Statistical analysis of constituency results in the last two general elections strongly suggests that ethnic minority (Black or Asian) candidates secure a smaller share of the vote for their parties than do white candidates. While it is not possible to prove from the evidence why this is the case, the obvious presumption must be that it is caused by racist voters being deterred from voting for an ethnic minority candidate. The effect was strongest in the case of Labour candidates, depressing their vote share by more than three-and-a-half percentage points, but was also present for Liberal Democrats; however, there was no statistically significant loss of votes found in the case of Conservative ethnic minority candidates.
There are two main types of analysis of voting behaviour. The first, of course, is survey research. Unfortunately, there have (as far as I know) been no published surveys measuring the effect of a candidate's race in recent British elections. (It is possible, of course, that one or other of the parties may have researched the question in their unpublished private polls but, if so, they have so far remained unpublished.)
The alternative is the analysis of aggregate voting data - in short, election results. It has the advantage over polls, of course, that the data is definitive - no questions of margins of error apply. On the other hand, there are far greater problems of interpretation. We do not know who has voted in what way, except on a very broad geographical categorisation. (In Britain, it is illegal for general election results to be published or indeed counted at any level below the constituency - no precinct-level data here!) Nor do we know why they vote in a certain way, except by inference - we can find differences in the character of constituencies, and associate them with differences in their voting behaviour, but it remains a moot point whether, and if so in what way, one causes the other. (A simple example will suffice. Analysis of the last election will show that the British National Party scored its highest votes in constituencies with high ethnic minority populations; but it would certainly not be a valid inference that Black and Asian voters support the BNP!)
The difficulty of analysing the effect of any factor on voting behaviour from constituency results is that so many possible factors are involved. Suppose we consider the case of one particular ethnic minority MP, Keith Vaz in Leicester East. Mr Vaz won 57.6% of the vote. Is that good or bad? It is higher than the average for Labour across the country, but then we would expect that in Leicester. But how much higher should it be, all other things being equal? How do we work out what the "normal" vote for Leicester East would be if Mr Vaz were not the candidate. (Since he has been the candidate for several elections, it is not easy.) And if it is worse than expected, is that because Mr Vaz is Asian, or because he was a government minister, or because of the widely-publicised allegations about his role in the Hinduja affair which surfaced before the election, or because the health problems that followed those allegations hindered his campaigning, or ...? Not easy, is it?
What we need as a basis for analysis, then, is some means of defining a norm for performance by a given party's candidate in a given constituency. We can find this by using the measure that tends to vary least across constituencies, the party's change in share of the vote between elections. If we assume that this should - all other things being equal - be the same across every constituency (effectively assuming "uniform swing"), then we can start looking for explanations of the variations that we find. Our basic data set, therefore, is the constituency results for the last two elections, 1997 and 2001.
The analysis was performed by comparing three classes of constituencies: those where the candidates in 1997 and 2001 were of different ethnicity, divided into those where the change was from white to non-white and those where it was vice-versa, and those constituencies where there was no change. This last class, which of course comprises the vast majority of constituencies, is our control group, where we can assume that the change in the party's vote would not be affected by the candidate's race, since this was the same in both cases.
First we can simply look at the average performance by each party in its control group of constituencies, and see whether the majority of ethnic minority candidates seem to have done better or worse than this average. If it is true that the voting public is prejudiced against and less likely to vote for black and Asian candidates, we would expect that the party would tend to do worse than average where a minority candidate in 2001 replaced a white candidate in 1997, and conversely to do better than average when a white candidate replaced an ethnic minority candidate (since that should have caused the party's vote to be worse than normal in 1997 but not in 2001, so that the effect would be an improvement in the change figure.)
What do we find? There are 641 constituencies in Great Britain, of which we must exclude 4 where all three major parties did not run in both elections (Wyre Forest, Tatton, West Bromwich West and Glasgow Springburn). Of the remaining 637, Labour's candidate was either white at both elections or black/Asian at both elections in 621 cases, and on average in these seats Labour's share of the vote fell by 1.9%. The Lib Dems had 603 constituencies where there was no change in candidate ethnicity, and here they averaged a 1.5% increase in vote share. The Conservatives had 616 constituencies where there was no change of their candidate's ethnicity, and they averaged an increase in share of 0.8%.
Labour had 11 non-white candidates in constituencies where the Labour candidate in 1997 was white; in 7 of these 11 the change in Labour share was worse than the -1.9% average. There were 5 white Labour candidates in constituencies where the Labour candidate in 1997 was non-white: in all 5 the change in the Labour share was better than the -1.9% average. The pattern fits what we would expect if Labour black and Asian candidates are at a disadvantage.
The Liberal Democrats had 21 non-white candidates in constituencies where the LibDem candidate in 1997 was white; in 16 of these 21 the change in LibDem share was worse than the +1.5% average. There were 13 white LibDem candidates in constituencies where the LibDem candidate in 1997 was non-white: in 10 of these 13 the change in the LibDem share was better than the +1.5% benchmark. Again, the pattern is entirely consistent with prejudice against ethnic minority Lib Dem candidates.
The pattern is less clear for the Conservatives, though. The Conservatives had 12 non-white candidates in constituencies where the Conservative candidate in 1997 was white; in 9 of these 12 the change in Conservative share was worse than the +0.8% average. But there were 9 white Conservative candidates in constituencies where the Conservative candidate in 1997 was non-white: in only 3 of these 9 was the change in the Conservative share better than the average, not worse.
The same pattern is clear when we compare the average results of each party in each group of seats.
|160||Average change in party share of the vote (number of cases in brackets) in constituencies where:|
|160||candidate was non-white in 1997, white in 2001||candidate ethnicity the same in 1997 and 2001||candidate was white in 1997, non-white in 2001|
|Labour||+4.2 (5)||-1.9 (621)||-4.4 (11)|
|Liberal Democrat||+2.9 (13)||+1.5 (603)||-0.1 (21)|
|Conservative||+0.3 (9)||+0.8 (616)||+0.7 (12)|
Finally, we make an overall estimate of the size of the effect and test whether the evidence is strong enough to be reliable. For this we use a technique called linear regression analysis. Again we treat each party's candidates separately. We give each constituency where a white candidate replaced a non-white candidate a value of -1, constituencies where the candidate was of the same ethnicity in both elections a value of 0, and those where a non-white candidate replaced a white candidate a value of +1. We could then graph all the results, with this value on the x-axis and the change in the share of the vote on the y-axis, and find the equation of the line of best fit that explains the relationship between the two.
The results are as follows:
|160||Co-efficient of regression||R-square||Significance|
What do these figures mean?
The co-efficient of regression in each case, which measures the slope of the line of best fit, is the measure of how much each party's vote seems to have been affected. Selecting an ethnic minority candidate cost Labour about 3.6% of the vote and cost the Lib Dems about 1.5% of the vote.
Both these figures are "statistically significant". This is what is measured in the significance column, which in effect measures the probability that the pattern might arise simply by chance. For the Liberal Democrats, the significance is 0.027; that is, there are only 2.7 chances in a hundred that the Liberal Democrat figures could have arisen by pure chance, good enough for us to rely on. (As is customary in opinion polling, we use the 95% threshold for significance - any significance figure up to 0.05, or five chances in a hundred of being a coincidence, is taken as being sufficiently reliable for our purposes.) In fact for Labour, the pattern is much clearer - there is less than one chance in a thousand that the Labour figure could have arisen by chance.
In the case of the Conservatives, however, although it is possible to derive a co-efficient from the regression, the significance figure is too high. If we were to pick this many constituencies at random, we would find this much deviation from the rest of the constituencies more than a third of the time. So the figures prove nothing about the results in constituencies where there was an ethnic minority Tory. In other words, there was no statistically significant effect detected in the case of Conservative candidates.
The figure in the middle column of the table, the R square, simply measures what proportion of the variation in each party's constituency vote change can be attributed to the race of the candidate. As might be expected, these figures are very low: even the highest, for Labour, finds that only 2% of the variation is explained by these figures. Of course there are a huge number of other factors that affect constituency results as well, and a candidate's race plays only a tiny part in the whole scheme of things - important for candidate selectors to remember. Nevertheless, one of those factors does seem to be the candidate's race.
Why? Clearly the most obvious, and indeed overwhelmingly likely, explanation, is that a small minority of voters are prejudiced against black or Asian candidates, and refuse to vote for them when they would otherwise vote for the party. Nevertheless, we have to bear in mind that we have not proved this, only that there is a consistent pattern to the results in the constituencies where such candidates are chosen. Other explanations are possible. For example, it is conceivable that rather than voters being prejudiced these poor results arise from party workers campaigning less hard for ethnic minority candidates. It is possible, though unlikely, that ethnic minority candidates are more likely to be selected in constituencies where their party is likely to do uncharacteristically badly than elsewhere. We could probably construct other conceivable explanations. But prejudice among the voters is probably the real reason.
Why should the effect be clear in the case of the Lib Dems and Labour but not in the case of the Conservatives, perhaps contrary to expectations? One factor is that the Conservatives performed worse than average across the country in marginal seats than in safe seats (which is why they gained so little reward for closing the gap on Labour in terms of votes). Since almost all the Conservative ethnic minority candidates were selected in seats safe for other parties, with only one (Shailesh Vara in Northampton) in a genuine marginal, any signs of a prejudicial effect against them may simply have been swamped by this other, more pronounced, factor.
But it is probably also true that it is "floating voters" who are most likely to be influenced in their vote by prejudice against a candidate; we can imagine that a party's core voters, committed to that party and probably long-time supporters of it, will be less likely to be swung in their voting whatever their opinion of the candidate. In both the 1997 and 2001 elections, of course, the Conservative vote has been reduced almost entirely to the party's hard core, while the vast majority of the floating vote has been with the other parties. It may simply be, therefore, that Conservative ethnic minority candidates have not suffered simply because the party was not going to get the votes of the potential switchers in the first place, regardless of who the candidate was.
It is important to understand that the figures prove nothing about individual constituencies or candidates. Any single candidate may have done worse than expected for any number of reasons quite unrelated to his or her race; and, indeed, as we have seen some ethnic minority candidates performed better than average. And, of course, assuming there is an anti-minority vote we would not expect it to be the same size in each constituency. The power of the analysis comes from its being spread over a sufficiently high number of cases. While any one minority candidate performing poorly may have been pure coincidence, so many have done so on a consistent basis that is possible to say that it is highly improbable that all these cases can be coincidence.
We should also note that there is no evidence of any instance where a candidate's race probably cost him or her the seat. None of the ethnic minority candidates in 2001 was defeated by a smaller margin than the apparent effect derived from the analysis. (This is partly because very few ethnic minority candidates were selected in marginals.)