¹ See, for example, Rose, Richard, ‘Britain: Simple Abstractions and Complex Realities’ in Rose, Richard, ed., Electoral Behaviour: A Comparative Handbook (New York: Free Press, 1974)Google Scholar, or Franklin, Mark and Mughan, Anthony, ‘The Decline of Class Voting in Britain’, American Political Science Review, LXXII (1978), 523–34.CrossRefGoogle Scholar

² Dunleavy, Patrick, ‘The Political Implications of Sectoral Cleavages and the Growth of State Employment’, Political Studies, XXVIII (1980), 364–83 and 527–49.CrossRefGoogle Scholar

³ Support for Labour in the public sector is especially strong in social grade B (lower managerial and professional) and within that group is strongest among employees other than managers.

⁴ Dunleavy, , ‘The Political Implications of Sectoral Cleavages’, pp. 536, 538, 542, 547.Google Scholar

⁵ This account nevertheless appeared attractive, as the combination of growth in public employment (see Semple, Matt, ‘Employment in the Public and Private Sectors’, Economic Trends, No. 313 (1979), 90–108Google Scholar) and the relatively middle-class character of public employment (the public sector has 33 per cent of its jobs in social grade categories A, B, and C1a compared with 23 per cent in the private sector) suggested the possibility of disproportionate upward mobility. Of course, one could tack sectoral self-interest onto this account to the extent that, once mobility was controlled for, public sector employees remained more likely than their private sector counterparts to vote Labour.

⁶ In view of Dunleavy's conclusions about the interrelationship of class, sector, unionization and political alignment, the tests should be made on the frequencies given in the last two columns of Table 1, which include unionization. Quite fortuitously, it turns out that union membership does not affect the central finding we discuss, which for ease of communication we present first in the context of the simpler four-way table.

⁷ Dividing the odds by the number equal to one more than the odds gives the proportion. For example, the first odds of 3·83 correspond to a proportion of 3·83/(3·83 + 1) or 3·83/4·83, which is equal to 0·79 or 79 per cent Conservative.

⁸ The models and methods are discussed in Upton, Graham, The Analysis of Cross-tabulated Data (London: Wiley, 1978)Google Scholar, and are those used by Dunleavy. We use his notation where possible to promote comparability of results.

⁹ If the association between A and B varies by C, it will also be the case that the association between A and C varies by B, and that of B and C will vary between categories of A. This leads to the necessity of parallel interpretations of high-order interactions, as discussed by Davis, James in ‘Contingency Table Analysis: Proportions and Flow Graphs’, in Alt, James, ed., Advances in Quantitative Analysis (Amsterdam: Elsevier, 1980).Google Scholar

¹⁰ An accessible example of these procedures appears in Kuklinski, James H. and West, Darrell M., ‘Economic Expectations and Voting Behaviour in United States House and Senate Elections’, American Political Science Review, LXXV (1981), 436–47.CrossRefGoogle Scholar

¹¹ Note that since the model is hierarchical, all the two-way associations in A F S are also included as parameters in the model, even though the F S association does not actually make any extra contribution. Its lack of effect confirms that background class has no effect on sectoral location other than that picked up by the A F S interaction (that is, contingent on political alignment). If both social-class variables were split between classes C1b and C2, Model 4 would again emerge as the final model, although the overall fit would be slightly worse (Y² = 7·2, 4 d.f.). There is no single term which could be added to this model, as in fact the extra Y ² is mostly associated with the A C F S interaction, reflecting a more idiosyncratic pattern of variation in the cell frequencies.

¹² Note that the C S association is still included as a component of the C S U term. The inclusion of these extra terms does not change the signs or magnitudes of the parameter estimates with which we are directly concerned.

¹³ A lot of research has turned up little else relevant to the relationship between sectoral location and political alignment. For details of relationships between sectoral location and other demographic characteristics, see Turner, Janet, Public Sector/Private Sector: A New Cleavage? (University of Essex M.A. dissertation, 1980).Google Scholar

¹⁴ If working-class occupations are less differentiated between sectors, or if those from working-class backgrounds are more likely to spend their lives alternating between sectors, then our contemporaneous measurements for these people would contain greater measurement error, obscuring any sectoral effects which might exist.

¹⁵ Details may be found in Turner, , Public Sector/Private SectorGoogle Scholar, Appendix B and in Dunleavy, , ‘The Political Implications of Sectoral Cleavages’, p. 537.Google Scholar

¹⁶ See, for example, Lehner, Franz, ‘Cognitive Structure, Uncertainty, and the Rationality of Political Action: A Synthesis of Economic and Psychological Perspectives’, European Journal of Political Research, 111 (1975), 275–91.CrossRefGoogle Scholar Lehner offers no empirical evidence but suggests that rational accounts of voting must be restricted to those having the appropriate ‘differentiated politics-related cognitive structures’. For the general argument that socialization experiences produce general orientations to politics as well as more specific partisan dispositions, see for instance Dawson, Richard, Prewitt, Kenneth and Dawson, Karen, Political Socialization (New York: Little, Brown, 1978).Google Scholar