¹ The British General Election Study is a collaborative venture between Oxford University and Social and Community Planning Research and is directed by Anthony Heath, Roger Jowell, John Curtice, Julia Field and Sharon Witherspoon. We are particularly grateful to Denise Lievesley, Sharon Witherspoon and Kate Melvin at SCPR for their help with this study, and to David Butler, Peter Cozens, Ivor Crewe and Byron Shafer for their comments on our earlier drafts.

The BGES is funded jointly by the Sainsbury Trusts, the ESRC and Pergamon Press. It continues the series of election studies conducted by Butler and Stokes in 1964, 1966 and 1970, by Crewe, Särlvik, Alt and Robertson in 1974 and 1979 and by Heath, Jowell, Curtice and Field in 1983. The 1987 study consists of a nationally representative probability sample interviewed in the weeks following the 1987 general election.

² Different published sources give rather different figures for turnout in 1987. We have preferred the figures given in the official publication, Return of Election Expenses (London: HMSO, 1988), p. 9.Google Scholar This shows that 74.8 per cent of the registered electorate in Great Britain turned out to vote. In addition to the non-voters that are accounted for here, there are of course many people who do not vote but are not counted in the official figures for turnout because they are not even registered as electors. Because the sample on which this research is based is drawn from the electoral register, we are unable to add anything about them here. See Todd, Jean and Eldridge, Jack, Improving Electoral Registration (London: HMSO, 1987)Google Scholar; Michael, and Pinto-Duschinsky, Shelley, Voter Registration: Problems and Solutions (London: Constitutional Reform Centre, 1987).Google Scholar

³ We are very grateful to Enid Smith, Jill Adam and Anthea Feries of the Lord Chancellor's Department for their efficiency and good humour in the mammoth task of locating the records that we wished to see, and we are also grateful to Debs Ghate of Nuffield College for her assistance in checking through them.

⁴ Return of Election Expenses (London: HMSO, 1988), p. 9.Google Scholar

⁵ The only other British study that we know used turnout data in combination with survey material was conducted by Catherine Marsh during the 1983 election. The sample of the pre-election poll conducted in Cambridge was checked against the official record of voting in order to test, amongst other things, the validity of the propensity to vote question as a predictor of turnout. See Marsh, , ‘Predictions of Voting Behaviour from a Pre-election Survey’, Political Studies, 33 (1985), 642–8.CrossRefGoogle Scholar In the United States various scholars have used official records similar to our own to check the validity of respondents' reports of their turnout. See Traugott, Michael W. and Katosh, John P., ‘Response Validity in Surveys of Voting behavior’, Public Opinion Quarterly, 42 (1979), 359–77CrossRefGoogle Scholar; Anderson, Barbara A. and Silver, Brian D., ‘Measurement and Mismeasurement of the Validity of the Self-Reported Vote’, American Journal of Political Science, 30 (1986), 771–85.CrossRefGoogle Scholar

⁶ Some of the missing cases were due to the illegibility of the records, but most were due to the fact that the records for some half dozen whole constituencies were not available.

⁷ See Crewe, Ivor, Fox, Tony and All, Jim, ‘Non-voting in British General Elections, 1966-October 1974’ in Crouch, Colin, ed., British Political Sociology Yearbook, Volume 3: Participation in Politics (London: Croom Helm, 1977), p. 46.Google Scholar In 1983, 83 per cent of BGES respondents claimed to have voted, compared with the official figure of about 73 per cent.

⁸ See Traugott, and Katosh, , ‘Response Validity and Voting Behavior’, p. 364.Google Scholar

⁹ As Crewe, Fox and Alt remind us, ‘the official turnout level is expressed as a percentage of all those listed as eligible voters in the electoral register [whilst] survey figures … are calculated as a percentage of those interviewed’. See ‘Non-voting in British General Elections’, p. 108.Google Scholar

¹⁰ See Todd, Jean and Butcher, B., Electoral Registration in 1981 (London: OPCS, 1982).Google Scholar

¹¹ The assumption that all people who voted in respect of a given address were resident there at the qualifying date may not be wholly correct. Among the 129 respondents that were traced from the issued address to a new one, but who voted in respect of the issued address, no fewer than 18 said that they had been resident at their new address for a year or more, and thus could not have been resident at the issued address on the qualifying date. But it is not impossible, of course, that this puzzling result is due to coding errors.

¹² Crewe, , Fox, and Alt, , ‘Non-voting in British General Elections’, p. 46.Google Scholar

¹³ We would urge some caution in using turnout regularity as an analytical tool, however. Firstly, our check on electoral behaviour in 1987 indicates that even for the most recent election, respondents' reports are often incorrect, and, when several elections are being looked at, errors will accumulate. Secondly, because it only takes account of electors eligible to vote on every occasion, the measure excludes a significant number of respondents, especially the young. And thirdly, if a panel is not used, it forces us to rely on respondents' recollections of their past behaviour, which we know from other studies is also often unreliable. For instance, see Himmelweit, H. T., Siberian, J. and Stockdale, J., ‘Memory for Past Vote: Implications of a Study of Bias in Recall’, British Journal of Political Science, 8 (1978), 365–75.CrossRefGoogle Scholar

¹⁴ Crewe, , Fox, and Alt, , ‘Non-voting in British General Elections’, p. 64.Google Scholar

¹⁵ Crewe, , Fox, and Alt, , ‘Non-voting in British General Elections’, p. 59.Google Scholar

¹⁶ Crewe, , Fox, and Alt, , ‘Non-voting in British General Elections’, p. 63.Google Scholar

¹⁷ Our dependent variable thus becomes the log odds of voting or not voting. We therefore fit the following model, which uses the notation of Fienberg. See Fienberg, S. E., The Analysis of Cross-Classified Data, 2nd edn (Boston, Mass: MIT Press, 1987), chap. 6.Google Scholar

A represents age (three categories), R represents length of residence (a binary variable scored 1 if the respondent had been resident at the address for less than one year, 2 otherwise), M represents marital status (a binary variable scored 1 if the respondent was married, 2 otherwise) and T represents housing tenure (three categories). We use the SPSSx paramaterization in which the parameters for the categories of a given variable sum to zero. When the dependent variable is official turnout, the model yields a satisfactory fit (maximum likelihood chi square =24.3, 27 degrees of freedom, p = 0.612); the fit is less satisfactory when reported turnout is the dependent variable (chi square = 46.8, 27 df, p = 0.01).

¹⁸ See Aldrich, J. H. and Nelson, F. D., Linear Probability, Logit and Probit Models (Beverly Hills, Calif.: Sage, 1984).CrossRefGoogle Scholar

¹⁹ Table 11 gives the results of our final logit model, which drops tenure and adds respondent's class (three categories – a collapsed version of that used by Heath, Jowell and Curtice) and household income (two categories). The fit is acceptable both when official turnout is the dependent variable (chi square = 71.8, 56 df, p = 0.08) and when reported turnout is the dependent variable (chi square = 64.8, 56 df, p = 0.20).

²⁰ Todd, and Butcher, , Electoral Registration in 1981.Google Scholar While there have been earlier studies of electoral registration, their focus has been on under-registration rather than on the redundancy in the register. See, for instance, Gray, P. G., Corlett, T. and Frankland, P., The Register of Electors as a Sampling Frame (London: Central Office of Information, 1950).Google Scholar It appears from these sources that under-registration was rather lower in 1950 than it is today, and we have therefore assumed that registration was more efficient then with respect to redundancy too.