1 See Cairns, Alan C., “The Electoral System and the Party System in Canada, 1921–1965,” this Journal 1 (1968), 55–80Google Scholar; and Johnston, Richard and Ballantyne, Janet, “Geography and the Electoral System,” this Journal 10 (1977), 857–66Google Scholar. For attempts to quantify this relationship for Canada see Sankoff, David and Mellos, Koula, “La régionalisation éléctorate et I'amplification des proportions,” this Journal 6 (1973), 380–98Google Scholar; Quaker, T. H., “Seats and Votes: An Application of the Cube Law to the Canadian Electoral System,” this Journal 1 (1968), 336–44; and William J. Linehan and Philip A. Schrodt, “A New Test of the Cube Law,” Political Methodology (forthcoming).Google Scholar

2 See Spafford, Duff, “The Electoral System in Canada,” American Political Science Review 64 (1970), 168–76. While Spafford's analysis is inspired by the votes/seats question, of particular interest is his specification of the impact of minor party candidacies on seat shares.CrossRefGoogle Scholar

3 Lovink, J. A. A., “Is Canadian Politics Too Competitive?,” this Journal 6 (1973), 343–79, contains information on the distribution of “safe,” “competitive,” and “marginal” federal seats in different historical periods. The “index of safety” used in that classification is based on frequency of turnover and margins of victory. The index is used as one of the independent variables in this study. For details of its calculation see p. 361.Google Scholar

4 Some empirical examples of the operation of this phenomenon are contained in Ennis, Philip H., “The Contextual Dimension in Voting,” in McPhee, W. N. and Glazer, W. A. (eds.). Public Opinion and Congressional Elections (Glencoe: Free Press, 1962)Google Scholar; and Kevin R. Cox, “The Spatial Structuring of Information Flow and Partisan Attitudes,” and Segal, David R. and Meyer, Marshall W., “The Social Context of Political Partisanship,” both in Dogan, M. and Rokkan, S. (eds.), Social Ecology (Cambridge: MIT Press, 1974), chaps. 7 and 9, respectively.Google Scholar

5 For indirect evidence of this see this author, “LIP and Partisanship: An Analysis of the Local Initiatives Program,” Canadian Public Policy 2 (1976), 17–32CrossRefGoogle Scholar;and Munro, John M., “Highways in British Columbia: Economics and Politics,” Canadian Journal of Economics 8 (1975), 192–204.CrossRefGoogle Scholar

6 For recent studies which use the individual as the unit of analysis but bring environmental information to bear, see William P. Irvine, “Testing Explanations of Voter Turnout in Canada,” and “Testing Models of Voting Choice in Canada,” in Budge, Ian, Crewe, Ivor and Fairlie, Denis (eds.), Party Identification and Beyond: Representations of Voting and Party Competition (London: Wiley, 1976)Google Scholar, chaps. 17 and 18, respectively; and Wright, Gerald C. Jr., “Contextual Models of Electoral Behavior: The Southern Wallace Vote,” American Political Science Review 71 (1977), 497–508CrossRefGoogle Scholar. For formal treatments of the testing of contextual propositions see Przeworski, Adam, “Contextual Models of Political Behavior,” Political Methodology 1 (1974), 27–61Google Scholar; and Sprague, John, “Improving Theoretical Parameter Estimation in Contextual Models by Combining Sample Survey and Aggregate Observations,” a paper presented to the annual meeting of the American Political Science Association,San Francisco, 1975.Google Scholar

7 For a statement of this position see Barton, Allan H., “Bringing Society Back In,” American Behavioral Scientist 12(1968), 1–9. As Barton expresses it, “…as usually practiced, using random sampling of individuals, the survey is a sociological meatgrinder, tearing the individual from his social context and guaranteeing that nobody in the study interacts with anyone else in it” (1).CrossRefGoogle Scholar

8 Classic examples are Key, V. O. Jr, Southern Politics in State and Nation (New York: Random House, 1949)Google Scholar; Tingsten, Herbert, Political Behavior: Studies in Election Statistics (Totowa: Bedminster, originally published in 1937)Google Scholar; and Siegfried, andre, Tableau Politique de la France de I'Ouest sous la Troisieme Republique (Paris: Armand Colin, 1964, originally published in 1913).Google Scholar

9 “A Study of the Economic and Racial Basis of Conservatism and Liberalism in 1930,” abridged in Courtney, John C. (ed.), Voting in Canada (Toronto: Prentice-Hall, 1967), 77, 80. Reid's article was first published in 1933.Google Scholar

10 “Politics and Social Class in Canada: The Case of Waterloo South,” this Journal 1 (1968), 307Google Scholar. This perspective, or variations of it, has guided the work of a number of students of Canadian politics, including this author. See “Regionalism in Canadian Voting Patterns,” this Journal 5 (1972), 55–81Google Scholar. For more recent examples see Schwartz, Mildred, Politics and Territory: The Sociology of Regional Persistence in Canada (Montreal: McGill-Queen's, 1974)CrossRefGoogle Scholar; and the examination of provincial party systems by Jenson, Jane, in Bellamy, D. J., Pammett, J. H., and Rowat, D. C., The Provincial Political Systems (Toronto: Methuen, 1976), chap. 9.Google Scholar

11 Irvine, “Voter Turnout,” 342.

12 “Voter Turnout Among the American States: Systemic and Individual Components,” American Political Science Review 69 (1975), 107–23.CrossRefGoogle Scholar

13 The authors acknowledge that they have undoubtedly underestimated the impact of individual factors because of their limited coverage of them and the possibility of interaction between individual characteristics and systemic factors. For an exploration of the interaction factor see the comment on their paper by Douglas D. Rose, Ibid., 125–31.

14 The principal investigator for this study was John Meisel. Total N for the study was 2,767. The data were made available by the UBC Data Library. Neither the principal investigator nor the Data Library are responsible for use made of the data in this paper.

15 There is a continuing debate in the literature as to whether least squares regression is appropriate given a dichotomous dependent variable because the procedure may produce predicted values outside the interval (0,1). A number of studies have compared the results obtained in analyzing such variables using regression, transformations of it, and maximum likelihood methods such as logit and probit. There is no consensus as to whether the statistically purer methods produce substantially different results or results different enough to justify their extra cost and lesser intelligibility to most political scientists. For comparisons of this sort see John Aldrich and Charles Cnudde, “Probing the Bounds of Conventional Wisdom: A Comparison of Regression, Probit and Discriminant Analysis,” American Journal of Political Science 19 (1975), 571–608; and Morley Gunderson, “Retention of Trainees: A Study with Dichotomous Dependent Variables,” Journal of'Econometrics (1974), 79–93. Neither study demonstrates that the use of ordinary least squares regression leads to appreciable errors of interpretation, hence our decision to use it.

16 These probabilities are undoubtedly too high. The equation was estimated after excluding nonvoters and those who refused to answer the education, income, religion, or vote questions. Moreover, the survey overestimates Liberal vote by about 10 percentage points. In the procedure outlined, adjustments are made for these factors.

17 Kimetal., “Voter Turnout,” 111, n. 26, report an R of .35 which indicates an R² of .12.

18 For details of the calculation procedure and a comparison of the survey and census distributions of independent variable characteristics see Appendix.

19 This procedure assumes that the individual-level relationship between vote and social characteristics is the same in every riding and that individual effects are additive. We are thus ignoring the possibility of interaction effects among these background characteristics and between them and systemic characteristics. On the basis of earlier work by the author (see above, n. 9) and others it would seem that these assumptions are unrealistic. However, careful examination of residuals should reveal any gross violations of these assumptions and help to indicate which systemic characteristics may be involved in any interaction. Determination of the latter is, of course, a major goal of this study.

20 This figure is arrived at by subtracting from the residual remaining after individual and systemic effects are removed in each constituency, the provincial mean value of that residual (column 10 in Table 3 for BC). The resultant values for each riding (column 11, Table 3) represent the deviation unexplained by individual, systemic, and provincial variation.

21 Our study is cross-sectional in design, but we are intrigued by the fact that R. P. Woolstencroft's analysis of the variance in Liberal party support over the three elections preceding that of 1968 reveals a large (73%) constituency-level component and a small (7%) provincial component. Our individual and systemic effects (explaining 78 per cent of the 1968 cross-sectional variance) are, of course, measured at the constituency-level and might be labelled “constituency effects.” See his “The Interplay Between Geography and Politics: The Case of Canada, 1953 to 1965,” a paper presented to the 1975 annual meeting of the Canadian Political Science Association, Edmonton, Table 15, p. 19. However, we note that Robert Jackman, using a variation of the procedure employed by Woolstencroft, forthe same party and time period, reports a 53 per cent provincial effect and a 45 per cent riding effect. See his “Parties, Voting and National Integration: The Canadian Case,” Comparative Politics 4 (1972), Table 3, p. 526.Google Scholar

22 “Voting Choice,” 361.

23 Ibid., 361–62.

24 This interpretation and the distinctiveness of Liberal/Social Credit contests suggests reconsideration of Irvine's notion of “extrinsic value” may be in order.

25 Although residuals are, on the average, negative for ridings featuring a Liberal/Social Credit clash, the mean residual is less negative (indicating less overprediction) where the Liberals defeated Social Credit contenders. The possibility that distinguishing between Liberal win/loss situations in the way we have (via interaction terms involving the dummy variable LIBWIN) introduces linear dependence in our model is addressed in the Appendix.

26 See Segal and Meyer, “The Social Context of Political Partisanship,” 220–24, for the impact of “neighbourhood SES” on the rate of Republican voting of high and low SES individuals. The study is based on a 1963 survey of nine towns in the northeastern United States. Of course, “occupation” was not used as a variable when individual effects were being calculated (see Table 1), and it might be argued that the significance of NONMANUAL is due to that omission. The fact that the related variables of income and education were utilized makes this possibility unlikely. The decision not to use occupation in addition to or instead of one of these variables when calculating individual effects was based on the difficulty of finding correspondence between census and survey classifications of occupation.

27 The Vertical Mosaic: An Analysis of Social Class and Power in Canada (Toronto: University of Toronto Press, 1965).Google Scholar

28 This is clearly suggested by the presence of important ethnic/regional interaction effects on Liberal and Conservative support between 1953 and 1965, particularly those involving Eastern Europeans in western Canada. See this author's “Regionalism in Voting Patterns.”

29 The significance of this variable does not seem to be merely an artifact of the decision not to include ethnic origin variables in the individual-level model. When constituencies outside Quebec are dichotomized into those with 33 per cent or more “noncharter” populations and those with less than that, the mean residuals are:

The number of constituencies in each group is given in parentheses below the mean. As suggested, problems of overprediction are most pronounced in the Liberal Loss/High “Non-Charter” category.

30 “Surge and Decline: A Study of Electoral Change,” in Campbell, Angus et al., Elections and the Political Order (New York: Wiley, 1966), chap. 3. Campbell's argument is not completely appropriate for our situation because it is bound up with questions of strength of party identification, something not considered in our approach, and focusses on change in turnout between elections. Because the 1968 election was the first held on a new set of riding boundaries, turnout change effects could not be evaluated.Google Scholar

31 For a fascinating study aimed at producing state-level predictions for the 1960 and 1964 US presidential elections, based on a much larger number of individual characteristics, see Pool, I. de S., Abelson, R., and Popkin, S., Candidates, Issues and Strategies (Cambridge: MIT Press, 1965).Google Scholar

32 An analogous phenomenon, but hurting the Liberals, by implication operates in working-class ridings. We note, however, the small size of the coefficient involved.

33 Party competition has an ambiguous status. On the one hand, individual behaviour is responsible for the degree of competitiveness which characterizes the riding. On the other hand, the degree of competitiveness is a factor individual voters may take into account in their voting decisions. The latter, of course, is one of Irvine's main interests.

34 See Barton, “Bringing Society Back In,” for suggestions.

35 Of course, one could, particularly using the Canadian national academic surveys which use federal ridings as a sampling basis, explore the behaviour of individuals in different socioeconomic contexts. The major drawback of this alternative from our perspective is that respondents are usually drawn from fewer than half the ridings, reducing the variability of the environment.

36 “Voter Turnout Among the American States,” n. 30.