Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

13
Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995 GEOFFREY EVANS 1 Nueld College, Oxford Introduction The years after the 1992 General Election saw a dramatic decline in support for the Conservative Party. Opinion polls charted a drop from the 43% obtained in the April 1992 Ballot to little more than 20% a year later. This precipitous fall levelled out, but by spring 1995 still only around a fifth of the electorate with a party preference intended to vote Conservative if there was a general election. The reasons for this change in fortune engendered much journalistic and academic speculation. They have also been examined empiric- ally through studies of the aggregate relationship between government popular- ity, trends in the economy and various political events since 1992. Very little is known, however, about the characteristics and motives of individuals who deserted the Conservatives during these years. To examine questions pertaining to the changing behaviour of individual voters we have to eschew the aggregate data favoured by economic modellers and explore instead the strengths of survey-based panel analysis in which far more detailed information on individuals’ social characteristics, political attitudes and behaviour can be examined. The aims of the analysis are two-fold. The first is primarily descriptive: who were these departing Tories and what might account for their defection? The second is more theoretical in focus. By examining the characteristics and perceptions of voters who switched from the Conservatives and comparing them with those who remained loyal, inferences can be drawn about the plausibility of certain theories of voting behaviour and the conditions under which they apply. In particular, the aim is to evaluate the plausibility of models of subjective economic voting advanced by David Sanders and his colleagues, 2 #Political Studies Association 1999. Published by Blackwell Publishers, 108 Cowley Road, Oxford OX4 1JF, UK and 350 Main Street, Malden, MA 02148, USA. Political Studies (1999), XLVII, 139–151 1 The British Election Panel Study data used in this analysis were collected under the auspices of the Centre for Research into Elections and Social Trends (CREST). CREST is an ESRC designated research centre (Award no.M 303 253 001) linking Nueld College and Social & Community Planning Research directed by Anthony Heath and Roger Jowell and co-funded by the Sainsbury Charitable Trusts. An earlier version of this paper was presented at the annual conference of the PSA specialist group on Elections, Public Opinion and Parties, Sheeld, September 1996. 2 See, among others, D. Sanders, ‘Government popularity and the next general election’, Political Quarterly, 62 (1991), 235–61; D. Sanders, ‘Economic performance, management competence and the outcome of the next general election’, Political Studies, 44 (1996), 203–31; S. Price and D. Sanders, ‘Economic expectations and voting intentions in the UK, 1979–87: a pooled

Transcript of Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

Page 1: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

Economics and Politics Revisited:Exploring the Decline in Conservative

Support, 1992±1995

GEOFFREY EVANS1

Nu�eld College, Oxford

Introduction

The years after the 1992 General Election saw a dramatic decline in supportfor the Conservative Party. Opinion polls charted a drop from the 43%obtained in the April 1992 Ballot to little more than 20% a year later. Thisprecipitous fall levelled out, but by spring 1995 still only around a ®fth of theelectorate with a party preference intended to vote Conservative if there was ageneral election. The reasons for this change in fortune engendered muchjournalistic and academic speculation. They have also been examined empiric-ally through studies of the aggregate relationship between government popular-ity, trends in the economy and various political events since 1992. Very littleis known, however, about the characteristics and motives of individualswho deserted the Conservatives during these years. To examine questionspertaining to the changing behaviour of individual voters we have to eschew theaggregate data favoured by economic modellers and explore instead thestrengths of survey-based panel analysis in which far more detailed informationon individuals' social characteristics, political attitudes and behaviour can beexamined.

The aims of the analysis are two-fold. The ®rst is primarily descriptive: whowere these departing Tories and what might account for their defection? Thesecond is more theoretical in focus. By examining the characteristics andperceptions of voters who switched from the Conservatives and comparingthem with those who remained loyal, inferences can be drawn about theplausibility of certain theories of voting behaviour and the conditions underwhich they apply. In particular, the aim is to evaluate the plausibility of modelsof subjective economic voting advanced by David Sanders and his colleagues,2

#Political Studies Association 1999. Published by Blackwell Publishers, 108 Cowley Road, Oxford OX4 1JF, UK and 350 MainStreet, Malden, MA 02148, USA.

Political Studies (1999), XLVII, 139±151

1 The British Election Panel Study data used in this analysis were collected under the auspices ofthe Centre for Research into Elections and Social Trends (CREST). CREST is an ESRC designatedresearch centre (Award no.M 303 253 001) linking Nu�eld College and Social & CommunityPlanning Research directed by Anthony Heath and Roger Jowell and co-funded by the SainsburyCharitable Trusts. An earlier version of this paper was presented at the annual conference of thePSA specialist group on Elections, Public Opinion and Parties, She�eld, September 1996.

2 See, among others, D. Sanders, `Government popularity and the next general election', PoliticalQuarterly, 62 (1991), 235±61; D. Sanders, `Economic performance, management competence andthe outcome of the next general election', Political Studies, 44 (1996), 203±31; S. Price andD. Sanders, `Economic expectations and voting intentions in the UK, 1979±87: a pooled

Page 2: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

who argue that either through personal economic expectations or assessmentsof economic management competence `(v)oters' economic perceptions continueto be central to the political fortunes of the government'.3

These questions are addressed by examining vote switching between 1992 and1995 among people who voted Conservative in the 1992 election. Switching ismodelled as conditional upon several types of respondent characteristics:(i) perceptions of personal and national economic fortunes; (ii) perceptions ofthe government's record; and (iii) party images. The data for these analyses aretaken primarily from the 1992 and spring 1995 waves of the British ElectionPanel Study (BEPS) conducted by the Centre for Research into Elections andSocial Trends (CREST).4 These data allow us not only to evaluate a wide rangeof possible explanations of Conservative decline but also to compare theirpredictive power in models that are far more highly speci®ed than modelsestimated in aggregate analyses of economic voting.

Analysis

The pattern of defection from the Conservatives in the BEPS panels is consist-ent with that observed in aggregate poll data. Defections took place relativelysoon after the 1992 election ± in the 1993 survey only 63% of those whoreported voting Conservative in 1992 still intended to do so ± with the trendbecoming rather more shallow over time so that by 1995 just 49% of the initialpool of Conservative voters retained a Conservative voting intention. Voteswitching away from the Conservatives appears to have been a fairly decisiveact, in that it was not associated with a marked tendency to return to the fold inensuing rounds of the survey. This one-step nature of defection from theConservatives means that vote switching between 1992 and any of the followingwaves can be treated as equivalent.5

The analysis models the probability of having switched from a Conservativevote to any other response (which includes supporters of all other parties as wellas abstainers and a small number of `don't knows'). As the dependent variable isdichotomous (Conservative voters in 1992 who intended to vote Conservative in1995 � 0; Conservative voters in 1992 who did not intend to vote Conservativein 1995 � 1), we use a logistic model, which is likely to provide more e�cientestimates than would OLS regression. Defection from the Conservatives is

cross-sectional analysis', Political Studies, 43 (1995), 451±71; D. Sanders and S. Price, `PartySupport and Economic Perceptions in the UK 1979±87: a Two-level Approach', in C. Rallings,D. M. Farrell, D. Denver and D. Broughton (eds), the British Elections and Parties Yearbook, 1994(London, Frank Cass, 1995). See also H. Clarke, M. C. Stewart and P. Whiteley, `Tory trends: Partyidenti®cation and the dynamics of Conservative support since 1992', British Journal of PoliticalScience, 27 (1997), 299±319.

3 Sanders `Economic performance, management competence and the outcome of the next generalelection', p. 223; see also Clarke et al., `Tory trends', p. 3.

4 The respondents in the ®rst wave of the BEPS were those obtained for the 1992 British ElectionStudy cross-section survey, the sample size for which (after re-weighting for a Scottish over-sample)was 2855. By the spring 1995 wave the number of respondents remaining in the panel had droppedto 1625. To correct for di�erential attrition, responses in all waves of the BEPS have been weightedby the distribution of votes in the 1992 sample. See B. Taylor, A. Heath and P. Lynn `The BritishElection panel study 1992±5; response characteristics and attrition', CREST Working Paper no. 40(London, SCPR & Nu�eld College 1996).

5 Where possible, the models shown below have been replicated across other waves of the study.

140 Research Note

#Political Studies Association, 1999

Page 3: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

conditioned on, in turn, economic experiences and perceptions, non-economicperformance and party images.6

Economic Experiences

In journalistic as well as academic commentary emphasis has been put on eco-nomic uncertainty associated with job insecurity and negative equity as a causeof the decline in Conservative support and also of the failure of a `feel-goodfactor' to emerge in more recent times.7 One set of pro�ered reasons concerns thepresence of what Sanders has referred to as the `new climate of post-Thatcheriteeconomic insecurity'.8 Despite national aggregate economic improvements,many people still feel vulnerable: job insecurity; negative equity; and uncert-ainty about the future all serve to prevent the return of feel-good and in con-sequence undermine the predictive power of models derived from economicindicators.

We test these claims by examining the role of respondents' economic circum-stances and experiences between 1992 and 1995 in accounting for patterns ofswitching. The 1995 wave of the BEPS contained several measures of retro-spective accounts of personal economic fortunes since 1992: experiences ofunemployment, redundancy and promotion (or lack of) are all likely to beconsiderations which bear upon respondents' and their families' feelings ofeconomic insecurity. Changes in circumstances between 1992 and 1995 canalso be calculated using the answers given by respondents to questions onhousehold income, unemployment and reported di�culty in paying mortgagesin each of the two surveys and calculating change over the period. We cantherefore assess if both retrospective accounts and measures of actual change inpersonal economic circumstances are associated with switching away from theConservatives.

Table 1 presents models predicting vote switching as a consequence ofindicators of personal economic experience between 1992 and 1995. It showsquite clearly that respondents' economic experiences do not account for theirvoting changes.9 In part, this non-e�ect will re¯ect the relatively small numberof respondents who report moving into or out of unemployment, for example,during this period. But even measures that do have considerable variance overthe period, such as that for household income, show no link to changes inpolitical preferences. Only the retrospective measure of whether respondents oranyone among their family had been unemployed for three months or more hasa signi®cant (if not particularly substantial) coe�cient. This ®nding must betreated with caution, however, as respondents were not asked exactly whenover the last few years they had experienced the events referred to in thesequestions ± the experiences may thus have occurred after any change of voting

6 There are also, of course, in¯uential models of voting behaviour that focus on voters' socio-logical characteristics. These are not well-suited, however, to explaining short-term political changeof the kind examined here. Analyses of vote switching which model the e�ects of social anddemographic characteristics ®nd little evidence of any pronounced tendency to defect di�erentiallyacross groups (details available on request).

7 See, for example, W. Hutton, The State of the Nation (London, Vintage, 1995).8 Sanders `Economic performance, management competence and the outcome of the next general

election', p. 223.9 Checks on the correlation matrix of the independent variables show that the lack of signi®cant

e�ects is not a result of multicollinearity (all inter-correlations are less than r � 0.30).

Research Note 141

#Political Studies Association, 1999

Page 4: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

allegiance. So even these rather minor e�ects of economic experiences are likelyto be overestimated.

The failure to account for vote switching using detailed measures of personaleconomic experience mirrors the lack of congruence between aggregate eco-nomic trends (de®ned in terms of in¯ation, interest rates and unemploymentrates) which generally tended to be stable or even positive during the 1992±95period, and the slump in Tory support over the same period. Objective eco-nomic trends provide no basis during 1992±95 for expecting a decline inConservative voting. Like others before us therefore we must turn away fromrelatively objective economic indicators to more subjective measures.

Economic Perceptions

Models of subjective economic voting di�er in the importance they attribute tothe types of economic perceptions held by voters. One distinction concerns thatbetween personal economic experiences and expectations versus perceptions ofthe state of the national economy ± so called, `pocket-book' and `sociotropic'models of voting.10 Another concerns the division between retrospective andprospective models.11 Sanders, in particular, has argued for the primacy ofperceptions of prospective personal (i.e. pocket-book) economic pay-o�s.Table 2 shows the pattern of responses in the BEPS to four standard indicatorsof economic perceptions which assess respondents' evaluations of change over

TABLE 1. Changing Economic Circumstances and Conservative Support. LogisticRegression of Probability of Defection from Conservatives

Economic experiences 1992±95

Parameter estimates

(standard errors)

(i) Measures of change calculated from information obtained

in 1992 and 1995Income change 1992±95 0.01 (0.03)Employment/unemployment 1992±95 0.30 (0.48)

Di�culty coping with mortgage in 1995 compared with 1992 0.25 (0.21)

(ii) Retrospective accounts obtained in 1995Self/family been made redundant since 1992 election 0.06 (0.37)Self/family experienced 3 months or more unemploymentsince election

0.57* (0.23)

Been promoted/got a better job since election ÿ0.01 (0.06)

Percentage of cases accurately classi®ed 55.9Improvement in ®t 9.07, 7 df

Weighted N � 644.

*Signi®cant at p5 0.05; **Signi®cant at p5 0.01.Coe�cient for missing income dummy variable not shown.

10 See D. Kinder and D. Kiewiet, `Sociotropic politics: the American case', British Journal ofPolitical Science, 11 (1981), 129±41.

11 M. Fiorina, Restrospective Voting in American National Elections (New Haven, Yale UniversityPress, 1981). H. Norpoth, `Presidents and the prospective voter', Journal of Politics, 58 (1996),776±92.

142 Research Note

#Political Studies Association, 1999

Page 5: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

the previous 12 months or the coming 12 months: (i) household prospective and(ii) retrospective economic evaluations and (iii) national prospective and (iv)retrospective economic evaluations, across the waves of the BEPS in which theywere obtained.12

Comparisons between these ®gures and aggregate trends are di�cult as theBEPS sample is not precisely representative of the population and thetrajectories of over-time responses to the various measures di�er according towhich one is being considered. There is also a problem with direct comparisonsof distributions on the questions asked in 1992±93 with those asked in 1994±95,because the latter present respondents with a response format that containsquali®ers distinguishing between a little stronger/weaker and a lot stronger/weaker, while the former does not. None the less, even a cursory examination ofthe relatively trendless patterns shown in Table 2 suggest that subjectiveappraisals cannot account for the dramatically changing levels of aggregateConservative support in the BEPS sample. However, they may still explain someof the switching from the Conservatives at the individual level.

To evaluate this possibility two approaches were adopted. The ®rst examinesthe possible contemporaneous e�ect of economic perceptions on vote switchingby modelling the e�ect of economic perceptions measured in 1995 on themeasure of vote switching between 1992 and 1995. This assesses the question ofwhether people who changed their allegiance had di�erent current perceptionsof the economy and their own personal ®nances than those who did not. Thesecond approach assesses whether changes in perceptions between 1992 and1995 are associated with changes in Conservative support. This is a moredemanding test of the e�ects of economic perceptions than the ®rst, as it testsfor the e�ect of change per se. It is not ideal, however, in that economicexpectations ± which have played a key role in Sanders' models ± were notmeasured in the 1992 survey and the e�ects of changes in them cannot beassessed. A combination of information obtained from both analyses istherefore likely to be useful.

TABLE 2. Economic Perceptions, 1992±95

1992 1993 1994 1995

Household income has gone up more than( fallen behind) prices over last year (%)

13 (44) 5 (60) 12 (49) 12 (49)

Household income will go up more than

( fall behind) prices in next year (%)

± 4 (60) 10 (46) 10 (45)

Economy has got stronger (weaker) in past year (%) 13 (55) 12 (57) 29 (33) 24 (33)Economy will get stronger (weaker) in nextyear (%) ± 31 (25) 35 (13) 24 (17)

Weighted N � 1654.

Note: these questions were asked with 3 response options in 1992/93; in 1994/95 theywere asked with ®ve response options.

12 (i) `looking back over the last year or so, would you say that your household's income hasfallen behind prices, kept up with prices, or gone up by more than prices?'; (ii) `looking forward tothe year ahead, do you expect your household's income to fall behind prices, keep up with prices, orgo up by more than prices?'; (iii) `looking back over the last year or so, would you say that Britain'seconomy has got stronger, got weaker or stayed about the same?'; (iv) looking forward to the yearahead, do you think Britain's economy will get stronger, get weaker or stay about the same?'.

Research Note 143

#Political Studies Association, 1999

Page 6: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

Table 3 presents a series of logistic models, the ®rst two of which containmore or less extensive measures of economic perceptions in 1995. Model 1contains the four measures of economic perceptions shown in Table 3. Themodel indicates that there are signi®cant e�ects of perceptions of the economyon changes in voting intention. The largest of these e�ects is found for theprospective sociotropic measure ± the indicator of prospective personalexpectations has no signi®cant e�ect.

Model 2 adds two further measures to those in model 1: these enquire as towhether (i) respondents's `own', or (ii) `the general' standard of living hasimproved or declined `since the last general election in June 1992'. Theseadditional measures do not change the conclusions drawn to any great degree,but they do further emphasize the predominance of the sociotropic measures ±all three of which are signi®cant predictors (@p5 0.05) of vote switching,whereas among the `pocket-book' perceptions only respondents' households'economic situation over the last year, and then only in model 1, is a signi®cantand not particularly powerful predictor of vote switching.

As with the retrospective measures of personal experiences modelled inTable 1, however, the grounds provided by these analyses for causal inference ±even for the sociotropic measures ± are weak. Despite evidence of an empiricalconnection between economic assessments elicited in 1995 and defection fromConservative support between 1992 and 1995, it remains quite possible thateconomic perceptions could have followed vote switching rather than precededor accompanied it, or that the respondents who were most likely to defect werethose who already had more sceptical views about the economy in 1992. Thuschanging economic perceptions over the 1992±95 period may not themselveshave been of consequence for Conservative Party popularity.

TABLE 3. Economic Perceptions and Defection from the Conservatives

Logistic regression parameter estimates (standard errors)

Model 1(Retrospective

e�ects)

Model 2(Retrospective

e�ects)

Model 3(Changee�ects)

Model 4(Retrospective

e�ects)

Economy past 0.29** (0.10) 0.21* (0.10) 0.17* (0.08) 0.36** (0.10)Household past 0.28** (0.11) 0.22 (0.12) 0.06 (0.10) 0.25* (0.11)

Economy future 0.57** (0.14) 0.49** (0.14)Household future 0.12 (0.13) 0.11 (0.13)Economy since 1992election

0.47** (0.12) 0.12 (0.08) 0.52** (0.11)

Household since 1992election

0.01 (0.12) 0.04 (0.08) 0.01 (0.12)

Percentage of cases

accurately classi®ed

63.7 64.9 56.9 62.9

Improvement ®t 64.6, 4 df** 82.9, 6 df** 10.9, 4 df* 69.0, 4 df**

Weighted N � 641.High scores � bad economic conditions. *Signi®cant at p5 0.05; **Signi®cant at

p5 0.01.

144 Research Note

#Political Studies Association, 1999

Page 7: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

The results of models 3 and 4 in Table 3 suggest that this later interpretationmay well be accurate. With the more demanding test of causal impact speci®edin model 3 ± which examines whether change in economic perceptions is relatedto change in vote preference ± there is little evidence of economic e�ects on voteswitching. The omission of the two economic expectations measures from thismodel makes direct comparison with models 1 or 2 di�cult, but if we comparethe e�ects of the four measures of change in retrospective evaluations with thoseof the same four perceptions measured in 1995 (shown in model 4), we can seethat they are very weak. Such economic evaluations may `predict' past voteswitching when measured at the point of outcome, but they do not appear to co-vary to any marked degree with vote switching over time.

There are, moreover, other reasons for not taking even the limited e�ects ofthe economic perceptions measures at face value. In particular, it would appearthat individuals' economic perceptions are not very stable over time. Anestablished literature in political science treats unstable responses as indicatingthe failure of many respondents to have views on the topic under considera-tion ± an argument formulated most in¯uentially in Philip Converse's cele-brated non-attitudes thesis.13 Yet the over time correlation between measures ofeconomic perceptions taken in 1992/3 and 1995 is only 0.2/0.3.14 This applies toboth personal and national economic situations. Thus although personalcircumstances among the same may well vary in ways that are likely to reduceover time correlations in perceptions of personal economic situations withoutnecessarily undermining our con®dence that these responses are valid indicatorsof such situations, perceptions of national economic conditions should be morestable ± after all, everyone lives in the same society. The observed lack of suchstability suggests that there is little in the way of a shared evaluation of the stateof the economy over time. Some respondents' expectations go up at the sametime as others' expectations go down. From a `Conversian perspective', thiswould imply that these perceptions are not ®rmly held beliefs, they display thecharacteristics of `non-attitudes'. Political preferences, by comparison, are farmore stable: even when measured three years apart, attitudes towards theConservative Party correlate at 0.67. This stability indicates that these attitudesare to some degree shared and presumably meaningful to respondents. It is hardto see how systematic changes in such well-formed attitudes could be explainedby the ¯uctuating responses to questions about topics, such as the economy, onwhich respondents display evidence of possessing `non-attitudes'.15

13 P. Converse, `The nature of belief systems in mass publics', in D. Apter (ed.), Ideology andDiscontent (New York, Free, 1964).

14 The correlation over the period 1992±95 for both household and national retrospectiveperceptions is only 0.20. Correlations for period 1993±95 for household and national prospectiveperceptions are 0.30 and 0.33 respectively.

15 There is also only a very weak association between subjective economic perceptions and theobjective economic variables measured in the BEPS surveys. For national economic expectationsthis is not particularly concerning, our measures of respondents' objective economic attributes areconcerned primarily with personal experiences, but even personal economic expectations andretrospective evaluations are di�cult to model using objective indicators ± a conclusion also drawnby Sanders and Price from their analysis of Gallup data; Sanders and Price `Party support andeconomic perceptions in the UK 1979±87', p. 59. But if economic perceptions are genuinelyappraisals of economic conditions ± as is assumed by advocates of such models of governmentpopularity ± then they should have a fairly solid basis in these conditions. That does not appear tobe the case (details available on request).

Research Note 145

#Political Studies Association, 1999

Page 8: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

Shifting the Focus from the Economy to Other Considerations:(i) the Government's Record

Given the evident limitations of measures of economic perceptions and theirrather weak link to vote switching, it is di�cult to attach much weight to thenotion that they drive government popularity. In the period since 1992, evenSanders has shifted emphasis to considerations of economic managementcompetence ± which are likely to involve political judgements as much as eco-nomic ones. However, we see no reason to restrict considerations of governmentcompetence to just questions concerning the economy and voters' standard ofliving. There are a range of issues, re¯ecting the pressing questions of the day,which also involve assessments of government competence: unemployment, thehealth service, education, and law and order commonly receive top billing inpublic assessments of the `urgent problems facing the country at the presenttime' and are thus likely candidates for public appraisal.16 Thus a more appro-priate model might be one that views popular appraisals of the government'scompetence as crucial to its popularity ± but does not restrict such assessmentsof competence to those concerning the performance of the economy per se.

In the BEPS surveys we have appraisals of several aspects of the government'srecord that are asked over time. These questions enquire about respondents'views regarding the government's record `since the last general election in June1992' on a range of topics. Two of these, `respondent standard of living' and `thegeneral standard of living' have been used in the analyses presented in Table 4.They represent appraisals of the government's record which focus explicitly onthe economy and its impact on standard of living. Four others, however, haveeither a less explicit connection with the management of the economy, or nodirect connection. These items cover the areas of unemployment, taxation,education and the National Health Service. The ®rst two are clearly not inde-pendent of concern about the economy, but they measure rather di�erentaspects than straightforward concern with standard of living ± the usual motivethat economic evaluations are assumed to tap. The last two are not economicevaluations, at least as usually de®ned.17 We can therefore compare the relativeimpact of appraisal of these di�erent areas of competence over the 1992±95period on vote switching during that period.

Table 4 follows Table 3 in presenting the results of two di�erent modellingstrategies: ®rst, a logistic regression analysis that models vote switching as afunction of post hoc appraisals taken in 1995 of the government record; andsecond, a model which presents the e�ects of changes between 1992 and 1995 inevaluations of the government's record.18

Considering ®rst the post hoc analysis, model 1 simply assesses the e�ects ofappraisals of the government's record on unemployment; taxation; education;the National Health Service; respondents' standard of living; and the generalstandard of living in the period 1992±95, but without controlling for these

16 This wording and the topics listed are taken from the Gallup series on this theme.17 This distinction is supported by a principal components analysis which indicates that assess-

ments of the government's record on the NHS and education formed the core of the ®rst principalcomponent ± with unemployment also loading somewhat more weakly onto this dimension ± whilenational and personal standard of living loaded onto the second principal component and taxationloaded clearly onto a third (details available on request).

18 In the 1992 survey respondents were asked about their appraisals of changes since the generalelection of June 1987.

146 Research Note

#Political Studies Association, 1999

Page 9: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

perceptions in 1992. From this model it can be seen that the government'srecord on the NHS has a stronger e�ect on vote switching than do appraisals onissues relating to the economy. None the less, there is a signi®cant e�ect for thenational, economic evaluation measure.

When the e�ects of change over time ± shown in model 2 ± are examined,however, we see a rather di�erent pattern. As we might expect, given the resultsof the analyses reported above, prediction is weaker when change over time isthe test of impact on vote switching rather than simply retrospective appraisal.More importantly, however, we also ®nd that the measures of economicperceptions do not survive this more demanding test. None of their e�ects aresigni®cant@ p5 0.05; the only signi®cant e�ects at this level are for taxation,while the NHS (p � 0.05) and unemployment (p � 0.06) are marginallysigni®cant.19

The results of these over-time analyses indicate that models of competencewhich focus only on the economy are under-speci®ed. Unfortunately, the mainpublic opinion surveys do not obtain appraisals of government's performanceon issues such as the NHS often enough to make them suitable for aggregate-level time-series analysis, so they do not get included in the models favoured by

TABLE 4. Assessments of the Government's Record 1992±95 and Defection from theConservatives

Logistic regression parameter estimates

(standard errors)Model 1

(Retrospective e�ects)Model 2

(Change e�ects)

National standard of living since 1992 0.36** (0.12) 0.10 (0.08)Household standard of living since 1992 0.14 (0.11) 0.03 (0.08)

Unemployment since 1992 0.11 (0.07) 0.11# (0.06)Taxation since 1992 0.25* (0.10) 0.17** (0.07)NHS since 1992 0.45** (0.09) 0.13# (0.07)Education since 1992 0.08 (0.10) 0.06 (0.08)

Percentage of cases accurately classi®ed 66.3 56.1Improvement in ®t 89.3, 6 df** 21.2, 6 df**

Weighted N � 641.High scores � negative assessment of record, *Signi®cant at p5 0.05; #Significant at

p5 0.10.

19 When changes in retrospective economic evaluations over the previous 12 months are added tothis model they also fail to have signi®cant e�ects. Further analysis of the pattern of relationsbetween appraisals of various aspects of the government's record and attitudes towards theConservatives shows that the links between party support and aspects of the government's recordchanged over time. Thus a principal components analysis shows that in 1992 both national andpersonal standard of living loaded onto the ®rst principal component along with attitudes towardsthe Conservatives. By 1994, however, these two items loaded onto a distinct second dimension;while attitudes towards the Conservatives loaded onto the ®rst principal component along withassessments of the government's record on the NHS, education, and unemployment (detailsavailable).

Research Note 147

#Political Studies Association, 1999

Page 10: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

aggregate analysts. None the less, even the items measuring the government'srecord on taxation, health and unemployment add little to the predictive powerof the regression models ± most vote switching remains unaccounted for byappraisals of record. Explanations of the government's haemorrhage of supportduring this period must lie elsewhere.

(ii) The Government's Image

Finally, we evaluate explanations of vote switching which explicitly incorporateperceptions of the parties themselves. The aim here is to assess whether defec-tion from the Conservatives was attributable to political events that served toundermine the government's reputation with the public. Certainly, the eventsfollowing the 1992 election victory from the ERM ®asco onwards, through thecorruption claims and the well-publicized ®nancial killings made following theprivatization of utilities seem designed to undermine the idea of a united,competent government concerned to further the interests of society as a whole,rather than `feathering the nests' of the already advantaged.

The BEPS contains several established measures of how the government isperceived in terms of its image as a party. These pertain to how `extreme',`divided', or `capable of strong government' a party is thought to be, andwhether a party is considered to be `good for one class or good for all'. That thegovernment's image changed for the worse on all these counts over the period1992±95 is indisputable. Table 5 presents illustrative evidence.

These dramatic falls in estimation mirror patterns observed in Gallup andother polls. They also display some similarity with the pattern of decline insupport for the Conservatives over this period. Once again, however, beforeinferring any causal connection between such images and changes in votingpreferences, we must undertake the more demanding test of assessing whetherchange in party image varies with the tendency to vote-switch at the individuallevel. To do this we again regress vote switching onto measures of change in theindependent variables, in this case the measures of party image, over the period1992±95.

Table 6 shows that changes in the image of the government can account forits plummet into unpopularity more e�ectively than can measures of changingeconomic conditions and perceptions, or broader assessments of the govern-ment's record. These e�ects are robust: even when a wide range of competing

TABLE 5. The Changing Conservative Party Image. Percentage agreeing that theConservative Party is . . .

1992 1994 1995

Good for one class (%) 54 73 72Capable of strong government (%) 86 32 27

United (%) 70 10 8Extreme (%) 29 41 39

Weighted N � 1654.

148 Research Note

#Political Studies Association, 1999

Page 11: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

explanatory variables is modelled concurrently ± see column 3 ± only imagematters.20 None of the measures of the economy or the government's record aresigni®cantly related to defection from the Conservatives.

Closer examination of the parameters for the four measures of Conservativeimage reveals a constant pattern of e�ects across all three models. Surprisingly,perceptions of divisions within the Conservative and Labour parties are notrelated to vote switching.21 Many Conservative supporters who perceived the

TABLE 6. Changes in Support for the Conservatives Regressed onto Changes in PartyImages, Assessments of Record and Economic Perceptions, 1992±95

Logistic regression parameter estimates

(standard errors)Di�erence scores 1992±95 Model 1 Model 2 Model 3

Conservatives extreme 0.45** (0.10) 0.43** (0.11) 0.44** (0.11)Conservatives divided 0.10 (0.11) 0.16 (0.12) 0.15 (0.12)Conservatives good for one class 0.56** (0.09) 0.58** (0.10) 0.56** (0.10)

Conservatives (not) capable ofstrong govt.

0.83** (0.10) 0.90** (0.10) 0.87** (0.11)

Labour extreme ÿ0.05 (0.09) ÿ0.06 (0.10)Labour divided 0.03 (0.09) 0.05 (0.09)

Labour good for one class ÿ0.30** (0.09) ÿ0.31** (0.09)Labour (not) capable ofstrong govt.

ÿ0.36** (0.09) ÿ0.34** (0.09)

Economy in past year 0.03 (0.10)Household in past year 0.00 (0.12)Economy since election 0.03 (0.10)

Household since election 0.00 (0.10)Unemployment since election 0.07 (0.07)Tax since election 0.03 (0.08)NHS since election 0.03 (0.09)

Education since election 0.00 (0.10)Percentage classi®ed accurately 71.1 72.9 73.3Improvement in ®t 184.3, 4 df** 223.5, 8 df** 221.9, 16 df**

Weighted N � 633.High scores indicate negative assessments. *Signi®cant at p5 0.05; **Signi®cant atp5 0.01.

20 They are also una�ected when controls for long-term stability in Conservative partisanship, asindicated by whether respondents reported voting Conservative in 1987 or not, are added to thesemodels ± thus indicating that these e�ects are not merely a proxy for weaker prior partisanattachments.

21 Given the surprising nature of this ®nding, further checks were made on the characteristics ofthe logistic models and the variable used to measure changes in perceptions of party divisions.These checks showed that there is a weak zero-order e�ect on vote switching for changes inperceptions of division within the Conservative Party, which is removed when other image measuresare modelled concurrently. None the less, the level of correlation between the image items is not sohigh as to produce undesirably high standard errors, while variance in the measure of changes inperceptions of Conservative Party division is not so truncated as to prevent its e�ects beingobserved. Thus the conclusion that changes in perceptions of division within the party are notresponsible for defection from Conservative support stands.

Research Note 149

#Political Studies Association, 1999

Page 12: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

party as more divided in 1995 than in 1992 still remained loyal. This was less thecase, however, for those who perceived the party as more likely to be `good forone class' or not as capable of `strong government', or who perceived theLabour Party as having become more positive in both these respects. Over-timechanges in both of these perceptions for each of the two main parties havepowerful e�ects on vote switching, as does the changing perception of theextreme character of the Conservative Party (but not Labour).

Conclusions

This paper has tried to explain why people deserted the Conservatives in theyears following the 1992 election by using individual-level panel data to examinethe plausibility of economic models of government popularity. It ®nds little orno support for the claims of various commentators that personal economicproblems, economic expectations, or economic management competence under-mined Conservative support during this period. Other aspects of the govern-ment's record, such as the management of the NHS, are at least as e�ective, ifnot more so, at predicting vote switching, while measures of the government'simage ± particularly perceptions of the Conservatives' and Labour's capacityfor strong government and their representation of the interests of di�erent socialclasses, and whether or not the Conservatives were extreme ± had by far thelargest impact on the Conservatives' popularity.22 We can conclude, therefore,that the links between political factors (i.e. party perceptions) and changes invoting preferences are far stronger than those between economic perceptionsand such changes.

Doubts about the explanatory power of economic models are also engenderedby evidence of individual-level instability in voters' economic perceptions. Ifmany voters do not have well-formed perceptions of the economy, then theseperceptions are unlikely to have a causal impact on their party preferences ±which as we have seen are by comparison far more stable and, we might infer,well-formed. Our ®ndings in this respect are consistent with other evidence onpublic ignorance about the economy23 and with evidence that economic percep-tions are not anchored in economic conditions. But if economic perceptions arenot explained by the economy itself, what does account for them? One plausiblesuggestion is that they re¯ect partisanship rather than in¯uence it ± perhaps asindicators of generalized approval of incumbents ± and this itself accounts fortheir links, where observed, with government popularity. Such speculationsawait further micro- and macro-level analysis, although, as we have observedabove, Sanders and Price have themselves presented evidence that non-economicpolitical events have a substantial in¯uence on economic expectations.24

22 These ®ndings might appear inconsistent with those of aggregate-level analyses of the role ofeconomic perceptions in government popularity. However, if only a small minority of voters reactpolitically to their own, or the country's, economic circumstances while the rest contribute randomerror we would ®nd an aggregate-level correlation between economic perceptions and governmentpopularity. Moreover, if party perceptions had been included in aggregate-level models ofgovernment popularity these analyses might also have found economic perceptions to be of minorsigni®cance.

23 See T. Holbrook and J. C. Garand, `Homo economus? Economic information and economicvoting', Political Research Quarterly, 49 (1996), 351±75.

24 Sanders and Price, `Party support and economic perceptions in the UK 1979±87', p. 59.

150 Research Note

#Political Studies Association, 1999

Page 13: Economics and Politics Revisited: Exploring the Decline in Conservative Support, 1992–1995

It is also probable, of course, that the performance of the economy, the NHS,or whatever, can themselves in¯uence the public's perceptions of parties. This isespecially likely to be so with respect to the rather general aspects of partycompetence tapped by the `capable of strong government' item. However, evenwhen examined without controlling for measures of party image (see Table 4)the impact on vote switching of changes in assessments of the government'srecord is very weak, while changes in party image have a substantial associationwith vote switching even when other factors are included in the predictivemodels (see Table 6).

But prediction is not explanation: claims concerning the causal impact ofparty images can be criticized on the basis that the relation between images andvoting intention is not causal; that images are instead merely the fellow-travellers of changing vote preferences and, therefore, that the use of partyimage measures to explain party support is tautological. Yet without any otherobvious candidates to account for vote switching and with a very salient set ofexogenous political events occurring during the period under considerationwhich themselves make changes in party image both likely and a reasonablebasis for re-evaluating party preference, it is hard to see on what grounds otheraccounts might be preferred. So for the time being at least, the economy wouldnot appear to be the central electoral issue in British politics. Whether or not ithas been securely supplanted in this role by party image remains to be seen. In arelatively healthy economic situation, the economy may not matter as much asother factors. If that situation should change, however, concern about theeconomy might be expected to make a re-appearance as a basis of voters'political choices.

(Accepted: 21 February 1998)

Research Note 151

#Political Studies Association, 1999