EfmoraiStxdies
(1989), 8: 1,59-74
The Components of Voting Change in England 1983-1987 GRAHAM J. G. UITQN~~
Department of Mathematics, University of Essex, Colcbester, Essex CO4 3SQ,
England
The change in the constisueney
vote between successive genera1 efections is broken down into a number of components. It is shown that there are regional geographical variations of up to 7.5 per cent of the rhree-party vote, that by-elections can have a major impact upon the subsequent general election vote, that there is an incumbency effect of undoubted statistical significance which is worth on average between&00 and 500 votes to the incumbent (a potential swing of 800 to 1,000 votes on retirement). The
incumbency effect is found to be considerably greater for Labour incumbents than for Conservative incumbents. There is also a gender effect of probable statistical significance, which takes the form of an average loss of between between
125 and 250 votes by a female candidate 250 and 500 votes>.
(a potential
swing
of
This paper presents an analysis of the changes in the voting patterns of 512 English constituencies between the general elections of 1983 and 1987. ’ The aim of the analysis is to obtain reasonably accurate estimates of the various components of change, having made allowance for the effects of other components, Seven components are considered and these are set out in the equation below: Constituency
change = National change -I- Previous election effect + Geographical effect + Intervening by-election effect + Incumbency effect + Gender effect + Lo&d effect. (11
fn this equation the final component, ‘local effect’, effectively represents the error component of the model, since it amounts to that part of the constituency change which cannot be accounted for by the previous six components. The bcaf effect will be influenced by differential turnout, specific local issues, racial effects and changes in the tocal socio-economic conditions of the constituency. The existence af the first five components of equation (1) is, of course, not in doubt. The use of the national change figures as a realistic overall summary was a commonplace until recent times when urban/rural and North/South divides have become apparent. These latter effects are encapsulated here as the geographical effect. With the advent of a relatively strong third party (in terms of its share of the three-party vote), there has been much discussion of potential tactical voting which is here contained within the ‘previous election effect’. “I would like to thank fvor Crewe, comments and suggestions.
Stephanie
Stray and a referee for a number
0261-3794J89/01/0059-16/$03.00 0 1989 Butterworth Sr Co (Publishers) Ltd
of helpful
60
The Components
of Votmg
Change
in England
1983-1987
By-elections are frequently spectacular-for example Orpington in 1962-and often have a carry-over effect to the next general election, leading to a result at that election which could not have been predicted from the previous general election result. In addition to these major components of constituency change, equation (I) also includes two minor factors relating to the candidates who stand. It will be shown that there is an undoubted advantage in being the sitting MP for a constituency, and that this advantage appears greater for Labour MI’s than for Conservative MI%. There is also a possible disadvantage in fielding a female candidate. National Change Over England as a whole there was remarkably little change in the voting proportions for the three main parties between the general elections of 1983 and 1987. The Conservative percentage remained virtually unchanged, while the Labour percentage rose by about 2.9 per cent and the Alliance percentage fell by about 2.8 per cent. However, these overall slight changes mask considerable variations from constituency to constituency. For example, in Ealing North the Conservative share of the three-party vote rose by 11.2 per cent, whereas in Liverpool Garston it fell by 14.0 per cent. Plainly the model Constituency is quite inadequate. statistic S2, defined
To measure by
S2 = c (observed
change = National the extent
%-Fitted
change
of the inadequacy
%)2
i=l,
. . . ,512;
(2) we use the summary
j=1,2,3.
&j The ‘fitted %’ in this definition refers to the value predicted by the model being studied, and the summation is over all 512 constituencies and the three main parties, with all percentages relating to the three-party vote. The over-simplistic model (2) gives rise to a value of 19190 for S”, and this value will serve as a baseline by which the effectiveness of introducing the other components of change can be judged. Previous Election Effect The publication of the results of a general election provides the voters at the next general election with a guide to the prospects of the various parties in this next eiection. In the case of a marginal seat this may well result in would-be adherents of the third party voting tactically against their least preferred party. There is aiso evidence (Upton and Stray, 1986) that the political identifications of other voters in a constituency have an effect on an individual’s own political identification. If all around support party X then only the most committed individual will support party Y. These two arguments provide separate support for the expectation that voting changes will show differences which are dependent upon the voting breakdown at the earlier of the two general elections. The evidence that this does occur is provided by Fig. 1. Fig. 1 shows a summary picture of the changes actually observed between 1983 and 1987. The vertices of the triangle correspond to the cases where 100 per cent of the
GRAHAM
J. G. &TON
61
Alliance
tabour FIG.
Conservative 1.
Aggregate three-party voting changes between 1983 and 1987.
three-party vote are cast for a single party. Diagrams of this type have been advocated for use with political data by Upton (1976) and Miller (1977) amongst others, and feature prominently in the recent book by Aitchison (1986). In the Fig. the start of each arrow represents the voting configuration for a group of 3 or more constituencies which :;ad very similar voting breakdowns in 1983. The arrowhead represents the corresponding aggregate voting breakdown in 1987. If there were no previous election effects then each arrow would (apart from ‘random’ variation) be pointing in the same direction, and would have the same length. Since, overall, Labour took votes from the Alliance, with the Conservative vote roughly unaltered, the result would be a uniform array of short equal-length arrows all parallel to the upper left side of the triangle and pointing downwards. This is manifestly not the case. There are two major groups of arrows showing distinctly different properties. In the bottom left are long arrows pointing to the Labour vertex and representing large increases in the Labour vote in Labour-held constituencies. Of particular interest is the top left arrow which illustrates the collapse of the Conservative vote in Labour/Alliance 1983 marginal constituencies-a clear instance of apparent tactical voting. It is apparent then that the 1983 political position did have an effect on the 1983-87 inter-election change. The change that occurred in a constituency dominated by the Conservatives in 1983 would not be a good guide to the change that occurred in a constituency dominated by Labour in 1983. However, whilst arrows in different parts of the triangle are very different, the arrows in neighbouring parts of the triangle are very similar to one another, so that knowing what happened to a constituency having a 60 per cent Conservative vote in 1983 would be a useful guide to what happened to a constituency having a 58 per cent Conservative vote in 1983. In general terms it is clear
62
The Components
of Voting
Change
in England
198%1987
that the more similar are the 1983 political profiles of two constituencies, the more similar will be their 1987 political profiles. A measure of the difference between the 1983 political profiles of constituencies Y and s is provided by P,, = (Con, - Con,)’ + (Lab, - Lab,)’ + (All, - All,)‘,
(3)
where Con, is the percentage of the three-party vote in constituency Yin 1983, and the remaining terms are defined in a similar manner. If we want to predict the ‘previous election effect’ for constituency Y, using the observed changes from the remaining 511 constituencies, then we must pay most attention to the constituency m, say, for which P,, is the smallest of P,,,,,.,, Pr,s12 (not counting P,,,), and least to constituency Q, say, for which Pr,Q is the greatest of the P,, values. A suitable procedure is to assign to each of the constituencies a weight dependent upon the choice of T. Thus constituency s would be assigned the weight
w,,,= PT>“l(P,T”
+ . , + Pr.5*;U)
(4)
where the summation in the denominator excludes I’,,;“, and a is some positive constant whose value will be fixed later. Also excluded from the denominator are constituencies affected by by-elections. The reason for using the remaining 511 constituencies in order to estimate the ‘previous election effect’ for constituency r is simply that the results for constituency Y itself only provide us with information on the overall totaf constituency change. By comparing constituency Twith its potential neighbours we can deduce how atypical its behaviour is, and hence the total magnitude of the remaining four components of the right-hand side of equation (1). We are attempting to predict the change in vote between 1983 and 1987 for constituency Y. This is equivalent to predicting the 1987 vote, since 1987vote
= 1983 vote + change in vote 1983-87
(5)
A problem arises here since the 1987 vote cannot lie outside the range 0 per cent to 100 per cent, and our procedure must ensure that this is the case. For this reason we work with the logarithms of ratios of votes. As a preliminary estimate G,,tya, of the 1987 vote for a given party for constituency r we compute iOg(V,t9s7) using Iog(‘E7,n?s,) = 1og(%,19*3) +
c 5fr
KlY.5~og(~~,~~*~/~~.,~~~)
(6)
where v,,, is the percentage of the vote at time t in constituency s for the given party. Repeating equation (6) for each of the three parties gives prehminary estimates of the predicted 1987 votes for these parties in constituency r. The procedure does not, however, guarantee that these predictions sum to 100, although in practice in almost every case their values had a total of between 98 and 102. As a final stage in the estimation of the 1987 votes the V values were scaled to give Gi.t9a7 values summing to 100. These G values can now be compared with the actual w values that occurred, using the statistic Sz to assess the contribution of the ‘previous election effect’. The value of S2 is dependent upon the choice of the constant a in the weighting factor defined by equation (4). Values of a=l, 1.5 and 2 gave rise to S’ = 17308, 16871 and 17310, respectively. Since smaller S2 values represent better fits, the value a = 1.5 was the best of those considered.
GRAHAM J.
G. UPTON
63
Comparing the value S2 = 16871 with that obtained earlier (3’ = 19190), we see that the introduction of the ‘previous election effect’ has accounted for about 12 per cent of the variation of the constituency changes about the overall national change.
Geographical
effect
In writing about the 1987 general election Cozens and Swaddle (1987) remark that ‘1987 witnesses the emergence of fully-fledged regional patterns in Britain’, though differences in the two-party Con/Lab swing had long been evident in the contrasts between the values obtained from urban and for rural districts in earlier elections. A recent account of the impact of geography in English politics is provided by Johnston (1985). After correcting the vote change for partyj in constituency i for its ‘previous election effect’, we have a net vote change eG defined by the equation ^ eG=Vij--t+j (7) which represents an estimate of the aggregate of the last four terms of equation (1) as applied to party j in constituency i. If there is a geographical effect then, barring ‘random’ variations, the e-values for nearby constituencies will be similar. An average of the e-values for the constituencies near to constituency I will provide an estimate of the geographical effect, gi;, for partyj and constituency i, and the difference f;j = eq - gf,
(8)
can then be attributed to the final three terms of equation (1). Clearly the nearer a constituency is to constituency r, the greater is its relevance, and it is therefore sensible to use a weighted average which places greatest emphasis on the nearest constituencies. This suggests using the weight zr,$ for constituency s, where z,,, is defined by the equation z,,, = ds,t;44: + * * * + 4,iiilq)t (9) and the summation in the denominators excludes d;,j and also omits constituencies affected by by-elections. The value of the constant b is positive and is to be determined. The quantity d+,$represents a measure of the distance between constituencies r and s, and is defined by 4$ = (x, - &)* + (3%- Yr)2,
00)
where (xi,yi) is the location of the centre of the population of constituency i. In practice the precise values of X; and yI are unknown, but small changes in these co-ordinates make negligible differences to the z-weights. The values chosen were based on a national grid map of England, with possible transcription errors being checked by using a plotter to display the connections between neighbouring constituencies, When the estimated geographical effects are added to the estimated ‘previous election effects’ there is a considerable increase in the precision with which the actual changes are matched, as would be expected. The extent of the improvement is a function of the constant b used in the distance-weighting factor defined in equation (9). The value b=I gives rise to S2 = 14376, whilst b=i! and 3 give S* = 13206 and 14375, respectively. The optimal value of b=2 corresponds to an inverse square law for geographical distance effects which is precisely what the laws of physics would have suggested.
64
The Components
ofVoting
Change in England 1983-1987
FIG. 2. Major estimated geographical effects for the Conservative Party.
The inclusion of the geographical effect has resulted in a further reduction of S2 from 16871 to 13206, which amounts to a reduction to 69 per cent of the value obtained assuming a constant national change. Figs 2 to 4 show the regions of major gains and losses for the three parties. These locations are based not on the observed changes for the various constituencies but on the estimated gi values (using b=2). This results in a much less ‘spotty’ effect for the maps, since, by definition, the local effects specific to individual constituencies have been eliminated. The principal patterns to be seen in Figs 2 to 4 are summarized and quantified in Table I. Although portrayed as a geographical variation it is plain that the most striking effect here is due to Dr Owen who as an Alliance leader evidently caused large pro-Alliance gains from erstwhile Conservative supporters in the neighbouring Plymouth seats. The Conservatives experienced huge losses in the central Liverpool seats, amounting to over 10 per cent in some cases. The use of weighted averages naturally results in a smoothing of the entire array of gains and losses, which is particularly marked in the case of Liverpool, and results in rather large estimated ‘local effects’.
GRAHAMJ. G.
65
UPTON
FIG. 3. Major estimated geographical effects for the Labour Party.
A noticeable feature of Table 1 is that, with the exception of the major variations involve interchanges between the Labour of the other two parties: on the whole there appears to have around the voting triangle, with the Conservatives losing voters losing voters to the Alliance.
Intervening
By-Election
the Plymouth seats, all party and one or other been a clockwise cycle to Labour and Labour
Effect
Anticipating that the more spectacular by-election swings will leave a lasting impression on the subsequent general election result, all the previous calculations have been made omitting the 14 English constituencies that were affected by a by-election during the period 1983-87, although the resulting estimates of the ‘previous election effect’ and of the ‘geographical effect’ were applied to these constituencies and the values quoted for S2 did include these affected constituencies. Table 2 records the seven most influential by-elections, whose dates appear to be distributed fairly haphazardly between the two general elections. With the exception of Truro, each of these by-elections exhibited a huge gain in the Alliance vote which
66
fbe Components
of Voting Change in England 1983-1987
FIG. 4. Major estimated geographical effects for the Alliance.
was largely maintained in the 1987 general election. Since the prevailing national trend was for a slight diminution of the Alliance vote, the departures from the ‘expected’ vote are the more marked and are summarized by the huge S values, which compare with an overall average contribution to S2 of 22 for the remaining constituencies. Upton and Sarlvik (1981) suggested a model for voting in general elections which implied that the act of voting itself was an important reinforcement of a voter’s political identification. The results of Table 2 suggest that the same is true of by-elections, although in this case the lower turnout for by-elections implies that only about 75 per cent of the voting population experience this reinforcement. The subsequent general election could then be expected to exhibit results that represented a compromise between the returns in the by-election and in the preceding general election, suitably modified by the shifts in opinion over the intervening time period. Modelling the intervention effect of by-elections is not the object of the present paper, however, and here the intervening by-election effect is taken to be constituency specific. Removal of the 14 affected constituencies reduces the overall S* by a total of 2091 (11% of its originaf vahte) so that we have now accounted for 42 per cent of the original deviations from the simple model of national change.
GRAHAM J. G. UPTOEJ
67
TABLE 1. Major geographical variations Number of constituencies Average chL;bpes (%) principally affected Con All
Region
3 12 8 2 3 13 6 12
Plymouth Liverpool London North Central Norwich Portsmouth and Gosport Tyne and Wear Bristol London, North West
Incumbency
-7.5 -5.0
-2.5 -3.0 +3.0
+6.5 +4.5 -5.0 -3.5 +5.0 +3.0 -3.0 -2.0
+3.5 +5.0 -2.5 +2.5
Effect
We now turn to consideration of the personal vote acquired by the sitting MP. By virtue of his or her status, an MP always features prominently in local newspapers and on local radio and television reports. This leads to a degree of local fame that can be anticipated as an ‘advertising’ advantage when it comes to election time. Probably the most direct evidence of the personal vote acquired by an MP is rather paradoxically available when that MP retires. There were 83 cases in which an English Ml? retired in either 1983 or 1987 and was replaced by a candidate of the same gender, with the candidates of the other main parties also being of the same gender on the two occasions. The need to be careful over the genders of the candidates arises because we anticipate that the replacement of a candidate by one of the opposite gender may also have an effect. In the previous stages of our analysis we have estimated and removed effects due to the national change, the previous election effect and the geographical effect. This leaves us with a set of residual figures representing the aggregate of the final three terms of equation (I)-we are omitting the constituencies affected by by-elections from all that follows. In order to make progress by an incumbency residuals
we assume temporarily
that, for constituencies
affected
change, there are no gender or local effects, so that our estimated
for these constituencies
are the estimates
of the incumbency
effect.
An
TABLE 2. Most important by-elections
Constituency Greenwich Portsmouth South Newcastle-underLyme Chesterfield Truro Derbyshire West Ryedale
Date of by-election
1983 result Con Lab All
By-election Con Lab All
1987 result Contribution Con Lab All to sa
262.87 14.6.84 17.7.86
35 51 36
39 23 42
26 26 22
11 35 20
34 27 41
54 38 39
24 44 28
35 13 41
41 43 31
599 391 290
1.3.84 12.3.87 8.5.86 8.5.86
32 38 56 59
48 5 17 10
20 57 27 31
15 49 32 7 40 20 41 8
36 61 40 50
25 41 53 53
46 10 12 8
30 49 35 39
241 152 150 133
The Components
68
of Voting Change in England i%-1987
alternative viewpoint, which may seem less drastic, is to regard the estimated residuals as being estimates of the incumbency effect which are contaminated by an error component comprising the gender and local effects. So long as it is reasonable to assume that these contaminating errors cancel one another out, the two viewpoints are equivalent. The figures in Table 3 are therefore summaries of the ‘residuals’ obtained after removing the previous election and geographical effects. The variability inherent in these figures is partly a consequence of the intrinsic variability in the size of an incumbency effect (it will vary from one Ml? to another) and partly as a consequence of the ‘unexplained’ variation due to gender and local effects. Table 3 shows that there were 45 relevant cases in which an MP retired in 1983 and was replaced by a new party candidate who was duly elected. This candidate, the new MP, had no personal vote in 1983, but had enjoyed four years of free publicity by 1987. As the Table shows, this is estimated to have been worth about 0.8 percent of the 1987 three-party vote. If we assume an average turnout rate of 75 per cent for an average constituency of 65,000 voters, then this is worth about 400 votes gained by the sitting MP and lost by the rival candidates-a swing of up to 800 votes. This agrees well with the findings of Curtice and Steed (1980) w h o estimated a personal vote of around 750 votes on the basis of their informal study of marginal constituencies in the 1979 general election. Table 3 shows that there were 38 relevant constituencies in which the sitting MP retired in 1987. The new candidate in 1987 has lost the benefit of the personal vote enjoyed by his or her predecessor and hence does rather worse than would otherwise have been expected. The estimated disadvantage is 1 per cent of the three-party vote (about 500 votes), which is rather more than the advantage gained in 1987 by the new MP. Obviously the new MI’ has had only 4 years to build up a personal vote, whereas the MI% retiring in 1987 had been in Parliament for longer-an average of about 21 years. The data for these 38 constituencies provide no evidence of any linear increase in the personal vote with time. Since 400 of the 500 votes accrued by the long-service Ml’s can be attributed to their first four years in office, this is not surprising. However, Mughan (1986) has found evidence of a time dependency and reports an estimate of 0.08 per cent per year. A second source of information about the personal vote arises from seats which changed hands in 1983. Suppose a seat that is contested by an MP representing party X was won in 1983 by the representative of party_ Y. The personal vote, due to being an MP is then lost by the party X candidate, who fares relatively badly in 1987, while the party Y candidate, the new MP, fares relatively well. Thus, on average, eve7z in rhe absence ofany generalpotitical change, a new MP will increase his or her majority in the following
general
TABLE 3. The
effect of the retirement of an MP on the fortunes
Type MP MP
election.
retired 1983 retired 1987
of his or her successor
in 1987
Number of seats of this type
Mean effect
Standard error of mean
t-value
Tail probability
45 38
0.846 -1.020
0.309 0.433
2.74 -2.35
co.005 -0.01
69
GRAHAM J. G. UPTON TABLE 4. The effect of the gain of a seat in 1983 on the fortunes
of the party representatives
in
1987
Type (i) Con or Lab gains from SDP 1987: Con or Lab candidate 1987: Alliance candidate (ii) Con gains from Lab or Lib 1987: Con candidate 1987: Lab or Lib candidate
Standard
Number of seats of this type
Mean
effect
error of mean
11 11
2.908 -1.835
0.809 1.419
3.59 1.29
18
1.223 -1.366
0.586 0.902
2.09 -1.51
18
t-value
Tail probability
<0.005
==0.025
After eliminating seats influenced by gender effects, there were 29 English seats that changed hands in 1983, and the results are summarized in Table 4. In 11 cases the sitting Ml’ had changed his or her original allegiance to the SDP, and these are treated separately from the remainder, since they might be expected to behave in a slightly different fashion. Because of the smaller sample sizes, the effects noted in Table 4 are rather less significant in statistical terms than those summarized in Table 3, though in absolute terms the magnitudes of the personal votes are appreciably greater, ranging between 600 and 1,100 votes. The reason for this is not clear, though it may be related to the greater degree of marginality of these seats (which changed hands, whereas those in Table 3 did not). Tables 3 and 4 give the standard error of the estimated mean effect for changes in incumbency, and the high significance of the results may suggest a greater degree of uniformity than is actually present in the individual estimates. In fact the 141 individual estimates of the benefit of incumbency (reversing the signs for the 67 cases of disadvantage) provide figures ranging from -6.2 per cent to 12.1 per cent with 44 of the 141 (about 30%) showing a negative benefit. The figure of 12.1 per cent refers to the exceptional case of Croydon North West, where the Alliance vote collapsed to 16 per cent, following figures of 32 per cent in 1983,40 per cent in the 1981 by-election and 10 per cent in 1979. Cain, et al. (1984) estimate that an excellent constituency MD could attract about 6.5 per cent more of the vote than a very poor one, and this seems a more reasonable guide to the true spread of the personal vote. TABLE 5. The effect of party
Party
Constituency
type
on the incumbency
effect
Number of seats of this type
Mean effect
Standard error of mean
t-value
Tail probability
Con
MP retired 1983 MP retired 1987 Overall
31 21 52
0.445 -0.679 0.539
0.364 0.352 0.258
1.22 -1.93 2.09
-0.1 co.05 KO.025
Lab
MP retired 1983 MP retired 1987 Overall
13 16 29
1.668 -1.380 1.509
0.560 0.925 0.560
2.98 -1.49 2.69
-0.005 CO.1 =o.oos
70
The Components of Voting Change in England 1983-1987
Table 5 shows a breakdown by party of the information given in Table 3 (with Alliance seats excluded). Although the sample sizes are now rather small some statistically very significant results appear. It appears that a Labour MP attracts, on average, a considerably greater personal following (1.5% of the vote or about 750 votes) than does a Conservative MP (0.5% of the vote or about 250 votes). The difference between the two overall estimates is significant at the 5 per cent level.
Gender
Effect
Over the years opinion polls have consistently revealed a preference by voters for male candidates. Ross (1955, Appendix III) reports a 1952 poll in which 16 per cent of voters stated that they were less likely to support a female candidate, while only 10 per cent stated that they were more likely to support a female candidate. Hills (1981) reports figures from surveys conducted by the European Commission in 1975 and 1977 showing a greater public confidence in male as opposed to female MI+, while Rasmussen (I 983b) notes that a poll conducted at the 1979 general election showed that 44 per cent of voters admitted preferring a male MP with only 11 per cent preferring a female MI’. Given these findings one might expect to find clear evidence from the voting figures of a disadvantage to a party in fielding a female candidate, but this has proved surprisingly elusive. An excellent summary of the work in this area is provided by Welch and Studlar (1986) who note that: ‘While social desirability may prompt some to give responses supporting a woman when they really would not, actual events may prompt others to support a woman even though they say they would not.’ In other words party allegiance is a dominating factor controlling the voter’s actions. Rasmussen (1983a) studies the effect of women candidature in the seven general elections from 1951 to 1979 inclusive. He does not eliminate the effects of other factors but rather compares the overall mean changes in cases where a male candidate is replaced by a female candidate with those in which a female candidate is replaced by a male candidate. Changes in incumbency are therefore confounded with changes in the candidature of a competing party and the general picture is thereby confused and Rasmussen does not find any clear evidence of an effect. Nevertheless, if attention is confined to cases which differ by more than 1 per cent of the vote, we discover four cases in which the replacement of female candidates by male candidates leads to an appreciable gain over the reverse situation and only one case of an appreciable loss. Each ‘case’ referred to here is an average for a particular party for all the relevant constituencies in a particular election. We now adopt a similar procedure to that applied in the previous section by considering the 101 English constituencies, having the same sitting MI’ in both 1983 and 1987, in which either a rival male candidate was replaced by a female candidate (56 cases) or a female candidate was replaced by a male candidate (45 cases). The restriction to these 101 cases ensures that there are no incumbency effects to consider. After removing the estimated politicai and geographical effects we are left with residual deviations that are attributable in part to the gender effect and in part to local effects. For the present we assume that the latter are non-existent, or, equivalently, that on average they have mean zero. The averages reported in Table 7 for these residuals are therefore assumed to represent the gender effects only. It will be seen that although these averages have the expected signs (showing a disadvantage to a party when a male candidate is replaced by a female and an advantage in the reverse circumstances), they
GRAHAM J. G. UF’TON
71
show too much variability to be judged significant. For these 101 constituencies there is an overall disadvantage to a female candidate of about 0.25 per cent, which is equivalent to about I25 votes. A further 48 constituencies experienced both a replacement of an unsuccessful candidate by a candidate of the opposite gender and the retirement of the local MP. Using the estimates of the incumbency effect from Table 5, and allocating the gains and losses due to incumbency between the rival candidates in proportion to their votes gained, enables us to retrieve further information about the gender effect. The resulting estimates are necessarily much less precise, since they include the imprecision attached to the incumbency estimate: this is reflected in the much larger standard errors for these estimates as indicated in Table 6. It is noteworthy that in every one of the six situations considered the estimates are consistent in their signs, showing female candidates to be at a disadvantage. To obtain an overall pooled estimate of the effect we use a weighted mean in which the weights are inversely proportional to the square of the standard errors. Combining all the results we find over the I49 affected constituencies that there is an effect that is statistically significant at about the 1 per cent level and amounts to a loss of about 250 votes (a potential swing of up to 500 votes). In Table 6 the figures that refer to cases of an MP retiring in 1983 appear considerably larger in magnitude than the other figures, suggesting that the personal votes for these MI’s may have been greater than the estimated values presented in Table 5. It would be possible to estimate the effects of gender and incumbency simultaneously, but since there are inevitable assumptions to be made concerning the source or destination of the votes that change hands, such statistical niceties appear misplaced. The separate estimates of 0.391 per cent and 0.644 per cent for the anti-female effects in 1987 and 1983 are suggestive of a decline in the effect, but these two estimates do not differ significantly and a longer time series would be required to gain any worthwhile evidence on this aspect. TABLE 6. The effect of a change in the gender of the candidate of a party
Gender of candidate
Number of constituencies
Standard Adjusted error of mean effect adjusted mean
Tail t-value probability
(i) Male ‘83, Female ‘87 No change of incumbent MP retired 1983 MP retired 1987 Overall
56 16 10 82
(ii) Female ‘83, No change MP retired MP retired Overall
45 18 4 67
0.495 1.142 0.080 0.644
0.364 0.733 1.518 0.323
1.36 1.56 0.05 2.00
<0.025
101 149
0.266 3.505
0.239 0.218
1.12 2.32
-0.01
Male ‘87 of incumbent 1983 1987
(iii) Combined effect No change of incumbent Overall
-0.082 -1.578 -0.222 -0.391
0.317 0.834 0.928 0.296
-0.26 -1.89 -0.24 -1.32
co.05 co. 1
The
72
Components
of Voting
TABLE 7. The major
Change
in England
estimated
local effects
Constituency
Observed change (% of three-party vote)
Blyth Valley Newham South Falmouth and Camborne Croydon North West Liverpool Mossley Hill Liverpool Garston Sheffield Hillsborough Southport Portsmouth North
Con -11.1 10.3 -6.4 4.0 -14.5 -14.0 -10.8 -5.8 0.0
Lab 2.6 -8.6 -0.5 12.4 11.9 7.1 6.8 -1.9 -1.1
Local By definition
local
effects
immediately
obvious
be produced
for several
constituencies vote are listed
explanation.
In the elections
estimated
All 8.5 -1.8 6.9 -16.4 2.6 6.9 4.0 7.7 1.1
local effect Lab -4.8 -11.3 -5.4 10.2 3.8 -1.9
Con -8.5 10.4 -5.7 0.6 -10.8 -7.9 -9.4 -3.8 2.4
specific
0.3
-5.5 -9.1
All 13.3 0.9 11.1 -10.8 7.0 9.8 9.1 9.3 6.7
and do not necessarily
Nevertheless,
of the constituencies
with local effects in Table 7.
Estimated
Effects
are constituency
political
1983-1987
immediate
that display at greater
of 1974 and in the 1979 election
major
have
explanations
local effects.
any can
All the
than 9 per cent of the three-party Blyth
Valley
was contested
by the
E. J. M’l1 ne, standing as an Independent Labour candidate. Mr Milne was successful in the February 1974 election and was only narrowly defeated in the October 1974 election. He evidently commanded a considerable personal vote and former
labour
MI’,
it seems likely that the consequent turmoil amongst Labour this day, since the rise in the Labour vote in Blyth Valley neighbouring Four
voters may continue was much less than
to in
seats.
constituencies
may have been
affected
by earlier
or nearby
by-elections.
The
rise and fall of the Alliance in Croydon North West has already been noted. It is interesting to note that Newham South borders Greenwich, Falmouth and Camborne borders Truro and Portsmouth North borders Portsmouth wich, Truro and Portsmouth South all feature prominently
South, in Table
relevant that Southport by-election in 1981.
subject
borders
Crosby,
which
was
the
and that Green2. It may also be of a high
profile
Liverpool Mossley Hill and Liverpool Garston are neighbouring constituencies for which it is clear that the geographic effect was much greater than has been estimated. Naturally Table 7 emphasizes the possible magnitude of the local effects. In general, however, these are much smaller, and a summary of their magnitudes is given in Table 8, in which each constituency is characterized by the magnitude of the maximum of the three separate party effects. It will be seen that typically these amount to less than 3 per cent of the three-party vote. Despite the undoubted significance of the incumbency effect and the probable significance of the gender effect, these are clearly much smaller in value than the typical local effects. The value of S’ is reduced by only a few hundred when corrections are made to take account of the incumbency and gender effects, so that some 60 per cent of the original S2 value (I 9190) remains to be accounted for by the local effects, which are
GRAHAM TABLE 8. Distribution
Absolute size of maximum local effect
of the sizes of local effects
Cl.0
C2.0
49
Number of constituencies
the product of local issues and changes electorate between successive elections.
73
J. G. UPTON
134
C3.0
C4.0
K5.0
112
a4
55
in the socio-economic
composition
>5.0 78
of the
Discussion This analysis of the components -of voting change has consisted of a series’ of adjustments, each conditional on the preceding adjustments. Although the sequence chosen appeared to be the most practicable, it is clearly not the only possible sequence and it is evident that slightly different results would have been obtained with an alternative sequence. Equally, a number of the effects could have been treated in rather different ways, and the particular case of the geographical effect is discussed below. However, the evidence in the literature suggests that all the effects found are real effects and the present paper appears to be the first that attempts a simultaneous quantification of these effects. Adherents of a regression approach in which all the effects are estimated simultaneously may object to any sequential approach, but the geographic effect does not lend itself to a regression approach unless one is prepared to postulate some underlying geo-political ‘surface’ which will manifestly be a rather crude approximation to reality. The approach adopted for dealing with the geographical effect relied upon the relative distances between the ‘centres’ of neighbouring regions. One flaw with this approach is that apparently close neighbours may not be so close in reality. A partial explanation of the presence of the two Liverpool constituencies in Table 7 is that in distance terms they are very close to the constituencies of the Wirral (North and South), Wallasey and Birkenhead, which lie on the other side of the Mersey. Further research will concentrate on finding alternative weighting procedures based on a mixture of distance and physical contact between constituencies, which should prove to provide a rather better set of estimates of the geographical effect. The retirement of MPs is generally beyond the control of a party-though it is clear that mid-term by-elections at times when the government is in disfavour are to be avoided, if at all possible, not only for their immediate effect, but also for the carry-over effect that may be manifest in the following general election.’ The estimate of the anti-female vote amounts only to a few hundred votes and may appear relatively unimportant. However, some of the majorities gained by ME’s in the 1983 election were also very small. Three constituencies (Mansfield, Norwich South and Portsmouth South) might have been lost if there had been a female candidate, while one (Dulwich) could well have been gained by a male candidate, since there the female Labour candidate lost by just 180 votes. There are of course other factors that play a part and that could have been deduced from the data. Some of the variation recorded here as local effect may be attributable to differences in the turnout rate, both between constituencies, and, more importantly in the present context, within a constituency between elections. The decision to abstain can have a crucial effect upon the outcome of an election, with differential abstentions
74
The Components
of Voting
Change
between majority
rn Engianci
19X-1987
the supporters of the major parties almost certainly of the change between the two 1974 elections. A second omitted is a potential colour bias. However, these are issues lying the present paper, which has already made the contemplation of quite sufficiently complicated.
accounting factor
for the
that has been
outside the range of voting change seem
Notes 1. The omitted
constituencies are Croydon North East, the speaker’s constituency, Bradford North, Erewash, Islington North and Newham North West, which were contested b? Independent Labour candidates in 1983, Hackney South and Shoreditch, Hammersmith and Liverpooi Broadgreen which were contested by Independent Liberal candidates in 1983, Bournemouth East and Spehhorne which were contested by Independent Conservative candidates in 1983 and Windsor and Maidenhead which was contested by an Independent Conservative candidate in both 1983 and 1987. 2. It is interesting to note that the loss of the personal vote goes a small way to explaining the anti-government swing in by-elections, since a majority of these occur in constituencies held by the government. During the period 1950-82 there were 294 by-elections, of which 63 per cent of the by-elections during Labour’s rule were Labour seats and 55 per cent of the by-elections during the period of Conservative domination were Conservative seats. See also Mughan (1986).
References J. Airchison, The Stutisticui Ana&s of Co~po~z~~o~~[ Data, (London: Chapman and Hall. 1986). B. E. Cain, J. A. Ferejohn and M. I’. Fiorina, ‘The Constituency Service Basis of the Personal Vote for U.S. Representatives and British Members of Parliament’, American PoLticalScience Review, 78, 1984, pp. 110-25. P. Cozens and IS. Swaddle. ‘The British General Election of 1987’, ElectoralStudies, 6, 1987, pp. 263-6. J. Curtice and M. Steed, ‘An Analysis of the Voting’ in D. Butler and D. Kavanagh, The British General Election of 1979, (London: Macmillan, 1980). J. Hills, ‘Candidates, the Impact of Gender’, Parliamentary Affairs, 34, 1980-81, pp. 221-S. R. J. Johnston, The Geography of English Politics, (London: Croom Helm, 1985). W. L. Miller, EIectoml Dynamics in Britarn since 1918, (London: Macmillan, 1977). A Mughan, ‘Toward a Political Explanation of Government Vote Losses in Mid-Term By-elections’, American ~o~itlcul Science Review, 80, 1986, pp. 761-7. J. S. Rasmussen, ‘Women Candidates in British By-elections: A Rational Choice Interpolation of Electoral Behaviour’, Poliricai Studies, 29, 1981, pp. 265-74. J. S. Rasmussen, ‘Women’s Role in Conremporary British Politics: impediments to Parliamentary Candidature’, Parilamentar? A,(f,zrs. 36, 1983a, pp. 300-15. J. S. Rasmussen, ‘The Electoral Costs of Being a Woman in the 1979 British General Election’, Comparative P&tics, 15, 1983b, pp. 461-75. J. F. S. Ross, Elections and Electors, (London: Eyre and Spottiswoode, 1955). G. J. G. Upton, ‘The Diagrammatic Representation of Three-Party Contests’, Political Studies, 24, 1976, pp. 448-54. G.J. G. Upton and B. H. Sarlvik, ‘A L oyalty-distanceMode forVotingChange’,Jotrrnafofthe Royal Statistical Society, Series A, 144, 1981 1 pp. 247-59. G. J. G. Upton and S. J. Stray, ‘The Effect of the Environment on Party Identification’, Environment and Planning, Series A, 18, 1986, pp. 1391-400. S. Wetch and D. T. Studlar, ‘British Public Opinion Toward Women in Politics: A Comparative Perspective’, western PoIztzcit1 Qtiarterly, 39, 1986, pp. 138-54.