Malcolm Brynin and David Sanders, "Party Identification, Political Preferences and Material Conditions: Evidence from the British Household Panel Survey, 1991-2, "Party Politics, 3 (January 1997), 53-77.

First Paragraph:
The sources of voting preferences in British general elections have been extensively analysed. Following the model established by the 'Michigan School' in the 1950s, analysts of UK election surveys have tended to place considerable stress on the explanatory role of partisan identifications . These relatively stable and enduring affective attachments to one or other of the major parties are held to predispose the electors thus affected to vote for the party with which they 'identify'. The empirical basis for the strong theoretical emphasis accorded to party identification in explanation of democratic voting behaviour derives in large part from the evidence garnered from the panel survey components of the US and British Election Studies. These have repeatedly shown that most identifiers acquire their party identification during their early formative years. Moreover, although there is always a limited amount of 'turnover' in reported identification, those individuals who retain their identification over time also tend to vote in same way in successive elections. As a result, party identification is usually accorded a pivotal causal role in the determination of voting preferences. In addition, because party identification acts as a filter for the interpretation of political and economic developments, it is also held to influence, inter alia , electors' ideological positions, policy preferences and economic perceptions (which themselves affect voting behaviour).

Figures and Tables:
Figure 1: Conservative support (percentage vote intention) and percentage identification with the Conservative, general elections, 1964-92.
Figure 2: Labour support (percentage vote or vote intention) and percentage identification with Labour, general elections 1964-1992.
Figure 3: Liberal (Democrat) support (percentage vote or vote intention) and percentage identification with the Liberals, general elections, 1964-1992.
Figure 4: Conservative Popularity (Percent Intending to Vote Conservative) and Percent Identifying with Conservative Party, 1992-1994.
Figure 5: Labour popularity (percentage intending to vote Labour) and percentage identifying with Labour Party, 1992-1994.
Figure 6: Liberal Democrat popularity (percentage intending to vote Liberal Democrat) and percentage identifying with the Liberal Democrats, 1992-1994.
Figure 7: Schematic outline of a typical party identification model of voting choice.
Figure 8: Schematic representation of a model used to conduct convergent validation tests of vote choice vs party identification.
Table 1: Logit models of Conservative vote (1992) and Conservative party identification (1991 and 1992).
Table 2: Logit models of Labour vote (1992) and Labour party identification (1991 and 1992).
Table 3: Logit models of Conservative vote and Labour vote (1992), including identifications terms as predictor variable.

Last Paragraph:
The simple conclusion implied by all of this is that models of the sort reported in Table 3, in which identification is employed as a predictor of vote choice, are woefully mis-specified. In these circumstances, analysts have three alternatives available to them: (1) abandon the concept of party identification as an 'explanation' of vote choice altogether; (2) seek to specify and estimate separate models of vote choice and of party identification which by definition exclude the other from the right-hand side of the specified equation; or (3) recognize that the current battery of party identification questions do not effectively measure respondents' 'enduring affective attachments' and contrive to establish new measures of 'real' party identification. Given that there are minor, and potentially interesting, variations in the equations for vote choice and identification shown in Tables 1 and 2, our preference would be for the second option.