¹ When referring specifically to the British case, we use ‘by-elections’; but when referring specifically to the American case or to the generic phenomenon, we employ ‘special elections’. The only paper comparing the two is a descriptive synthesis: Shelley, Mack II, ‘By-Elections and Government Stability in Britain and the United States, 1824–1983’, paper presented at the annual meeting of the American Political Science Association, Washington, D.C., 1984.Google Scholar

² Sigelman, Lee, ‘Special Elections to the U.S. House: Some Descriptive Generalizations’, Legislative Studies Quarterly, VI (1981), 577–88.CrossRefGoogle Scholar

³ Kincaid, Diane D., ‘Over His Dead Body: A Positive Perspective on Widows in the U.S. Congress’, Western Political Quarterly, XXXI (1978), 96–104.Google Scholar

⁴ Another point concerning terminology: in the United States, the party controlling the White House is the ‘presidential party’; in Britain, the party of the Prime Minister is the ‘governing party’; and in the generic case we employ this latter term.

⁵ Sigelman, , ‘Special Elections to the U.S. House’Google Scholar; Taylor, Stan and Payne, Clive, ‘Features of Electoral Behaviour at By-Elections’, in Cook, Chris and Ramsden, John, eds, By-Elections in British Politics (New York: St. Martin's Press, 1973), pp. 330–56.CrossRefGoogle Scholar

⁶ Taylor, and Payne, , ‘Features of Electoral Behaviour at By-Elections’Google Scholar; Stray, Stephanie and Silver, Mick, ‘Do By-Elections Demonstrate a Government's Unpopularity?’ Parliamentary Affairs, XXXIII (1980), 264–70CrossRefGoogle Scholar; Mughan, Anthony, ‘Toward a Political Explanation of Government Vote Losses in Midterm By-Elections’, American Political Science Review, LXXX (1986).Google Scholar

⁷ Although we began with all 362 British by-elections and all ninety-seven American special elections, the analyses reported below are based on 242 of the by-elections and sixty-six of the special elections. From the British data set we have dropped: by-elections in Northern Ireland; by-elections uncontested by one of the two major parties; pre-1950 university seats; and cases in which major boundary revisions before the 1950, 1955, February 1974 or 1983 general elections made it impossible to compare a by-election with the succeeding general election. Boundary changes rated as ‘minor’ – less than 5 per cent change by either Craig, F. W. S., British Electoral Facts, 1832–1980 (Chichester, Sussex: Parliamentary Research Services, 1981)Google Scholar, or the appropriate edition of The Times Guide to the House of Commons (London: Times Newspapers Limited, 1950–1983)Google Scholar – were considered still to be, in effect, the same seat. The 242 cases analysed here are, then, instances on which comparable data are available for a full three-election cycle – the general election preceding the by-election, the by-election itself, and the general election following the by-election. By-elections were coded by location (England, Scotland or Wales) and separate analyses were run for each region. These analyses are not reported below, since in every instance the regional findings were consistent with those that emerged in the analysis of all 242 elections. From the American data we have dropped: uncontested special elections; special elections for which all the information our models require is not available – a far greater problem for the United States, where election data are maintained (often informally) by separate boards in each state, than for Britain; and, as in Britain, special elections held in districts whose boundaries changed over the course of the three-election series on which our analysis focuses.

⁸ Measuring the governing party's performance in terms of its share of the electorate – registered voters in a constituency – indicates the party's success in mobilizing popular support. Although descriptive studies in both countries show turnout to be lower in special elections, Mughan, , ‘Toward a Political Explanation of Government Vote Losses’Google Scholar has cast doubt on whether turnout is as influential on British by-election results as previously thought. Despite the lack of turnout data for the United States, we analysed the effects of our independent variables on the government's share of the electorate in Britain. The results were very similar to those reported below for the governing party's share of the votes, but turnout itself was important. Full results are available from the authors on request.

⁹ Tufte, Edward R., ‘Determinants of the Outcomes of Midterm Congressional Elections’, American Political Science Review, LXIX (1975), 812–26CrossRefGoogle Scholar; Kernell, Samuel, ‘Presidential Popularity and Negative Voting: An Alternative Explanation of the Midterm Congressional Decline of the President's Party’, American Political Science Review, LXXII (1978), 44–66.Google Scholar

¹⁰ Kincaid, , ‘Over His Dead Body’.Google Scholar

¹¹ Kernell, Samuel, ‘Explaining Presidential Popularity’, American Political Science Review, LXXII (1978), 506–22.CrossRefGoogle Scholar

¹² Stimson, James A., ‘Public Support for American Presidents: A Cyclical Model’, Public Opinion Quarterly, XL (1976), 1–21CrossRefGoogle Scholar; Stray, Stephanie and Silver, Mick, ‘Government Popularity, By-Elections, and Cycles’, Parliamentary Affairs, XXXVI (1983), 49–55.CrossRefGoogle Scholar

¹³ For an elaboration, see Stray, Stephanie and Silver, Mick, ‘The Measurement of Change in the Popularity of Governments in United Kingdom By-Elections’, Political Methodology, VIII (1982), 93–106.Google Scholar

¹⁴ Recall, however, that the coefficient for presidential party control of a seat is not statistically significant. This effect, though large, has an almost equally large standard error attached to it, and the standard error of the prediction is consequently relatively large.

¹⁵ In fact, the government has a negative base (cushion) in the electorate (–4.1 per cent) even before party voting loyalty is taken into account; in other words, some former government supporters do not even turn out to vote. As noted in fn. 8, results are available from the authors on request.

¹⁶ There is no indication, however, that the time elapsed between the by-election and the succeeding general election has a positive effect on voters returning to the government fold, as claimed by Stray, and Silver, , ‘The Measurement of Change in the Popularity of Governments in United Kingdom By-Elections’.Google Scholar

¹⁷ Lewis-Beck, Michael S., ‘Election Forecasts in 1984: How Accurate Were They?’ PS, XVIII (1985), 53–62CrossRefGoogle Scholar. Strictly speaking, our models do not fulfil Lewis-Beck's remaining criterion of applicability before the election, because whether a minor party will surpass the 5 per cent threshold used in our definitions of intervention and withdrawals cannot be known before the fact. Had we not employed some such threshold, the concepts of intervention and withdrawal would be rather vacuous; intervention could occur, for example, if a new party received a single vote – hardly a meaningful intervention.