David Sanders, Jonathan Burton and Jack Kneeshaw, "Identifying the True Party Identifiers: A Question Wording Experiment," Party Politics, 8 (March, 2002), 193-205.

First Paragraph:
Since 1964, successive British Election Studies (BES) have measured party identification (or partisanship) using a modified version of the question that was originally devised for the American National Election Study. Although minor modifications have been made to the wording, we now have a lengthy time-series of measures of 'very strong', 'fairly strong' and 'not very strong' party identifiers covering every general election since 1964. The need for 'continuity' in survey measures over time argues for maintaining the current form of the question. We argue here that the evidence is now overwhelming that the current question wording produces a flawed measure of partisanship. Accordingly, future election studies should aim to provide an accurate measure of partisanship, even if this involves 'breaking the time-series'. Part 1 of the paper reviews the now considerable body of evidence that the existing BES question battery does not adequately capture voters' 'real' party identifications. Part 2 proposes an alternative measure of partisanship based on changes in the way the party identification question is asked and reports the results of a 12-month survey experiment conducted during 1998--9. Part 3 explores some of the problems of model mis-specification that arise from using the 'traditional' measure of partisanship and, by comparing simple models of vote choice using both the 'traditional' and 'alternative' measures, suggests that the continued use of the 'traditional' measure of partisanship may cause us to incorrectly estimate models of vote choice.

Figures and Tables:
Table 1. Measured levels of party identification in Britain, 1964-97, using the 'traditional' question sequence
Table 2. Dynamic measures of partisanship derived from the British Household Panel Survey, 1991-7
Table 3. Results of 12 split-sample surveys of British voters, November 1998 to October 1999
Table 4. Models of Conservative and Labour voting intentions, using 'traditional' and 'revised' measures of party identification; pooled monthly survey data, November 1998 to October 1999

Last Paragraph:
Party identification has long been both pivotal and contentious as a variable that contributes to the explanation of voting in British elections. It has been pivotal because there are -- still -- clearly many voters who think of themselves as long-term party supporters and who remain loyal to the same party across different elections. It has been contentious because there have always been general doubts about how partisanship should be measured and specific concern that the 'traditional' BES measure is so encompassing that it is tautologically related to vote. We have shown here that, if survey respondents are given a clear opportunity to indicate that they do not think of themselves as long-term supporters of a particular party, then the measured level of partisanship is significantly lower than that obtained using the traditional BES approach. Indeed, we estimate on this basis that little more than half of the UK electorate can realistically be regarded as genuine partisans. We also recognize that, if more dynamic, panel-based measures are employed, then this figure falls even further. However, given that it is still necessary to measure partisanship in cross-sectional surveys, the sort of approach to partisanship measurement we advocate here has important lessons for the way that party identification is measured in future British Election Studies.