are responsive to this autonomous force, and that it can be fairly accurately measured using polls (Geer, 1996; for a review see Shapiro and Jacobs, 1989). More critical perspectives challenge the theory of government responsiveness by focusing on the frequent instances in which opinion and policy diverge. Some scholars have even rejected the entire enterprise of searching for some kind of correlation between opinion and policy. They have suggested that mass opinion as measured by polls is so unstable, so affected by measurement error such as question wording and order effects (Achen, 1975; Converse, 1964), or so easily influenced by changing context (Kuklinski and Quirk, 2000; Zaller and Feldman, 1992) that to speak of “public opinion” at all may be a misnomer, and therefore to presume that public officials can sensibly respond to or manipulate such a slippery entity is a mistake. However, in Canada, Europe and the United States, there is ample evidence that despite these legitimate methodological and theoretical critiques, when individual opinions are aggregated and tracked over time, collective opinion looks far more stable and reasonable than individual responses to survey questions, and demonstrates consistent correlations with policy and changes in the environment (Bélanger and Petry, 2002; Feld and Groffman, 1988; Isernia et al., 2002; Johnston, 1986; Page and Shapiro, 1992).
Another reason as to why policy might diverge with opinion as measured by polls is that, although elected officials may try to respond to public opinion, they do not think that polls accurately represent the real state of public opinion. They may therefore rely on other indicators or interpretations of public opinion that they find more useful than polls to assess where majority opinion stands on an issue.1
For example, Entman and Herbst (2001) see public opinion as made up of four referents: mass opinion refers to the aggregate of individual opinions found in polls; activated opinion includes the opinion of engaged, organized and informed groups and individuals, including lobbies and experts; latent opinion is shaped by underlying beliefs behind opinions and is where the collective stance ends up after debate; and perceived majorities are the perceptions of the mass audience, journalists and political actors of where majority opinion stands on an issue.
Another argument in support of non-responsiveness emphasizes the autonomy of the state from public opinion, which allows politicians to ignore or deviate from mass preferences and get away with it. These deviations may reflect the greater influence of organized interests—party activists, interest groups, business associations—on policy makers, a process characterized by Brooks as democratic frustration (Brooks, 1985). The responsible party model is a variant of this view, in which government decision makers formulate policy on the basis of their own, often strongly held, policy preferences, and these may come into conflict with public majorities that support the policies proposed by opposition parties, thereby prompting non-responsiveness.2
As Schedler (1998) points out, the responsible party model also implies that policy makers should not shift previously announced policy positions in response to changing public opinion for fear of appearing untrustworthy and morally inconsistent, even at the cost of non-responsiveness.
The democratic frustration perspective presumes that opinion would correlate with policy if only policy makers were responsive. This notion can be challenged by arguing that correlations between opinion and policy in no way provide evidence that policy makers are responsive. Correlations could just as easily be interpreted as evidence of a counterfeit consensus, whereby policy makers are able to lead members of the public away from their true interests by endorsing elite programs. Under such a scenario, where public opinion is the dependent rather than the independent variable in the equation, any correlation between policy and mass opinion is a hollow imitation of the theory of democratic responsiveness (Margolis and Mauser, 1989). Seen in this light, the absence of correlation between opinion and policy would presumably imply that the public is capable of resisting government efforts at manipulation.
At this point of the discussion, it should be clear that both the government responsiveness and the counterfeit consensus perspectives are caricatures of how opinion and policy are actually related. A more balanced view holds that the relationship between opinion and policy is reciprocal, with neither clearly leading the other.3
For example, Jacobs (1993) argues that the growth of a public opinion apparatus within the executive branch of the American and British governments originally intended to help manipulate popular preferences through public relations campaigns has in fact backfired and increased policy makers' interest in tracking and responding to public opinion. He calls this the “recoil effect.” Another example is Wlezien's work (1995; 1996) on the ‘thermostatic’ opinion-policy dynamics, where he finds that US policy makers respond to public preferences and that the public responds to changes in policy. See Soroka and Wlezien (2003) for evidence of a thermostatic opinion-policy relationship in Canada.
The broad theoretical perspectives outlined above provide us with the beginnings of a road map, but much of the detail remains missing. Many questions emerging from this theoretical review require the use of such qualitative methods as archival research, interviewing elite informants, and participant observation. There are, however, pressing questions which can be appropriately addressed in a quantitative inquiry such as this one. Under what circumstances are opinion and policy correlated? How has the correlation evolved over time? Why has consistency declined in recent years? To answer these questions, it is useful first to remember that poll results only measure “mass opinion” and that they may not be a good indicator of other attributes of public opinion. Second, it is also useful to distinguish between various aspects of “mass opinion”: its existence, its direction and its magnitude or intensity (Schuman and Presser, 1981). Previous work is often insufficiently subtle to recognize that the impact of mass opinion on policy is likely to depend on the size of public majorities, and on the conviction that it really exists and does not simply reflect “non-attitudes.” Once we accept such a conceptualization, it is not surprising that different kinds of “public opinion” will impact differently on policy, depending on the circumstances. Geer (1996) parsimoniously suggests that political leaders will respond to mass opinion as measured by polls on high profile issues, but will be free to respond to other definitions of public opinion on lower profile issues.
Third, mass opinion is more likely to influence policy if public majorities and government policies follow the same ideological direction on issues than if one takes a left and the other a right direction. A government of the left is expected to follow public opinion more closely if public majorities support left policies than if they support right policies. Ideological considerations are relevant here because the change of government from Brian Mulroney to Jean Chrétien coincided with a clear ideological shift to the left as far as federal policies were concerned. There is no evidence, however, that a similar shift to the left has occurred in Canadian public opinion since 1994. It is therefore important to examine whether consistency has depended on the left-right ideologies of successive governments.
Reviewing the Comparative Evidence
A number of different methodological choices are possible in order to determine quantitatively the conditions which facilitate an opinion-policy nexus. In the US, dyadic studies have examined the state of constituency opinion and the voting behaviour of individual legislators in an attempt to quantify the impact of mass opinion and other factors on legislators' decisions (see for example Bartels, 1991; Wetstein, 1996; Burstein, 1998). This type of study is well suited to an American congressional system, but has little to offer in Westminster-style systems.
Another approach that is better suited to the Canadian case consists of comparing actual government decisions (rather than individual legislators' voting) with measures of public opinion on a large number of issues. One group of studies within this approach looks for opinion-policy congruence by tracking, over time, summary measures of mass opinion and policy outputs. The seminal study within the congruence tradition is that of Page and Shapiro (1983), who tracked policy preferences on 231 separate issues between 1935 and 1979 and compared them to trends in policy decisions over the same period, and concluded that there was congruence 66 per cent of the time, with opinion usually leading policy decisions. Studies of congruence need not look at questions on an issue-by-issue basis. For example, Stimson, MacKuen and Erickson (1995) constructed highly aggregated liberal-conservative trends in the public mood and law making by Congress and the Presidency since the 1950s, and found that as the public mood shifts to a more liberal position, more liberal legislation is passed into law.
A second group of studies looks for consistency rather than congruence. These studies do not track opinion over time but instead examine issues at a single point in time, dichotomizing public preferences as supporting either the status quo or change, and examining subsequent policy outputs to see if government decisions are consistent with mass opinion. Evidence of consistency is established when the public supports change and the government follows, or when the public supports the status quo on a given issue and the government takes no action on that issue. Monroe (1998) found a consistency rate of 55 per cent in the US in the period 1980–1993, a drop of 8 points relative to his earlier study over the 1960–1975 period (Monroe, 1979). Comparable rates have been found in other countries. In a series of comparative studies on the US, Canada, Great Britain, France and West Germany, however, Brooks (1985; 1987; 1990) finds consistency rates that are approximately 20 percentage points lower than other authors.4
Petry (1999) reports a consistency rate of 60 per cent in Canada over the period 1968–1993. Brettschneider (1996) finds that government actions in postwar Germany are consistent with majority opinion 71 per cent of the time. He also finds that changes in public opinion and government decisions are congruent 65 per cent of the time. The lower consistency rates found by Brooks are explained, in large part, by differences in methodology and time periods.
One alleged advantage the congruence approach has over the consistency approach is that it can tell us whether shifts in mass opinion occur prior to changes in government policy or not. Examining the temporal order of change in opinion and government decisions may, therefore, offer some clue about the direction of causality in the opinion-policy relationship (something the consistency approach alone cannot do). However, the congruence approach has several drawbacks that lead us to adopt instead the consistency approach in this research. Since it must rely on repeated identical survey questions only, the congruence approach automatically excludes issues for which repeated polls are not available. This is a problem because repeated polls normally represent only a small subset of all available polls. Another problem with the congruence approach is that, if there is no change in the distribution of opinion over time, it is not possible to determine congruence. When we take sampling error into account, many instances of opinion changes are too small to be statistically significant. These instances must be eliminated from the analysis. Consequently, the sample of usable repeated surveys is likely to be too small to permit us to draw meaningful statistical inference from change in public opinion. To illustrate the difficulties and limits of the congruence approach, consider that, out of a total of more than 800 surveys, there were only 55 usable repeated surveys in this study. And out of those we found only nine unambiguous cases of opinion leadership (either by the masses or by government).5
There were five cases of mass opinion leadership involving the following: deficit reduction; lower government spending on foreign aid; forbidding doctors from charging user fees; banning adds for cigarettes; and legislation to protect gays and lesbians. Four cases of leadership by the government involved regulating the sale of firearms, TV program censorship, healthcare spending and cutting the number of public employees. Another illustration of the limits of the congruence approach is Brettschneider's study of the opinion-policy relationship in Germany. He starts with a sample of 331 usable polls (from a population of more than 1000 public opinion surveys). He finds 94 instances of measured opinion change in the sample (remember this is over the period 1949–1990—forty years). Out of these 94 instances of opinion change, he finds 14 cases of “mass opinion leadership” when opinion changed first and congruent government activities followed. In most of the remaining 77 instances, he finds that it is impossible to determine the direction of the relationship. So, in the end, out of a population of more than 1000 polls, he is able to infer leadership of mass opinion over policy in only 14 cases.
Whether opinion influences policy or policy influences opinion is doubtless an important issue. However, we do not think that the quantitative method we use in this essay is appropriate to the task. Page and Shapiro (1992, 26–27) themselves have acknowledged the futility of trying to measure the temporal congruence between changes in opinion and policy even in the ideal situation where there is a large number of usable cases. In this article, we assume that opinion precedes policy without attempting to establish a causal direction by way of the congruence approach or any other method.6
As a consequence, consistency is interpreted as government responding to public opinion rather than public opinion supporting government. Whether consistency is interpreted one way or the other has some minor influence on methodological choices (the selection of cases and independent variables, for example). However, formulating consistency in terms of opinion support rather than government responsiveness does not fundamentally alter our overall conclusions.
The Study
Cases were selected for analysis as follows. Available published national survey data from November 1993 through 2000 were inspected for items matching federal policy actions. The data were found in the online survey archives of the Centre for the Study of Democracy at Queen's University, the social science data collection at Carleton University and the University of British Columbia numeric data services. The survey items came from five reputable polling institutions: Gallup, Angus Reid, Environics/CROP, Decima and Pollara. Only those polls dealing with public preference on an identifiable question of national policy were used in the analysis. This raises the question of what exactly constitutes national policy questions as opposed to provincial or local issues. A number of survey items deal with functions carried out primarily at the provincial level, including education, health and social welfare. But the reality of Canadian federalism is that it is very often the case that federal legislation, regulations or court decisions can make policy changes implied in the questions about these provincial functions. The reason why national survey questions about provincial functions are asked is precisely because the issue has policy ramifications at the federal level. Consequently, provincial issues were included in the analysis to reflect the realities of Canadian federalism, but only federal policy decisions were considered in the outcome.
Well over 800 polls were found, most of which did not meet the pre-established criterion. A small number of polls asking respondents to explicitly approve or disapprove of recently adopted governmental decisions were excluded because they contradict the posited temporal sequence between opinion and policy. Polls appearing repeatedly in identical form were considered as separate cases unless they appeared within the same calendar year, in which case they were averaged and treated as one poll. This was to avoid the possibility of coding public opinion as both consistent and inconsistent for the same issue. Sampling error was also taken into account. Since the samples used in the surveys have a three per cent margin of error, all polls indicating a difference between majority and minority opinion falling within the six per cent range were excluded from the analysis. Only those polls that gave respondents a dichotomous choice of agreeing or disagreeing with a specific policy statement were kept in the analysis. Polls that gave respondents more than two preference alternatives (e.g., increase, maintain or decrease the level of program spending) were therefore eliminated because they cannot be operationalized within this dichotomous framework. After elimination, 230 cases remained, substantially more than those reported during the governments of Brian Mulroney (162) and Pierre E. Trudeau (186) in Petry (1999).
The next step was to record government policies in a similar dichotomous fashion on the same issues. The record of federal actions on each case was inspected from November 1993 through November 2001 using the following archival sources: Canadian News Facts; The Canadian Annual Review of Politics and Public Affairs; Facts on File; The Official Report of the House of Commons Debates; the annual reports of federal ministries and agencies; and various volumes of How Ottawa Spends?. The objective was to classify policy outcomes as having resulted in the policy change implied by the survey or as maintaining the status quo within a twelve-month period following the date of the survey. This was not as problematic as one might suppose because survey items typically deal with specific policy proposals that are under consideration at the time. For government spending cases, the amount of money allocated had to shift by at least five per cent from the previous year (after adjusting for government inflation) in the direction implied by the survey question to qualify as a policy change. Otherwise, it was reported as maintaining the status quo. Issues that did not permit a clear-cut classification were submitted to the judgement of policy experts. A few instances where it was not possible to satisfactorily determine the policy outcome were excluded from the analysis.
Findings
Table 1 presents a first look at the findings by comparing the state of mass opinion, dichotomized into majorities for change and majorities for the status quo, and actual policy outcomes in the subsequent twelve-month period. We report results for the Chrétien, Mulroney and Trudeau eras separately.7
The data analyzed here are accessible on the Web site of the Centre for the Analysis of Public Policy at Université Laval, available at www.capp.ulaval.ca.
Table 1: Consistency and Majority Preference for Policy Change (1968–2001)

Several empirical points should be underlined. First, consistency under Chrétien looks unusually low only when compared with consistency during the Mulroney era, but is in line with consistency during the Trudeau era.8
Variation in consistency across governments is not due to methodological differences. We used identical methods for data collection and analysis during all three periods. Cases (surveys and decisions) associated with the Clark (1979), Turner (1984) and Campbell (1993) interludes were deleted from the analysis.
It is possible that the change in the level of consistency between the Mulroney and the Chrétien periods is simply an artifact of change over time in the distribution of issues. There may have been, for instance, a larger proportion of low consistency issues raised by surveys during the government of Jean Chrétien as compared to the government of Brian Mulroney. The relative number of surveys on high consistency issues may also have dropped since 1993. To examine this possibility, Table 2 presents the level of opinion-policy consistency under Chrétien and Mulroney, broken down by issues. From the table, we see two low consistency issues that are new to the Chrétien government (privatized health care; young offenders). Other low consistency issues, while not entirely new, have taken on more importance in recent opinion polls (mercy killing; tough on crime; workfare). Furthermore, some high consistency issues (economic regulation; Quebec/Canadian unity) have received less attention in recent polls. On the other hand, we also see high consistency issues that were raised more often in recent surveys than during the previous government (government cuts; deficit reduction; sexual orientation issues). Consistency during the Chrétien years is lower in roughly similar proportions across pretty much the whole issue spectrum. The only exception is the crime issues complex, where the level of consistency under Chrétien (17%) is half the level of consistency under Mulroney (33%). On balance, pending a more accurate assessment by way of multivariate regression analyses, there is no clear evidence from Table 2 that the distribution of issues under Chrétien is skewed toward low opinion-policy consistency when compared with the distribution of issues under Mulroney.
Table 2: Consistency Broken Down by Issues 1984–1993 and 1994–2001

Aside from methodological considerations, there are substantive reasons as to why consistency may have declined after 1993. One explanation of lower consistency under Chrétien is that there was an increase in government bias against policy change in recent years. Previous research has uncovered what appears to be a general government bias against change and in favour of the status quo in policy making (Brettschneider, 1996: 301; Monroe, 1998: 18). Given this general bias, government will appear more responsive—and consistency will be more likely—when public opinion favours the status quo as compared to when public opinion supports a policy change. Based on this, we hypothesize that inconsistency will occur more frequently when the public supports change and the government opts for the status quo than when the public prefers the status quo and the government favours change. Table 1 indeed shows that inconsistency is far more frequent overall when the public supports change but none is forthcoming from the government (95 + 32 + 80 = 207 cases overall), than when the public supports the status quo and government initiates change (22 + 18 + 10 = 50 cases overall), suggesting that there are institutional forces which lead to government inertia and resistance to change, even when the public is supportive. There is, in other words, a bias against change and in favour of the status quo in policy making, which is consistent with German and American findings.9
The number of cases in which opinion supports change while government supports the status quo is typically higher in Canada than in the US. There is a methodological reason, involving repeated polls, for the discrepancy. We consider repeated polls as separate cases and compare policy outcomes with majority opinion on each case, while Monroe treats repeated polls as one poll and reports only the earliest survey in the series. Our method reduces consistency relative to Monroe's method because repeated polls have a tendency of targeting issues in which the public supports change and government favours the status quo (this is presumably why pollsters keep repeating polls). However, the methodological bias against consistency resulting from counting repeated polls as separate cases is not as severe as one might expect. When repeated polls are counted as one case, consistency rises to 55 per cent under Trudeau, 70 per cent under Mulroney and 53 per cent under Chrétien. Thus, the increase in consistency is very modest. More importantly, the relative magnitude of the variation in consistency across governments remains basically the same irrespective of the method used.
Public majorities supported change more often during both the Trudeau era (83 per cent favouring change) and the Chrétien era (70 per cent) than in the Mulroney era (62 per cent).
Why was there greater inertia and resistance to change during the government of Jean Chrétien than during the government of Brian Mulroney? One reason may be that they had markedly different leadership styles. Brian Mulroney was prime minister at a time of profound socio-economic transformations, and took a number of important initiatives that reversed the policies pursued by his predecessor Pierre E. Trudeau and radically transformed the political landscape (the Free Trade Agreement, GST, closer ties with the US). By contrast Jean Chrétien deliberately avoided fundamental changes after a balanced budget had been achieved. His stay in office in the post-deficit period was characterized by prudent management.11
For contrasting accounts of Brian Mulroney's and Jean Chrétien's leaderships, see Nossal (2003) and Savoie (2003). However, it should be noted that the passage of important measures in the environmental and social policy fields in the final year of Jean Chrétien's mandate (not recorded in this study) represented a reversal of his earlier record.
A more valid explanation of the recent decline in opinion-policy consistency finds its origin in the ideological divergence between public majorities and government policies. There is evidence from case studies that the Canadian public has been for some time now more conservative than policy makers on a variety of issues such as wiretapping (Fletcher, 1989), election rules (Blais and Gidengil, 1991), gun control (Mauser and Margolis, 1992) and unreasonable search and seizure (Fletcher and Howe, 2000), and there is no reason to believe that this trend has reversed in recent years. In fact, experts argue that Canadian values have undergone a shift to the right during the 1990s (Nevitte et al., 2000; see also Soroka and Wlezien, 2003). At the same time, it is a truism that many federal policies shifted to the left under Jean Chrétien relative to Brian Mulroney. So the gap between the ideologies of the Canadian public and the federal government, which was small under Mulroney, has increased under Chrétien. This suggests that inconsistency has increased in recent years as a result of a combination of the public supporting the right position on many issues and the government of Jean Chrétien opting for the left position on the same issues.
To test whether or not ideological divergences between the Canadian public and the federal government could be the source of the observed increase in inconsistency, we assigned a right or left position to public majorities based on how individual survey respondents voted at the last election.12
Many Canadians have difficulties recognizing ‘left’ from ‘right.’ However, electoral studies have shown that Canadian citizens organize their political values along clearly identifiable left-right dimensions, regardless of whether they recognize the dichotomy.
Scores for Liberal party and Bloc Québécois votes were ignored in the calculation. See below for further analyses incorporating a variable for Liberal party voting. We did not include other attributes (socio-economic status, occupation, sex, religion) of individual survey respondents because there were no relevant theoretical links to our research problem.
Table 3 cross-tabulates the direction of mass opinion, dichotomized into right and left public majorities, and records the subsequent consistent or inconsistent government actions. We report results only for the Chrétien era because the data about how survey respondents voted in the last election were too scarce for the Mulroney and Trudeau eras to generate reliable statistical results. The data show that inconsistency is far more frequent overall when public majorities take a right position on issues and the government adopts a left position on the same issues (36 cases overall), than when the public supports a left position and government prefers a right position (15 cases overall). It therefore appears that, aside from its reluctance to initiate policy changes desired by the public, the government of Jean Chrétien took a more leftist position than the Canadian public on many issues.
Table 3: Consistency and Majority Preference for the Right Side of Issues 1993–2001

Is there a relationship between government biases against change and against the right? The answer is found in Table 4, which cross-tabulates the ideological direction of public majorities, dichotomized into right or left positions, and the change vs. status quo dichotomy. The data indicate that there is a strong association between public majorities for change and support for right-wing positions on issues. Rightist public majorities are approximately three times more likely to support change (41 cases) than the status quo (14 cases). The evidence suggests that public majorities in support of policy change are much less likely to get what they want from the government, especially if the policy changes public majorities support have a more right-wing flavour than what the government is ready to deliver.
Table 4: Majority Preference for Change and for the Right Side of Issues 1993–2000

A more accurate understanding of consistency during the government of Jean Chrétien can be obtained by using a multivariate analysis. Table 5 presents six multivariate logistic regression models of consistency. The dependent variable takes the value 1 when policy is consistent with opinion and 0 otherwise. There are four control variables in the models. It is expected that a large majority regarding an issue will carry more weight than a bare majority. Consequently opinion-policy consistency should correlate positively with the extent of majority opinion. The variable is expressed as the difference between majority and minority opinion percentages on each poll. We also record the number of undecided respondents in the poll, a measure, albeit an imperfect one, of how salient the issue is to the public and therefore how “real” their expressed opinions are. Issue saliency is measured by the inverse proportion of respondents answering “don't know” or “no opinion” to survey questions. The higher the value of the indicator, the higher the saliency. Governments should feel more pressure to respond to issues that are salient to the public. Consequently, we hypothesize that consistency correlates positively with the measure. The models also include a dummy variable for seven newly introduced issues during the government of Jean Chrétien. Issues reported twice as often or more under Chrétien as compared to Mulroney are coded 1. They are: government cuts, deficit reduction, privatized health care, gun control, sexual orientation, mercy killing and young offenders (see Table 2). Other issues are coded 0. We hypothesize that the coefficient for new issues is negative.
Table 5: Multivariate Logistic Analyses of Consistency

We also look at the electoral cycle. It is generally expected that a party will try to pass key policies at the beginning of its term in office, in the year following an election, especially those policies on which it has been elected and that are supported by public majorities. Once it has passed its key policies, the governing party will be faced with issues that enjoy less popular support. New issues can also arise at that time that are not necessarily popular. We should, therefore, expect higher levels of consistency in the year immediately following an election. On the other hand, politicians have an incentive to respond to public opinion because they will be held accountable for their actions at election time. As election time approaches, the government might face greater pressure to respond to public preference. Consequently, we should expect that consistency will increase again as elections draw nearer. Here we hypothesize a curvilinear relationship between consistency and elections, and we operationalize the electoral cycle variable as a dummy taking the value 1 if a survey is administered in the years immediately before and immediately after an election, and 0 otherwise.14
We wish to thank the anonymous referee who pointed out the correct way to operationalize the electoral cycle variable.
Models 1 and 2 measure the effect of the variable for change on consistency during the period 1968–2001. The variable is a dummy coded 1 when majority opinion prefers change and 0 when the majority prefers the status quo. We hypothesize that change correlates negatively with consistency. We know from the previous discussion that the Mulroney government was less biased against policy change than the Trudeau and Chrétien governments, and consistency was substantially higher during the Mulroney years as a consequence. We therefore hypothesize a significantly positive coefficient for the Mulroney variable in the additive model (1) and for the change × Mulroney term in the interaction model (2).
From the additive model (1), we see that the estimate for the change variable is significant and in the correct direction. All other things being equal the predicted odds ratio of consistency when majority opinion supports change is e−0.75 = 0.46, less than half the odds of consistency when majority opinion supports the status quo. The coefficient for the Mulroney government variable is significantly positive, as predicted. The odds of consistency under Mulroney are twice the odds of consistency under Trudeau. The coefficient for the change × Mulroney variable in the interaction model (2) is positive and significant. This must be compared with the significantly negative coefficient for the change variable (−1.08) which represents the effect of change on consistency during the Trudeau years. Despite its overall negative effect on consistency, change is significantly more likely to be associated with consistency under Mulroney than under Trudeau. The coefficient for change in interaction with Chrétien is negative but not statistically significant, suggesting that there is not much difference in the effect of change on consistency between the Chrétien and Trudeau eras.
The coefficient for the extent of majority opinion is of correct sign and statistically significant, indicating that a greater consensus in mass opinion in favour of a particular direction produces more responsiveness from the government. This result is directly relevant to the issue of the use of polls by government decision makers. One side of the issue argues that decision makers trust polls as an accurate reflection of public opinion and often use them to gauge the direction of public opinion. The other side argues that decision makers don't trust polls and that they will often construct a view of public opinion more consistent with their own preferences and those of people around them, rather than consulting the actual state of mass opinion. What our finding shows is that such willful blindness is less possible when the majority in favour of a particular direction is very large. The estimates for issue salience are not significant, though we interpret this as evidence of the weakness of our measure.15
A high proportion of “don't know” responses may be a sign of low salience, but it may also be a sign of indecision on salient and contentious issues.
The next four models only consider the determinants of consistency during the Chrétien years. Models 3 and 4 measure the impact of the conservative public ideology on consistency in addition to, and in interaction with, the change vs. status quo direction of mass opinion. From the additive equation of model 3, we see that public majorities that support the right have a negative and significant impact on consistency. All other things being equal the predicted odds ratio of consistency when majority opinion supports the right side of an issue is e−0.64 = 0.42, indicating that public support for the right reduces the odds of consistency by more than half. The significantly negative coefficient for the change × ideology variable in model 4 suggests that there is a strong interaction between the two variables. Public majorities for change and public majorities in support of the right powerfully combine their effects to lower opinion-policy consistency.
Models 5 and 6 examine whether consistency during the Chrétien government is higher when public majorities are positively associated with a vote for the Liberal party. The variable for Liberal support is coded 1 whenever public majorities (whether in favour of change or the status quo) are positively and significantly associated with a Liberal vote in the last election, and 0 when public majorities are negatively associated with Liberal support. Cases in which the association is not statistically significant are eliminated from the sample. Note that a public majority in support of or against the Liberal party can also be a bipolar public majority. We could identify 99 cases (out of 123 cases that report the respondents' votes in the last election) in which public majorities were significantly associated with a vote for or against the Liberal party at the last election.
We expect that the Chrétien government is more responsive to public majorities that support the Liberal party. We also expect that Canadians who voted for the Liberals in the last election will support the policies of the Liberal government more often than those who voted for the opposition or abstained. It is therefore hypothesized that the coefficient for the Liberal support variable will be positive. From Model 5, we see that the additive effect of Liberal support is statistically significant and positive. The odds of consistency are multiplied by approximately four when a public majority is positively associated with Liberal support. However, the coefficient for the change × Liberal interaction variable in Model 6 is not statistically significant. This suggests that the effect of the variables for change and Liberal vote on consistency is additive rather than multiplicative. The Chrétien government is significantly more responsive to public majorities that support the Liberal party whether they prefer policy change or the status quo. Note also that Models 5 and 6 offer the best fit (R2 = .25). This is testimony to the powerful impact of public ideology on government responsiveness.
Conclusion
In this research, we attempted to provide convincing explanations as to why the government of Jean Chrétien was so much less responsive to majority opinion as measured by polls than the Mulroney government. It is now time to summarize our main findings and discuss their implications.
The analysis shows that lower responsiveness during the government of Jean Chrétien cannot be attributed to methodological reasons alone. There are also substantive explanations for the decline in consistency after 1993. Low responsiveness during the Chrétien government is associated with an increased bias against change in policy making as compared with the Mulroney government. The data suggest that the Chrétien government was significantly less responsive when mass opinion favoured a change of policy than when it supported already existing policy. Remarkably, there was no such bias against change and in favour of the status quo in policy making during the government of Brian Mulroney. The Mulroney government deferred to mass opinion more often when the public favoured policy change.
The increase in government resistance to change does not completely account for the difference in consistency from Mulroney to Chrétien, however. Another explanation for the observed variation in consistency is ideology. There is evidence that public majorities have supported the conservative side on a large number of issues throughout the period of analysis. It is no coincidence, therefore, that public opinion was highly consistent with the conservative policies of the government of Brian Mulroney, and that the election of the more liberal government of Jean Chrétien has led to lower consistency after 1993. The data also show that public majorities that were associated with Liberal party support were more likely to get what they wanted from the Chrétien government, irrespective of whether that was a policy change or the status quo. The Chrétien government appears to have been especially unresponsive on issues that were favoured by opposition parties' supporters. This suggests that officials in the Chrétien government made a distinction between the opinion of all Canadians and a Liberal public opinion, and that they were prepared to be more responsive toward the latter because it was more important to them. The finding could also be interpreted to mean that public majorities consisting of many Canadians who had voted for the Liberal party at the last election were more likely to support the policies of the Liberal government.
Our diagnosis is of course limited by the quantitative nature of our design. This research raises several important questions that cannot be answered in full at the moment because we could not address them satisfactorily within a quantitative framework. Only further qualitative research will allow for a proper response.16
We are currently pursuing such qualitative research involving elite interviews and tracing the opinion-policy relationship on an issue-by-issue basis using the case study method. For preliminary results of this research, see Mendelsohn and Petry (2001).
Another limit we faced in this quantitative analysis was that we could only measure consistency on equally weighted issues, although some issues clearly have a higher profile than others. We think that the considerations surrounding high profile issues are directly relevant to the apparent paradox of presumed attentiveness to public opinion and actual low responsiveness that was raised in the introduction. Actually, there is a further enigma involved here. If the policies of the Chrétien government were unresponsive to mass opinion or—using the alternative formulation—if a majority of the Canadian public does not support the policies of their government on a large number of issues, why then do Canadians keep re-electing the same party in government? Is it because Canadians are unaware that their government is unresponsive? We think that the answer depends largely on whether the issue is high profile or not. Recent electoral research suggests that Canadians are indeed aware when government is or is not responsive on high profile issues and rewards or punishes the government accordingly. Nevitte et al. (2000) show that government responsiveness on high profile issues such as deficit reduction generated extra support for the Liberal government in the 1997 election, and government non-responsiveness to public opinion on high profile issues such as keeping the GST cost the government some votes in the same election. Consequently, government officials have an incentive to be responsive to mass opinion as measured by polls on high profile issues. But on other, less important issues, whether the government is responsive or not—or whether Canadians agree or disagree with government policy—does not seem to have a perceptible effect on electoral support. We interpret this to mean that the Canadian public is either unaware of or unmoved by government non-responsiveness on lower profile issues. Since they anticipate that non-responsiveness on low profile issues is unlikely to cost them votes at the next election, government officials will be free to be non-responsive or to respond to their own definitions of public opinion on lower profile issues.
A related explanation as to why Canadians keep re-electing the Liberals despite government lack of responsiveness is that their assessment of government performance on high profile issues at election time is disconnected from their opinion of the day-to-day low profile decisions of the Chrétien government. A majority of Canadians did indeed support Jean Chrétien on his two most vital initiatives (deficit elimination and Canadian unity) whereas they were more ambivalent on lower profile issues. Interestingly, the pattern of public support (or government responsiveness) on high vs. low profile issues is reversed under Brian Mulroney. Whereas the Canadian public supported most of the Mulroney government's day-to-day decisions, the two most important measures enacted by the government (the Free Trade Agreement and the Goods and Services Tax) were never popular with the masses. The evidence suggests a counterintuitive negative correlation between government popularity (and attentiveness) on high profile issues and government responsiveness on lower profile issues. The more responsive government of Brian Mulroney was also less popular than the less responsive government of Jean Chrétien. However, the expected positive correlation is reestablished when consistency is redefined based on high profile issues only. Government popularity has perhaps less to do with responsiveness to mass opinion on day-to-day issues than to public preference on high profile issues.
Acknowledgments
This study was supported by the Social Sciences and Humanities Research Council of Canada. We wish to thank Mélisa Leclerc and Guillaume Laperierre for their research assistance and the Journal's anonymous referees for their suggestions.
References

Table 1: Consistency and Majority Preference for Policy Change (1968–2001)

Table 2: Consistency Broken Down by Issues 1984–1993 and 1994–2001

Table 3: Consistency and Majority Preference for the Right Side of Issues 1993–2001

Table 4: Majority Preference for Change and for the Right Side of Issues 1993–2000

Table 5: Multivariate Logistic Analyses of Consistency
- 37
- Cited by