¹ For Britain, see Benney, M., Pear, R. H. and Gray, A. P., How People Vote (London: Routledge and Kegan Paul, 1956)Google Scholar; Milne, R. C. and McKenzie, H. C., Straight Fight (London: Hansard Society, 1954)Google Scholar; and Milne, R. C. and McKenzie, H. C., Marginal Seat (London: Hansard Society, 1958).Google Scholar For the United States, see Lazarsfeld, P. F., Berelson, B. and Gaudet, H., The People's Choice (New York: Columbia University Press, 1944)Google Scholar; and Berelson, B., Lazarsfeld, P. F. and McPhee, W., Voting (Chicago: University of Chicago Press, 1954).Google Scholar

² Butler, David and Stokes, Donald, Political Change in Britain, 1st edn. (Harmondsworth, Middx.: Penguin, 1971), p. 516.CrossRefGoogle Scholar

³ Butler, David and Stokes, Donald, Political Change in Britain, 2nd edn. (London: Macmillan, 1974), p. 486.CrossRefGoogle Scholar

⁴ This volatility index is suggested in Butler, D. and Kavanagh, D., The British General Election of February 1974 (London: Macmillan, 1974), p. 137.CrossRefGoogle Scholar The October figures are calculated from Butler, D. and Kavanagh, D., The British General Election of October 1974 (London: Macmillan, 1975), pp. 190–1.CrossRefGoogle Scholar

⁵ The data used in this analysis were collected by the Gallup polling organization in a two-wave panel study conducted during the October 1974 campaign. A random sample of 1,015 respondents was interviewed in the early days of the campaign and 744 of these same respondents were reinterviewed two or three days before the actual election. This second sample, weighted to take account of changes in the final days of the campaign, is the one on which my analysis is based. Of these 744 respondents, only those intending to vote for a specific party in both waves of the study are included in my data base. This is because Gallup did not give respondents the option of answering ‘don't know’, ‘undecided’ or ‘no answer’ to the voting intention question so that their residual category is very ambiguous in meaning. This left 728 respondents in the weighted sample, ninety-four of whom changed their vote. When broken down by party, the number changing to or from a particular party ranged from eighteen to forty-seven. To minimize the possible inferential problems involved in working with relatively small numbers of changers, my conclusions will be based on patterns in the data rather than on individual parameter estimates. For the sake of comparability with the Butler and Stokes data on campaign vote changers, this figure of 10·9 per cent is based on the unweighted sample. In the weighted sample, the same figure is 12·9 per cent.

⁶ The 1966 figure is calculated from Butler, and Stokes, , Political Change, 1st edn., p. 516.Google Scholar

⁷ Rose, R., ed., The Polls and the 1970 Election (Glasgow: University of Strathclyde Survey Research Centre, Occasional Paper Number 7, 1970), p. 53.Google Scholar

⁸ King, A., ‘The Election that Someone Won – More or Less’, in Penniman, H. R., ed., Britain at the Polls: The Parliamentary Elections of 1974 (Washington, D.C.: American Enterprise Institute, 1975), pp. 192 and 195.Google Scholar For a similar description of the pattern of campaign change in February, see King, A., ‘The Election that Everyone Lost’Google Scholar, in Penniman, , Britain at the Polls, p. 27.Google Scholar

⁹ Rose, R., ‘The Polls and Public Opinion in October 1974’Google Scholar, in Penniman, , Britain at the Polls, p. 228.Google Scholar

¹⁰ Butler, and Kavanagh, , The British General Election of February 1974, p. 137.Google Scholar

¹¹ Various explanations of the electorate's greater volatility have been explored in great detail in I. Crewe, Särlvik, B. and Alt, J., ‘Partisan Dealignment in Britain 1964–1974’, British Journal of Political Science, VII (1977), 129–90.Google Scholar See also, Franklin, M. N. and Mughan, A., ‘The Decline of Class Voting in Britain: Problems of Analysis and Interpretation’, American Political Science Review, LXXII (1978, forthcoming).Google Scholar As regards the study of electoral change during the campaign, it should be acknowledged at this point that the form of my analysis has been severely determined by the type of data available. That is, the party evaluations are used because they are the only really feasible correlates of voting change in the data set. Ideally, one would have designed a questionnaire similar in detail and complexity to those that go into the Reid immediately after an election and applied it to a random sample of the electorate at several points during the campaign. However, the immense financial and technical problems involved in launching such a project in Britain, especially when an election is called prematurely as in October 1974, forced me to rely on data collected for other purposes by commercial polling agencies. But, although limited in many respects, studies of this kind should be judged by the validity and interest of their findings and not the extensiveness of their data base. Thus, I hope that the value of my own study will become apparent as the analysis proceeds.

¹² The exact questions are: ‘Taking everything into account, which party has the best policies?’, ‘And which party has the best leaders?’, ‘Who would make the better Prime Minister: Mr Wilson, Mr Heath or Mr Thorpe?’, and, in response to questions asking which are the most and next most urgent domestic and foreign affairs problems facing the country at the present time, ‘Which party do you think can best handle that problem (the important problems facing the country)?’ All of them are closed-ended and respondents were presented with a list of parties or party leaders before answering the individual questions.

¹³ Butler, and Kavanagh, , The British General Election of October 1974, p. 102Google Scholar and Chap. 5 generally. See also Pinto-Duschinsky, M., ‘False Cairn: Party Strategies in October 1974’Google Scholar, in Penniman, , Britain at the Polls, pp. 201–22.Google Scholar

¹⁴ Buller, and Kavanagh, , The British General Election of October 1974, p. 82.Google Scholar

¹⁵ Crewe, I., Särlvik, B. and Alt, J., ‘The Why and How of Voting in February 1974’, in Rose, R., ed., Studies in British Politics, 3rd edn. (London: Macmillan, 1976), p. 247.Google Scholar

¹⁶ Butler, and Kavanagh, , The British General Election of October 1974, p. 272.Google Scholar

¹⁷ For a detailed analysis of Liberal support in 1974, see Alt, J., Crewe, I. and Särlvik, B., ‘Angels in Plastic: The Liberal Surge in 1974’. Political Studies, XXV (1977). 343–68.CrossRefGoogle Scholar

¹⁸ Regression analysis is used because it overcomes the problems of bivariate correlation analysis and because its coefficients are easily interpreted. Its use is not meant to imply a uni-directional causal flow from evaluation to voting change. In all probability, some part of the total relationship between the two types of change is attributable to some respondents rationalizing their party evaluation to conform to a prior decision to change their vote. The respective contributions of the two patterns of change to the total relationship can only be estimated by means of a complex, non-recursive causal model that is simply beyond the scope of this data. The regression coefficients, therefore, should be interpreted as measures of association and not of cause and effect.

¹⁹ An alternative strategy to this simplifying assumption would be to enter the four variables corresponding to defection from Labour and the eight measuring favourable change of evaluation for the Conservative and Liberal parties into the regression equation at the same time. This would have produced independent estimates of the impact of all twelve variables. However, because multiple regression averages the variance shared by inter-correlated predictor variables, some of these variables might have been made artificially insignificant (in a statistical sense) solely because their shared variance would have been averaged over so many variables. To avoid this problem, I made the simplifying assumption described in the text. For a fuller discussion of the shared variance problem, see Gordon, R. A., ‘Issues in Multiple Regression’, American Journal of Sociology, LXXIII (1968), 592–616.CrossRefGoogle Scholar

²⁰ These are phi correlation coefficients, which are measures of association between dichotomized variables that are the equivalent of Pearson's r. For an extended and illustrated discussion of multiple regression analysis with dichotomized variables, see Franklin and Mughan, ‘The Decline of Class Voting in Britain’. The measurement of evaluation change has already been described in the text. Voting change is measured in the same way. If a respondent changes his vote from Labour to Liberal between the two waves of the panel he is classified as both a Labour defector and Liberal switcher, although his behaviour is analysed separately in the two roles to coincide with corresponding evaluation changes. It would be interesting to have been able to specify not only which party voters switched to or defected from, but also which party switchers originally supported and which one defectors ended up supporting. Thus, for example, one would not just have examined evaluation change among Labour defectors, but would have distinguished between Labour defectors who switched to the Conservatives and those who switched to the Liberals. Unfortunately, this more detailed analysis was not possible because of the small numbers of switchers in the sample.

²¹ For a general discussion of the more immediate and long-term problems facing the country in 1974, see Butler, and Kavanagh, , The British General Election of October 1974, Chap. I.Google Scholar

²² Obviously, if a respondent were to make similar changes of evaluation on three or more dimensions he would be still more likely to make a corresponding vote change. However, the number of vote changers to and from each party were too small to allow the construction of interaction terms measuring similar changes on more than two evaluation dimensions.

²³ Individual regression coefficients for each term in the equation are not presented in this table because of the severe problems of collinearity that the addition of interaction terms brings to multiple regression analysis. See Blalock, H. M. Jr., Social Statistics, 2nd edn. (New York: McGraw-Hill, 1972), pp. 463–4.Google Scholar