1 See, for example, Markus, Gregory B., “Stability and Change in Political Attitudes: Observed, Recalled, and ‘Explained,’” Political Behavior 8 (1986), 21–44CrossRefGoogle Scholar; Niemi, Richard G., Katz, Richard S. and Newman, David, “Reconstructing Past Partisanship: The Failure of the Party Identification Recall Questions,” American Journal of Political Science 24 (1980), 633–51CrossRefGoogle Scholar; van der Eijk, C. and Niemoller, B., “Recall Accuracy and its Determinants,” Acta Politica 14 (1979), 289–342Google Scholar; Smith, T. W, “Recalling Attitudes: An Analysis of Retrospective Questions on the 1982 GSS,” Public Opinion Quarterly 48 (1984), 639–49CrossRefGoogle Scholar; Weir, B. T., “The Distortion of Voter Recall,” American Journal of Political Science 19 (1975), 53–62.CrossRefGoogle Scholar

2 Recall data is a respondent's current report of a behaviour or attitude that is said to have been engaged in or held sometime in the past. As most election studies are post-election interviews, even the respondent's reported vote is recalled, though in most instances the time span is short and the incidence of elections sufficiently extraordinary that the party voted for is unlikely to be forgotten.

3 Markus, “Stability and Change in Political Attitudes,” 40–41.

4 Campbell, Angus, Converse, Philip E., Miller, Warren E. and Stokes, Donald E., The American Voter (New York: Wiley, 1960), 133.Google Scholar

5 Ibid., 141.

6 Markus, “Stability and Change in Political Attitudes,” 22.

7 Data from the 1974, 1979 and 1980 Canadian National Election Studies, funded by the Social Sciences and Humanities Research Council of Canada, were made available by the study's principal investigators. The data were collected by Harold D. Clarke, Jane Jenson, Lawrence LeDucand Jon H. Pammett. The original collectors of the data and the SSHRC bear no responsibility for the analysis and interpretations presented here.

8 It is common practice to assign an identification to respondents who must be coaxed into admitting attachment to a party. As usual, the numbers in this instance are small: 47 respondents in 1974, 35 in 1979 and 47 in 1980 had to be coaxed into declaring an identification. Two further considerations argued for the inclusion of these respondents: many required cajoling on just one occasion, so that their exclusion would have reduced the sample by as much as 15 per cent; and those who required prodding to reveal their partisan feelings were not a great deal more likely to be inconsistent in their reports of past identification than were those who admitted an identification following just one probe. Of all those who had to be coaxed into claiming an identification at one wave or another (N = 129), 38 per cent had consistent records of recall while 62 per cent committed one of the errors described below.

9 The current reports of party identification were compared to determine whether a respondent consistently claimed the same identification. Those with missing data at any one of the three waves were considered to be consistent in their identification if they reported identifying with the same party in the two remaining waves.

10 I have used the terms “inconsistency” and sometimes “error” in the text below. The use of these words is driven by theoretical expectations: from the point of view of party identification theory, an “inconsistency” may be consistent with the theory in that a new attachment will overcome the memory of a prior one, but from the point of view of the researcher matching responses, an “inconsistency” appears less problematic.

11 “Inconsistency” refers to the record of responses and not to their theoretical interpretation.

12 Respondents in the three-wave panel showed a consistent pattern of heightened interest during elections and diminished interest between campaigns. When respondents were asked, “We have found that people sometimes don't pay too much attention to elections. How about yourself? Would you say that you were very interested in the recent federal election, fairly interested, slightly interested, or not at all interested in it?”, 35 per cent of the panel respondents said that they were very interested in the 1974 election. That figure increased to 40 per cent in 1979 and 43 per cent in 1980. However, when the same respondents were asked whether they paid much attention to politics generally, “from day to day when there isn't a big election campaign going on,” only 17 per cent said that they followed politics very closely in 1974, 1979 and 1980.

13 It is conceivable that some respondents could have interpreted the question about ever having felt closer to another party as not implying a different past identification. In other words, a Liberal identifier might at some time have felt closer to the Conservative party than at the time of the current interview while remaining a consistent Liberal identifier.

14 Of the 134 respondents with a Type 1 inconsistency, 77 also had a Type 2 and 3 also had a Type 4 inconsistency. Type 2 inconsistents, of which there were 276, included 28 who had a Type 3 and 30 who had a Type 4 error. Sixteen of the 60 Type 3 inconsistents also had Type 4 errors.

15 Research by van der Eijk and Neimoller reports a strong relationship (tau b=.64) between the consistency of voting for the same party and the ability to recall past votes accurately. See van der Eijk and Neimoller, “Recall Accuracy and its Determinants,” 312.

16 Campbell et al., The American Voter, 144, and Clarke, Harold D., Jenson, Jane, Leduc, Lawrence and Pammett, Jon H., Political Choice in Canada (Toronto: McGraw-Hill Ryerson, 1979), 306.Google Scholar

17 Clarke et al., Political Choice in Canada, 306.

18 Interest in politics was measured using the 1974 question. The question wording and response frequencies are summarized in note 12 above.

19 The education measure used was the respondent's 1979 report of the number of years in school: “How many years of school did you attend?” The responses were recoded into three categories: those who completed primary school only, attended high school but did not complete it, and completed high school and went on to more advanced levels.

20 When entered into the regression analysis below, neither the average strength of attachment calculated across the three waves nor the strength of attachment to a party in 1980 were important predictors of inconsistencies in recall when all other variables were controlled for. The mean strength of attachment across all three waves for those whose recall was completely accurate was 1.4 on a scale from one to three, while for those with inconsistent recall the mean was 1.5.

21 The memory variable was an additive scale ranging from one to three. If respondents claimed the same federal party identification for their fathers in 1974 and 1979, they scored one point. Similarly consistent reports of mothers’ federal party identification in 1974 and 1979 were awarded one point. A third point was scored if respondents recalled in both 1974 and 1979 that they voted for the same party in the 1972 election.

22 Those considered to have consistent vote records either voted for the same party in all three waves or for the same party in two waves if the third record was missing.

23 Forty-nine per cent of the respondents (N = 816) voted and identified with the same party in all three waves, 10 per cent voted for the same party but claimed different identifications, 26 per cent changed their vote and their identification and 15 per cent identified with the same party but voted for different parties.

24 Niemi et al., “Reconstructing Past Partisanship,” 649.

25 Campbell et al., The American Voter, 128–36.

26 Ibid.