¹ Crewe, Ivor, ‘Is Britain's Two-Party System Really About to Crumble?’ Electoral Studies, 1 (1982), 275–314.CrossRefGoogle Scholar

² See Rose, Richard, The Problem of Party Government (London: Macmillan, 1974), Chap. 2CrossRefGoogle Scholar; Crewe, Ivor, Alt, James and Särlvik, Bo, ‘Partisan Dealignment in Britain 1964–1974’, British Journal of Political Science, VII (1977), 275–314Google Scholar; Franklin, Mark N. and Mughan, Anthony, ‘The Decline of Class Voting in Britain: Problems of Analysis and Interpretation’, American Political Science Review, LXXII (1978), 523–34CrossRefGoogle Scholar. This last exposition also contains a discussion of the different ways in which the decline can be measured and interpreted.

³ Franklin, Mark N., ‘Demographic and Political Components in the Decline of British Class Voting, 1964–1979’, Electoral Studies, 1 (1982), 195–220.CrossRefGoogle Scholar

⁴ Franklin, , ‘Demographic and Political Components’, p. 217.Google Scholar

⁵ Franklin, , ‘Demographic and Political Components’, p. 218.Google Scholar

⁶ Franklin, , ‘Demographic and Political Components’, pp. 203–9.Google Scholar

⁷ Miller, William L., ‘Social Class and Party Choice in England: A New Analysis’, British Journal of Political Science, VIII (1978), 257–84.CrossRefGoogle Scholar

⁸ The first three of these surveys were conducted by David Butler and Donald Stokes and kindly made available by the Inter-University Consortium for Political and Social Research at Ann Arbor, Michigan. Those fielded in 1974 and 1979 were conducted by Ivor Crewe, Bo Särlvik, James Alt and David Robertson, and made available by the British SSRC Data Archive at the University of Essex. These surveys were acquired by the Social Statistics Laboratory at the University of Strathclyde, and their analysis was much facilitated by preliminary work performed by the staff of the Laboratory, and particularly by Ann Mair for whose patient assistance I am much indebted. At the time of writing, no surveys incorporating childhood characteristics are yet available for the 1983 general election. Data from the 1983 Gallup surveys which was ‘spliced’ to the series illustrated in Figure i is not appropriate for the analyses conducted in the remainder of this article.

⁹ Because of the increasing proportion of children attending comprehensive schools, years of education were dichotomized to distinguish those who left school at the minimum legal age from those who stayed on at school after that time. It was felt that this would provide an indicator of classroom environment that should remain relatively stable over the fifteen-year period covered by our study.

¹⁰ The impact of childhood socializing variables on adult political behaviour has been extensively analysed by Butler, David and Stokes, Donald in their Political Change in Britain (London: Macmillan, 1974)CrossRefGoogle Scholar, Chap. 5. The variables chosen to indicate adult influences are the most important of those whose impact was evaluated in Franklin, and Mughan, 's ‘The Decline of Class Voting in Britain’, passimGoogle Scholar. The impact of socializing forces throughout adult life as well as during childhood is so well established as to appear beyond the need for documentation; but in recent works, Patrick Dunleavy has asserted that ‘contagion’ (as he conceives the forces of socialization) has no theoretical basis and rests on mere ‘empiricist’ observation. See his ‘The Urban Basis of Political Alignment: Social Class, Domestic Property Ownership and State Intervention in Consumption Processes’, British Journal of Political Science, IX (1979), 409–43.Google Scholar

On the contrary, I would contend that socializing forces are well-understood, theoretically grounded, empirically confirmed mechanisms going back at least to Sumner, William Graham's Folkways (New Haven, Conn.: Yale University Press, 1916)Google Scholar, if not to de Toqueville. To dismiss them is to dismiss at a stroke the major findings not only of electoral research but of most of sociology, anthropology and developmental psychology. Anyone seeking to sweep so clean must surely do more than deny ‘that political alignment brushes off on people by rubbing shoulders in the street’. See Dunleavy, , ‘The Urban Basis of Political Alignment’, p. 413Google Scholar. That is a parody of the socialization thesis to which no ‘empiricist’ would subscribe. See Franklin, Mark N. and Page, Edward, ‘A Critique of the Consumption Cleavage Approach in British Voting Studies’, Political Studies, XXXII (1984)Google Scholar; and also Harrop, Martin, ‘The Urban Basis of Political Alignment: A Comment’, British Journal of Political Science, X (1980), 288–98.Google Scholar

¹¹ Katz, Richard S., Niemi, Richard D. and Newman, David, ‘Reconstructing Past Partisanship in Britain’, British Journal of Political Science, X (1980), 505–15.CrossRefGoogle Scholar

¹² Jennings, M. Kent and Niemi, Richard D., Generations and Politics (Princeton, N.J.: Princeton University Press, 1981), p. 42Google Scholar. In the case of parents' class, we can obtain further confirmation of the small degree of contamination by considering the extent of response change between successive waves of a panel survey. If respondents were to misremember childhood influences to bring them in line with current perceptions then any change in current perceptions should be reflected in some change in perceived childhood influences. Between 1963 and 1966 only 55 per cent of respondents to the post-election surveys gave consistent and unchanging responses to questions on childhood class and present party identification, but only half of one per cent changed both responses in such a way as to keep them consistent, and the correlation between change in perception of childhood class and change in current party identification was -0·076. Small negative relationships were also obtained between change in childhood class and change in current social class, both occupational and self-assessed.

¹³ Franklin, , ‘Demographic and Political Components’Google Scholar. Details of this coding scheme are given in a separate appendix available from the author on request.

¹⁴ When Conservative voting is taken as the dependent variable, class voting appears generally less, but the trends established in this article remain much the same.

¹⁵ Regression analysis is not necessarily the best method of computing average coefficients from data that could be represented as a multi-way contingency table. Because it assumes normally distributed interval variables, it can be subject to bias when making estimates from dichotomies. See Goodman, Leo, ‘The Relationship between Modified and Usual Regression Approaches to the Analysis of Dichotomous Variables’ in Heise, David R., ed., Sociological Methodology 1976 (San Francisco: Jossey-Bass, 1975), pp. 83–110Google Scholar. However, other research has shown such bias to be negligible in a situation similar to the one that faces us here. See Franklin, Mark N. and Mughan, Anthony, ‘Communication’, American Political Science Review, LXXIV (1980), 462–4CrossRefGoogle Scholar. The method is employed for lack of an alternative means of causal analysis suited to this number of variables.

¹⁶ The coefficients are Pearsonian product-moment correlations. The model differs from the only other British voting model of similar (or greater) complexity in a number of respects. See Himmelweit, Hilda T., Humphreys, Patrick, Jaeger, Marianne and Katz, Michael, How Voters Decide (London: Academic Press, 1981), especially pp. 68–71Google Scholar. Himmelweit and associates were concerned with individuals all of whom entered the electorate in 1959, and each of whom thus had the opportunity of voting in an identical number of general elections. Past voting behaviour becomes a plausible component of such a model, replacing the role of parents' party in our own model and dominating the resulting picture at the expense of such variables as housing and union membership. Moreover, attitudinal variables are introduced as mediating between social characteristics and voting choice. However, the major difference between this model and my own is the inclusion of a variable, educational attainment, which I eliminated because, despite its large effect for the small number of individuals who did receive higher education, added very little to the overall variance explained.

¹⁷ I.e., 50 plus 38 out of 1,208 (Figure 2).

¹⁸ This is similar to the ‘backward elimination’ procedure detailed in Draper, N. R. and Smith, H., Applied Regression Analysis (New York: Wiley. 1966), pp. 167–8.Google Scholar

¹⁹ Franklin, , ‘Demographic and Political Components’, pp. 203–12.Google Scholar

²⁰ They are non-comparable in the extent of minor-party voting, but this difficulty will be allowed for in the analysis that follows. All the models were subjected to a number of tests to determine their adequacy, as recommended by Asher, Herbert B. in his Causal Modeling (Beverley Hills, Calif.: Sage, 1976), pp. 33–4Google Scholar. Since the information needed to replicate them for the models of interest is included in the article, all that is necessary here is to report that the outcome of these tests was satisfactory.

²¹ This is explained in Franklin, and Mughan, , ‘The Decline of Class Voting in Britain’, especially in Table 3.Google Scholar

²² All the analyses reported in this article were performed using pairwise deletion of missing data. The danger inherent in this method lies in the possibility that the case base for some coefficients could be very different from the case base for other coefficients. In order to guard against the possibility of being misled in this fashion, all the analyses were repeated with missing data being replaced by least-squares estimates of their true values. Happily the results of all these analyses corresponded closely to the results presented here, though the coefficients were generally of a somewhat lower magnitude, reflecting errors in estimation. The data manipulation involved in all these analyses was extensive and could hardly have been attempted without instantaneous access to intermediate findings as different coding schemes were tried and different regression models evaluated. The computer package used was SCSS. See Nie, Norman H., Hull, C. Hadlai, Franklin, Mark N. et al. , SCSS: A User's Guide to the SCSS Conversational System (New York: McGraw-Hill, 1980).Google Scholar

²³ See for example Butler, and Stokes, , Political Change, pp. 70–3Google Scholar; Rose, , Party Government, pp. 143–52Google Scholar; Crewe, , Särlvik, and Alt, , ‘Partisan Dealignment’, pp. 168–81Google Scholar. The major dissenting voice among researchers into British voting behaviour comes from Patrick Dunleavy who suggests in ‘The Urban Basis of Political Alignment’ that British voting behaviour is determined not by the social influences people experience in the home, school and workplace but by real economic interests which have in recent years come to be determined by consumption patterns (but see above, fn. 10). Because of his focus on interests rather than influences, Dunleavy's analysis falls outside the tradition within which the present argument is constructed.

²⁴ Separate analysis of the Butler and Stokes 1964 survey shows that two-thirds (68·1 per cent) of the individuals whose class self-image conflicted with their occupational status had class origins consistent with this self-image.

²⁵ The model leaves out certain variables that Butler and Stokes emphasize in their discussions of voting choice, particularly the effects of the local political environment (Political Change, Chap. 6). However, Butler and Stokes did not conduct a multivariate analysis with all the variables they considered important. When these are brought together in a model such as the one depicted in Figure 4, some of them turn out to be redundant. This was shown in Rose, , Party Government, Chap. 2Google Scholar, and in Franklin, and Mughan, , ‘The Decline of Class Voting in Britain’, pp. 529–30Google Scholar. These other factors are already subsumed within a model as detailed as the one presented here. A respondent with all six working-class characteristics included in the model is most likely also to live in a heavily working-class environment, and the reverse is true of one with six middle-class characteristics.

²⁶ Pulzer, Peter, Political Representation and Elections in Britain (London: Allen & Unwin, 1967), p. 89.Google Scholar

²⁷ Not all past researchers have considered supportive variables to be ‘embellishments’. Rose, for example (in Party Government and elsewhere), considers class to be a matter of reinforcing or cross-cutting social characteristics which yield a few ‘ideal type’ class members with all the characteristics typical of that class. Other members of the population approximate more or less to the ideal. However, Rose does not construct a causal model of class voting, and it is not implausible to suggest that my model underlies even Rose's eclectic approach.

²⁸ Franklin, Mark N., ‘The Rise of Issue Voting in British Elections’ (Glasgow: Strathclyde Papers on Government and Politics No. 3, 1983).Google Scholar

²⁹ Butler, David and King, Anthony, The British General Election of 1964 (London: Macmillan, 1975). p. 143.Google Scholar

³⁰ Butler, and King, , The British General Election of 1964, p. 139.Google Scholar

³¹ Franklin, and Page, , ‘The Consumption Cleavage Approach’, Table 3.Google Scholar

³² It must be emphasized that the attribution of influences to prior variables in this fashion results directly from the causal ordering of the model, which was determined on the basis of theoretical considerations. Once this ordering is established, indirect effects result from the rules of algebra, and do nothing to confirm the correct definition of the model.

³³ The same result would have been obtained in a stagewise regression analysis which took out the effects of central variables before investigating other influences.

³⁴ McAllister, Ian, ‘Housing Tenure and Party Choice in Australia, Britain and the United States’ (Canberra: Australian National University Research School of Social Sciences mimeo, 1983).Google Scholar

³⁵ Heath, Anthony, Social Mobility (London: Fontana, 1981), Chap. 3Google Scholar. Successive studies of voting choice do not provide an ideal means of measuring social mobility because only in the re-interviewing of the same people who are asked anew about their social characteristics does mobility data arise. However, attempts at reinterviewing the same people will fail to find many of those who have moved house. The young and the old will also be underrepresented if the waves are fielded several years apart. Nevertheless, our panels of voters can tell us about any trends in the extent of social mobility, and analysis of this data confirms that while the trend in occupational mobility was upwards during the decade, with mobility per annum between working-class and middle-class occupations in 1970–74 up by a quarter of its extent in 1963–66; at the same time the trends in housing tenure and union membership were downwards, with some 30 per cent less mobility per annum at the end of the period than at the start.

³⁶ Miller, , ‘Social Class and Party Choice in England’Google Scholar. See also Curtis, John and Steed, Michael, ‘Electoral Choice and the Production of Government’, British Journal of Political Science, XII (1982), 249–98.CrossRefGoogle Scholar

³⁷ Miller, , ‘Social Class and Party Choice in England’, p. 276Google Scholar. It may not be immediately clear how a variable measured at the constituency level can explain so much of the variance in votes cast when essentially the same variable measured at the individual level can explain little more than 10 per cent of the variance in voting choice (8 per cent in 1974). The problem arises because, at the individual level, someone either votes for a particular party or they do not. They cannot cast a proportion of a vote to match the probability associated with their occupation. On the other hand, in analysing constituency election returns this problem does not arise. At the constituency level, the Labour vote does not have to be (and indeed never is) 100 per cent. So readers should not be overly impressed by the high level of variance explained in the constituency analysis. What is important is the different direction of the trend.

³⁸ Miller, , ‘Social Class and Party Choice in England’, p. 283.Google Scholar

³⁹ Miller, , ‘Social Class and Party Choice in England’, p. 281.Google Scholar

⁴⁰ Franklin, , ‘The Rise of Issue Voting in British Elections’, pp. 18–19Google Scholar. Further analysis of the same data shows that the relationship between housing tenure and constituency complexion increased threefold between 1964 and 1974, from a correlation of 0·055 to 0·176; while the relationship between occupation and constituency complexion fell from 0·212 to 0·190. The small magnitude of these relationships reinforces my general feeling of scepticism for the value of inferences made from aggregate data (see below), and the direction of change reinforces my argument about the likely reason for Miller's findings.

⁴¹ Miller, , ‘Social Class and Party Choice in England’, p. 276.Google Scholar

⁴² Because of difficulties referred to earlier in defining a single causal model correct for all the elections held during our time period, the analyses reported in Figure 5 consist of straightforward predictions of voting choice from all six of our independent variables, including respondent's education. The coefficients for education and parents' party are not reported but remain fairly constant over the period, except for a notable drop in 1970 and revival thereafter.

⁴³ See Franklin, , ‘Demographic and Political Components’, pp. 208–12Google Scholar, for a discussion of the way in which swing can be expected to affect the estimation of class influences. Curtis, and Steed, , in ‘Electoral Choice and the Production of Government’Google Scholar, find another change which coincided with the rise of Miller's area effect: the demise of the ‘cube law’ relating seats to votes in British elections (p. 278), Such a change might indeed exaggerate a pre-existing relationship, again casting doubt on the substantive importance of Miller's area effect.

⁴⁴ Miller, , ‘Social Class and Party Choice in England’, p. 283.Google Scholar

⁴⁵ Miller, , ‘Social Class and Party Choice in England’, p. 283.Google Scholar

⁴⁶ And the decline was not associated with a reassertion of class voting, as already shown in Figure 1.

⁴⁷ Franklin, and Page, , ‘The Consumption Cleavage Approach’, passimGoogle Scholar. Indeed, the new party system could be based on the same class cleavages as the previous party system, in which case class voting would have returned in a new guise.