1 See Table 13.1 in Dalton, Russell J., “Political Cleavages, Issues, and Electoral Change,” in LeDuc, Lawrence, Niemi, Richard G. and Norris, Pippa, eds., Comparing Democracies: Elections and Voting in Global Perspective (Thousand Oaks: Sage, 1996), 325.Google Scholar

2 Blais, André, Gidengil, Elisabeth, Nadeau, Richard and Nevitte, Neil, “Accounting for the Vote in the 1997 Canadian Election,”paper presented at the annual meeting of the Canadian Political Science Association, Ottawa, 1998.Google Scholar

3 Weaver, R. Kent, “Improving Representation in the Canadian House of Commons,” this Journal 30 (1997), 473–512.Google Scholar

4 The gap in Reform support between the West and Atlantic Canada cannot be examined. Support for Reform was so low in Atlantic Canada that the dependent variable would be too skewed to allow for reliable inferences. The gap in Liberal support between Quebec and Ontario is not susceptible to the analytic approach adopted here. We focus on regional differences in the effects of a common set of explanatory factors, but in Quebec, the most important correlate of vote choice was attitudes about sovereignty (see Blais et al., “Accounting for the Vote”). Given the available sample size, we are necessarily ignoring intraregional differences in vote choice.

5 See, for example, Irvine, William P., “Assessing Regional Effects in Data Analysis,” this Journal 4 (1971), 21–24Google Scholar; Blake, Donald E., “The Measurement of Regionalism in Canadian Voting Patterns,” this Journal 5 (1972), 54–81Google Scholar, and “Constituency Contexts and Canadian Elections: An Exploratory Study,” this Journal 11 (1978), 279–305Google Scholar; and Schwartz, Mildred, Politics and Territory (Montreal: McGill-Queen's University Press, 1974).CrossRefGoogle Scholar

6 Simeon, Richard and Elkins, David J., “Regional Political Cultures in Canada,” this Journal 7 (1974), 397–437.Google Scholar

7 Blake, “The Measurement of Regionalism”; and Irvine, “Assessing Regional Effects.” Compare Ian McAllister and Studlar, Donley T., “Region and Voting in Britain, 1979–87: Territorial Polarization or Artifact?” American Journal of Political Science 36 (1992), 168–199.Google Scholar

8 Compare Simeon and Elkins, “Regional Political Cultures”; and McAllister and Studlar, “Region and Voting.”

9 Why these differences might exist is beyond the scope of this article, but possible candidates include formative historical events and the cultural baggage of the first significant waves of settlers. See, for example, Bell, David V. J., The Roots of Disunity: A Study of Canadian Political Culture (2nd ed.; Toronto: Oxford University Press, 1992)Google Scholar; Lipset, Seymour M., Continental Divide (New York: Routledge, 1990)Google Scholar; Stewart, Ian, “All the King's Horses: The Study of Canadian Political Culture,” in Gagnon, Alain-G. and Bickerton, James P., eds., Canadian Politics: An Introduction to the Discipline (2nd ed.; Peterborough: Broadview, 1994), 75–92Google Scholar; and Wiseman, Nelson, “The Pattern of Prairie Politics,” Queen's Quarterly 88 (1981), 298–315.Google Scholar

10 Blais, André and Nadeau, Richard, “Measuring Strategic Voting,” Electoral Studies 15 (1996), 39–52CrossRefGoogle Scholar; Blais, Andreé, Nadeau, Richard, Gidengil, Elisabeth and Nevitte, Neil, “Voting Stategically Against the Winner: The 1997 Canadian Election,”paper presented at the annual meeting of the Midwest Political Science Association, Chicago, 1998Google Scholar; and Cox, Gary, Making Votes Count: Strategic Coordination in the World's Electoral Systems (Cambridge: Cambridge University Press, 1997).CrossRefGoogle Scholar

11 Kraus, Vered, “Group Differences: The Issue of Decomposition,” Quality and Quantity 20 (1986), 181–190.CrossRefGoogle Scholar

12 The original solution to the problem of which group to choose as the standard was to use the average of the two groups, but using a mean of means only makes sense when the two groups are of roughly equal size. This is the case, for example, in decomposing gender gaps. See Gidengil, Elisabeth, “Economic Manmdash; Social Woman? The Case of the Gender Gap in Support for the Canada-United States Free Trade Agreement,” Comparative Political Studies 28 (1995), 384–408CrossRefGoogle Scholar; and Gilens, Martin, “Gender and Support for Reagan: A Comprehensive Model of Presidential Approval,” American Journal of Political Science 32 (1988), 19–49CrossRefGoogle Scholar. Kraus's reformulation gives identical results when the population groups are the same size.

13 The study consists of a three-wave survey conducted by the Institute for Social Research at York University. The study was funded by the Social Sciences and Humanities Research Council of Canada under its Major Collaborative Research Initiatives Programme and by Elections Canada. The data are available at: www.isr.yorku.ca/ISR. During the campaign, 3,949 eligible voters were interviewed, of whom 3,170 were re-interviewed after the election and 1,727 returned a mailback questionnaire. The campaign survey response rate was 59 per cent. See Northrup, David A., The 1997 Canadian Election Study: Technical Documentation (North York: Institute for Social Research, 1998Google Scholar). The data analyzed here are drawn exclusively from the campaign and post-election surveys. The relatively small size of the sample in Atlantic Canada has necessitated the use of corrective weights to bring respondents' reported vote into correspondence with the actual vote. Correcting only for oversampling of the Atlantic provinces, the Liberal vote was underestimated by 5 percentage points. For Ontario, the reported Liberal vote was 1.1 percentage points higher than the true value, while the Reform vote was within 0.5 percentage points of the true value. For the West, the Liberal vote was underestimated by 0.9 percentage points, while the Reform vote was overestimated by 2.8 percentage points.

14 With a binary dependent variable, estimation by maximum likelihood might be preferred, but the results of a decomposition using estimation by MLE would be much more complex and difficult to interpret. Estimation by ordinary least squares will normally give similar results, provided that the aggregate predictions lie within the .20 to .80 range. See Gillespie, M. W., “Log-Linear Techniques and the Regression of Dummy Dependent Variables: Further Bases for Comparison,” Sociological Methods and Research 6 (1977), 103–122.CrossRefGoogle Scholar

15 This represents a theoretically meaningful origin since a neutral position cannot influence vote one way or the other. The neutral position includes respondents who answered “don't know.” While this was done partly to conserve sample size, it is defensible on theoretical grounds since a lack of opinion should not move the vote for or against any one party.

16 Age, religion and ethnicity were all converted into a series of dummy variables.For age, the dummies were under 25 years, 25 to 39 years, 50 to 64 years and 65 years or older. For religion, the dummies were Catholic, non-Christian and none.For ethnicity, the dummies were French, northern European and non-European. The regional numbers for southern and eastern Europeans were too small for meaningful analysis.

17 Education was transformed into two dummy variables, representing university graduates and high school dropouts. Household income was adjusted for household size and transformed into two dummy variables representing the lowest and highest income quintiles.

18 Scales were used to measure orientations toward outgroups, accommodating Quebec, continentalism, cynicism, free enterprise and moral traditionalism. See the Appendix for details of the factor analysis procedure used to derive these scales. Single items were used for the federal government's treatment of the province (cpsj12) and orientations toward women, which were represented by views about how much should be done for women (pesel). When items from the mailback are included in the factor analysis, the latter item loads on a more general orientations toward women factor.

19 Economic perceptions included retrospective and prospective evaluations of personal finances (cpscl, cpsc2), Canada's economy (cpsg1, cpsg3a) and the provincial economy (cpsg2, cpsg3b), fear of losing one's job (cpsm8a), and perceived change in the unemployment rate and the future unemployment rate (cpsc5, cpsc6). Based on a factor analysis (see Appendix), four scales were used to measure attitudes toward spending cuts, taxes, deficit reduction and job creation. Single items were included for gun control (pese12), youth crime (cpsj21) and immigration (cpsj18). The dimensions of Liberal performance examined were preserving national unity, reducing the deficit, creating jobs, keeping election promises, fighting crime and protecting social programmes (cpsf1Oa, b,c,d,f,g). Leader evaluations were the difference between evaluations of the party's leader and the other party leader with the highest evaluation.

20 LeDuc, Lawrence, Clarke, Harold D., Jenson, Jane and Pammett, Jon, “Partisan Instability in Canada: Evidence from a New Panel Study,” American Political Science Review 71 (1984), 470–484CrossRefGoogle Scholar. But see Johnston, Richard, “Party Identification Measures in the Anglo-American Democracies: A National Survey Experiment,” American Journal of Political Science 36 (1992), 542–559.CrossRefGoogle Scholar

21 Kam, Christopher, “Party Identification in Canada: Re-Examining the Concept,”paper presented at the annual meeting of the Canadian Political Science Association, Ottawa, 1998, 28.Google Scholar

22 Each party's chances in the riding are expressed as the difference between its expected standardized chances and those of the top contender among the other parties. If the difference is positive, the variable equals 0. Thus the variable is on a scale from -1 to 0. The assumption is that the more the party was perceived to be trailing in the riding, the less likely people would be to vote for it. Chances of forming the official opposition is a dummy variable that equals 1 if the party was perceived to have the best chances of forming the official opposition and 0 otherwise.

23 The contribution of differences in effects can be found in the column headed “effect component.” The figures in that column give the estimated contribution to the vote gap of regional differences in the effects of social background on the vote. The figures should again be multiplied by 100 to obtain the percentage point contribution. To take the West-Ontario gap in Reform voting as an example, the fact that rural residence had a much stronger effect on Reform voting in the West contributed 1.3 percentage points to the total gap. The positive sign indicates that the gap would have been 1.3 percentage points less if rural residence had had the same effect in both regions.

24 Our model over explains the gap by .8 percentage points. Obviously, it should be bome in mind that our estimates of the contribution of the various factors are just approximations.

25 It should be noted, however, that the effect of the latter is of borderline statistical significance.

26 It might be objected that the Reform party is too new for identification to be considered deeply rooted. Reform voters, however, did show impressive levels of loyalty in the 1997 election. See Blais et al., “Accounting for the Vote.” Even if we enter party identification after economic perceptions, issue positions, incumbent performance and leader rating—and thus make the unrealistic assumption that any link between these judgments and party identification runs entirely from the former to the latter—a 1 percentage point contribution remains.

27 Even (unrealistically) entering party identification after economic perceptions, issues, incumbent performance and leader evaluations leaves a net contribution of 4.3 percentage points attributable to party identification alone.

28 Blake, “The Measurement of Regionalism,” 59. The other manifestation was “inconsistencies in the strength or direction of party support on the part of members of the same religious, ethnic, or class group living in different regions” (60).