1 Ford, Richard, Independence Day (Toronto: Little Brown, 1995), 36.Google Scholar

2 The year 1974 marks a qualitative shift in the inter-election pattern, as earlier government shares exhibited no special inter-election rhythm. Monthly analysis is possible only for the period since 1974, and only with the Gallup poll, There is little reason to believe that analyses with other polls would alter this article's conclusions, although this conjecture begs substantiation. To my mind, the outstanding question about the Gallup poll's generalizability is the wording of its vote question (see Appendix), which does not name any parties. This places a premium on recall of party labels, and may privilege big parties over small ones and old parties over new ones.

3 First observed by Nadeau, Richard, “L'effet lune de miel dans une contexte parlementaire: le cas canadien,”this Journal 23 (1990), 483–497.Google Scholar

4 The canonical source on the theory of the political business cycle is Nordhaus, William, “The Political Business Cycle,” Review of Economic Studies 42 (1975), 169–190CrossRefGoogle Scholar. A useful, acerbic, review of the model is Lewis-Beck, Michael, Economics and Elections: The Major Western Democracies (Ann Arbor: University of Michigan Press, 1988), chap. 9.Google Scholar

5 Monroe, Kristen and Erickson, Lynda, “The Economy and Political Support: The Canadian Case,” Journal of Politics 48 (1986), 616–647CrossRefGoogle Scholar, and Clarke, Harold D. and Zuk, Gary, “The Politics of Party Popularity: Canada 1974–79,” Comparative Politics 19 (1987), 299–315CrossRefGoogle Scholar.

6 Findings are not clear-cut, however. Happy, J. R. found unemployment not to matter in “Voter Sensitivity to Economic Conditions: A Canadian-American Comparison,” Comparative Politics 19 (1986), 45–56CrossRefGoogle Scholar, and in “Economic Performance and Retrospective Voting in Canadian Federal Elections,” this Journal 22 (1989), 377–387Google Scholar, but the contrary in “The Effects of Economic and Fiscal Performance in Incumbency Voting: The Canadian Case,” British Journal of Political Science 22 (1992), 117–130CrossRefGoogle Scholar. Calum Carmichael found outright reversal of signs for key economic variables between 1972 and 1974 (“Economic Conditions and the Popularity of the Incumbent Party in Canada,” this Journal 23 [1990], 713–726Google Scholar). The Happy and Carmichael studies all employed time-series/ cross-section setups. The one national time-series estimation for electoral data identified the impact of unemployment (Nadeau, Richard and Blais, André, “Explaining Election Outcomes in Canada: Economy and Politics,” this Journal 26 [1993], 775–790Google Scholar).

7 Archer, Keith and Johnson, Marquis, “Inflation, Unemployment and Canadian Federal Voting Behaviour,” this Journal 21 (1988), 569–584.Google Scholar

8 Clarke, Harold D. and Kornberg, Allan, “Support for the Canadian Federal Progressive Conservative Party since 1988: The Impact of Economic Evaluations and Economic Issues,” this Journal 25 (1992), 29–53.Google Scholar

9 Robert M. Campbell questions the depth of postwar Keynesian commitments but also makes clear that the inflation which began to plague the system in the late 1960s shifted the terms of debate (Grand Illusions: The Politics of the Keynesian Experience in Canada 1945–1975 [Peterborough: Broadview, 1987Google Scholar]). Campbell identifies most budgets brought down by Minister of Finance John Turner in the early 1970s as embodying a supply-side logic and, as such, neither contractionary nor dirigiste. Stephen McBride argues forcefully that 1975 signalled the final breakdown of the Keynesian project, most importantly a shift to anti-inflationary monetary growth targets (Not Working: State, Unemployment and Neo-Conservatism in Canada [Toronto: University of Toronto Press, 1992]Google Scholar).

10 One of this Journal's referees suggested shifting analysis from macro-economic targets to budgetary instruments, with special attention paid to the previous year's budget. I translated the suggestion into analysis in various ways, but none seemed persuasive. The most powerful setup involved replacing macro-economic outcome variables with the budget deficit/surplus as a percentage of GDP and with the bank rate, each variable measured at t–1. This produced slightly less robust results than those in Tables 1 and 2. The impact of the bank rate was negative, as we would expect. But the impact of the deficit was positive (bigger deficit, lower popularity), which suggests that what really drove the relationship was not the deficit as such, but the deficit as an indicator of deficient demand. This leads us back to the variables in the body of the text. I also moved the time lag back to t – 12, to try to capture the previous year's budget, but this effort led nowhere.

11 A useful summary of work outside the US is Lewis-Beck, Economics and Elections, esp. chap. 2. For a comprehensive US summary, with references down to 1992, see Ostrom, Charles W. and Smith, Renée M., “Error Correction, Attitude Persistence, and Executive Rewards and Punishments: A Behavioral Theory of Presidential Approval,” in Freeman, John R., ed., Political Analysis, Vol. 4:1992 (Ann Arbor: University of Michigan Press, 1993Google Scholar), Table 1.

12 Okun's Law is first stated in Okun, Arthur, “Potential GNP: Its Measurement and Significance,”American Statistical Association,Proceedings of the Business and Economic Statistics Section, 1962Google Scholar. The most spirited discussion of the law in the popularity-function literature, by Hibbs, Douglas A., is accompanied by routine inclusion of all three variables (The American Political Economy: Macroeconomics and Electoral Politics in the United States [Cambridge: Harvard University Press, 1987]Google Scholar).

13 Some studies use the Implicit Price Index (the GDP deflator) as the basis for inflation calculations. See, for instance, Monroe and Erickson, “The Economy and Political Support.” Most use the CPI, however, on the grounds that this is what citizens see, even if it is an inferior measure of true inflation. For personal income change, two choices had to be made. First is the period for measuring change. Year-over-year change seems to be the norm, and seems intuitively closest to the spirit of most discussions of income effects. See, for example, Kiewiet, D. Roderick and Rivers, Douglas, “The Economic Basis of Reagan's Appeal,” in Chubb, John E. and Peterson, Paul E., eds., The New Direction in American Politics (Washington: Brookings Institution, 1985), 69–90Google Scholar. The second choice is between total or disposable income. As a monthly series, the former is markedly more accessible, and earlier Canadian work (see, for instance, Johnston, Richard, Public Opinion and Public Policy in Canada: Questions of Confidence [Toronto: University of Toronto Press, 1986]Google Scholar) suggests that the two are close substitutes. Hence income change here means year-over-year change in real personal income per capita. For more detail on the data, see the Appendix.

14 Holbrook, Thomas M., Do Campaigns Matter? (Thousand Oaks: Sage, 1996), chap. 4.Google Scholar

15 On this, see Courtney, John C., Do Conventions Matter? Choosing National Party Leaders in Canada (Montreal: McGill-Queen's University Press, 1995), esp. 85ff.Google Scholar

16 The 1990 Liberal selection is the hardest to characterize, since John Turner announced his intention to step down on May 3, 1989, long before the party was able to convene, and it became necessary to declare a campaign moratorium that summer. Although covert campaigning occurred, it received little media coverage. As a news event the campaign for the June 1990 convention began in January of that year, so Courtney suggests (Do Conventions Matter? Table 5–3). The other three campaigns were naturally demarcated by the resignations in 1984 and 1993 of prime ministers Trudeau and Mulroney, respectively, and in 1983 by Joe Clark's announcement that his leadership review support was insufficient and his call for the leadership convention later that year. The 1976 Conservative race presented a challenge akin to that for the Liberals in 1990, but data limitations put that race outside the period for estimation.

17 Clark and Zuk find an impact from the 1976 Conservative leadership change, for instance (“The Politics of Party Popularity”).

18 Both Clark and Zuk (ibid.), and Monroe and Erickson (“The Economy and Political Support”) model the 1977 event as a crisis. Monroe and Erickson also include the 1970 “October Crisis.”

19 On specification issues, see, for instance, Hibbs, The American Political Economy, 162–64. The heart of the matter is whether or not a political variable is correlated with an economic one. Exclusion of the political variable would then constitute specification error, which in turn produces bias and inconsistency in the economic estimates. I experimented with political variables to test robustness in estimates. Singling out 1977 as a dummy variable to capture the post-1976 panic affected neither the economic coefficients nor their standard errors. Including dummy variables for each of the five campaigns modestly reduced the size of the income and inflation coefficients.

20 The expression seems to originate with Ostrom and Smith (“Error Correction”).

21 This article finesses the relationship between “approval” and “popularity.” But see Hudson, John, “The Relationship between Government Popularity and Approval for the Government's Record in the United Kingdom,” British Journal of Political Science 16 (1985), 165–186CrossRefGoogle Scholar, and the discussion below.

22 Partial adjustment setups once seemed the industry standard. For substantive arguments in their favour, see Kernell, Samuel, “Explaining Presidential Popularity,” American Political Science Review 72 (1978), 506–522CrossRefGoogle Scholar. But Nathaniel Beck argues that Kernell's logic is in fact better captured by the error-correction model, and that the setup in equation (3), a Koyck lag, is best thought of as a proxy for an exponentially distributed lag process (“Comparing Dynamic Specifications: The Case of Presidential Approval,” in Stimson, James A., ed., Political Analysis, Vol. 1: 1991 [Ann Arbor: University of Michigan Press, 1992], 64–67CrossRefGoogle Scholar). Hibbs (The American Political Economy) specifically links his Koyck setup (chap. 5) to the language of distributed lags.

23 An incidental effect of the fact that A_t–1 is now a lagged endogenous variable is that R² will be much higher for estimations based odsn (3) than for ones based on (2).

24 Remember that δ has a negative value, so –δ is positive. The expression in the text is the sum of an infinite series, reflecting the Koyck-lag setup in (3). Although only the previous period's value for A appears on the right-hand side, earlier periods are present by implication. For an interval of k months, the impact at t of A_t–k is (1 + δ)^k. This A_t–k carries impact from X_t–k–1, yielding atime path of marginally diminishing effect.

25 Estimation starts in 1977 rather than 1974 to accommodate availability of the seasonally adjusted unemployment series in Cansim. I was unable to link this series to earlier hard-copy versions of the series. For reasons described below, this also helps with graphics. Estimation assumes first-order autoregressive errors, and captures them by a Hildreth-Liu grid search. For each change of party in government (1979, 1980, 1984), post-election observations are purged to suspend impact of economic forces in place under the previous government; the data are then Kalman-filtered. In the error-correction framework of Table 1, three observations are purged; the partial adjustment setup requires only one deletion per change of government. Estimation is by SHAZAM. I considered estimating a cointegrated system but decided that straightforward estimation of single-equation models was both sufficient to the task and more consistent with the logic of government popularity. First of all, co-integrated systems link variables that are random walks, and the dependent variable is not one, nor should we expect it to be. Random walks presuppose that earlier information remains fully relevant to later transactions, as in efficient markets; for such an argument see Sheffrin, Steven M., Rational Expectations (Cambridge: Cambridge University Press, 1983Google Scholar), chap. 4.1 do expect government popularity to have a memory (precisely this is embodied in the coefficients involving δ) but not an infinitely long one; for an argument along these lines, see Beck, Nathaniel, “The Methodology of Cointegration,” in Freeman, , ed., Political Analysis, Vol. 4, 240–241Google Scholar. Co-integrated systems further assume that either part of the system can respond to the other, in this case, popularity to the economy and vice versa. But in this article's time scale, causality flows strictly from economy to popularity, in contrast with, say, the mutual adjustment of various interest rates to each other. Again, see ibid., 241–42.

26 In the partial-adjustment setup, the real income coefficient has less than a 2 percent chance of being the product of random covariation; the chance for the inflation coefficient is 3 per cent. In the error correcting setup, the chances are, respectively, 7 per cent and just under 10 per cent. All tests are two-tailed.

27 I also performed estimations for 1977–1984 and 1984–1993 subperiods. The pattern in (1) comes through most clearly for the second period, while unemployment dominates the first period. Both patterns point to an emphasis on aggregate demand as the key to government popularity.

28 Predicted values cannot be generated directly from the 1977–1993 equation in Table 2, as the closeness of the relationship between A_t, and A_t–1 makes predicted and observed lines track each other with misleading accuracy. The trick is to confine prediction to truly exogenous variables, yet allow their effects to be distributed in time. To this end, I lagged values on relevant exogenous variables 24 months, incorporated all lags into the estimation, and discounted each lag term by 1 + δ taken to the relevant exponent, thus:

where β_i is the coefficient on X_i-. The 24-month cumulation brings us very close to the asymptotic impact implied by β_i1–δ. Using an error-correction setup would necessitate another set of lag terms for each variable, with little impact on values of Â_t. I used only real income and inflation terms to estimate Â_t, as only these were statistically significant and as doing so allows us to extend 24-month lag terms back before 1977. This exercise is akin to taking an indefinite integral and so lacks a constant, but adding the difference between the predicted and observed 1976–1993 means to the predicted series brings predicted values into the domain of observed ones.

29 It thus partly reaffirms claims by Clarke and Komberg (“Support”) and in Clarke, Harold D., Jenson, Jane, LeDuc, Lawrence and Pammett, Jon H., Absent Mandate: Canadian Politics in an Era of Restructuring (3rd ed.; Toronto: Gage, 1995).Google Scholar

30 The most extensive documentation is Brody, Richard A., Assessing the President: The Media, Elite Opinion, and Public Support (Stanford: Stanford University Press, 1991Google Scholar). James A. Stimson argues for a full cycle, although weaker in recovery than in drop (“Public Support for American Presidents: A Cyclical Model,” Public Opinion Quarterly 40 [1976], 1–21CrossRefGoogle Scholar). Brody finds no such pattern. The classic evidence on mid-term losses is Tufte, Edward R., “Determinants of the Outcomes of Midterm Congressional Elections,” American Political Science Review 69 (1975), 812–826.CrossRefGoogle Scholar

31 A striking early claim for a UK cycle in popularity was by Goodhart, C. A. E. and Bhansali, R. J., “Political Economy,” Political Studies 18 (1970), 43–106CrossRefGoogle Scholar. Further claims can be found in Miller, W. L. and Mackie, M., “The Electoral Cycle and the Asymmetry of Government and Opposition Popularity: An Alternative Model of the Relationships between Economic Conditions and Political Popularity,” Political Studies 21 (1973), 263–279CrossRefGoogle Scholar, and in Pissarides, C., “British Government Popularity and Economic Performance,” Economic Journal 90 (1980), 569–581CrossRefGoogle Scholar. For a counter-claim see Hudson, “The Relationship.” On by-elections, see Miller and Mackie, “The Electoral Cycle.” The only relevant Canadian evidence is provided by Barry J. Kay who finds that governments typically lose ground in by-elections but do not recover it at the next election (“By-Elections as Indicators of Canadian Voting,” this Journal 14 [1981], 37–52Google Scholar). By-elections, thus, have been reasonable precursors of the next outcome rather than manifestations of an endogenous cycle. It may be telling, though, that Kay's analysis ends in 1979.

32 This was explicitly the argument in Stimson, “Political Support,” and seems implied in Miller and Mackie, “The Electoral Cycle.”

33 Mueller, John E., War, Presidents, and Public Opinion (New York: Wiley, 1973Google Scholar), chap. 9. Mueller's setup purports to represent linear growth in a “coalition of minorities.”

34 Brody, Assessing the President, esp. part 2.

35 The language in the body of the text hints that approval and popularity are not the same thing, a point central to the analysis by Hudson (“Trie Relationship”). The process may be like “negative voting,” as in Kemell, Samuel, “Presidential Popularity and Negative Voting: An Alternative Explanation of the Midterm Congressional Decline of the President's Party,” American Political Science Review 71 (1977), 44–66.Google Scholar

36 For the mid-1950s, see the discussion in Johnston, Richard, “Canada,” in Shafer, Byron E., ed., Postwar Politics in the G-7: Orders and Eras in Comparative Perspective (Madison: University of Wisconsin Press, 1996), 62–63.Google Scholar