British_Electoral_System

1. Introduction The British electoral system is often defended for producing workable Commons majorities, with the governing party accountable to voters rather than to other parties (Popper, 1988; Chandler, 1992; Johnson, 1992; Johnson, 1998; Norton, 1992; Pinto-Duschinsky, 1999a and Pinto-Duschinsky, 1999b). But since the 1970s, this defence has been undermined by four developments: the decline of the cube law; the increase in minor-party seats; the decreased cohesiveness of legislative majorities; and the rise in pro-Labour bias. The first three developments make workable Commons majorities less likely, and the fourth may hinder accountability by letting a party come first in votes but second in seats. The nightmare scenario for first-past-the-post’s defenders is an election which produces a hung parliament with the Conservatives first in votes, Labour first in seats, and the Liberal Democrats holding the balance of power and insisting on proportional representation (PR) as their price for supporting a government. These diseases poisoning first-past-the-post have not been properly recognised by its defenders or critics. Its defenders know that it is imperfect, but they have not grasped how unreliable it has become. Although most recent elections have seen large government majorities, this is partly fortuitous: the winning party has usually had a big lead in votes over the second-placed party. First-past-the-post’s frailties should be clearer once the winning party’s lead in votes shrinks. Critics of the system should also take note. Several discuss the cube law’s decline, but I will suggest that the rise in minor-party seats is more important. The increase in bias, often mentioned by critics, may not be as troubling as many imply. Meanwhile, neither critics nor defenders of first-past-the-post have spotted that decreased legislative cohesiveness aggravates the other developments. Both sides of the electoral reform debate should attend to these findings. The four developments are sometimes related. For example, increased minor-party voting has contributed to the cube law’s decline, the rise in minor-party seats, and pro-Labour bias. However, this paper is more normative than empirical, so I discuss effects more than causes. It is the cumulative effect of these developments that matters most — a double, triple or even quadruple whammy, potentially a knockout blow. The comparative dimension is noteworthy. No two single-member plurality (SMP) electoral systems work identically, so in this paper, ‘first-past-the-post’ only refers to Westminster elections. But similar systems have faced related problems. American congressional elections witnessed something similar to the cube law’s decline (Gelman and King, 1994), and in 2000 a president was elected with fewer votes than his opponent, under multi-member plurality voting. Minor parties have weakened SMP in Canada ( Johnston, 2000). In New Zealand, minor parties and two successive biased elections contributed to wholesale electoral reform ( Denemark, 2001). Every electoral system is partly contingent on exogenous factors, but SMP seems especially delicate. First-past-the-post’s problems are partly its own fault. This paper assesses the actual effects of these factors at past elections, and their possible effects in the future. New methods are presented for investigating the cube law, bias, and the effect of minor-party seats. I conclude that the standard defences of first-past-the-post have been weakened but are still viable. However, this paper’s central message is that although first-past-the-post is still defensible in theory, electoral reform is becoming more likely in practice. 2. The decline of the cube law In January 1950, an anonymous article in the Economist described what came to be called the ‘cube law’. At its simplest, the cube law means that in a two-party election, a party with a 1 percentage point lead in votes would have a 3-point lead in seats, and a party with a 2-point lead in votes would have a 6-point lead in seats. The article’s author, David Butler, concluded that first-past-the-post ‘exaggerates narrow majorities in votes into sizeable majorities in seats’ (Butler, 1950). Readers should note carefully the terms used here. Exaggeration, also called the winner’s bonus or responsiveness, refers to the advantage in seats that the party with most votes receives over the party coming second in votes. The cube law has an exaggerative index of 3, a square law has an exaggerative index of 2, and so on1. (The cube law is not a genuine law; some scholars prefer ‘cube rule’.) Strictly speaking, exaggeration refers to the two-party vote (votes cast for the two major parties — in Britain, the Conservatives and Labour) not the total vote (votes cast for all parties). The winner’s bonus may be increased or decreased by bias (sometimes called partisan bias), discussed in section 5. The size of the exaggerative index depends on the relative proportion of marginal and safe seats. For example, where all seats are extremely marginal, a party with just over half of the vote would win all of the seats in a two-party election, but with a small shift in votes the other party would win all of the seats — an extremely high exaggerative index. By contrast, where all seats are safe, a party will win the same number of seats whether it has 51 or 60% of the vote, say — a very low exaggerative index. Fig. 1 compares the pattern of marginals and safe seats at the 1970 and 1992 general elections. The 1992 histogram, which is wider and flatter, clearly has many fewer marginals: the exaggerative index should thus be lower. [pic] [pic](11K) Fig. 1. The frequency distribution of the Conservative share of the two-party vote. Notes: British seats won by major parties only. The Conservatives win all seats to the right of the 50% line, Labour wins all seats to the left. [pic] Curtice and Steed, 1982 and Curtice and Steed, 1986 have explained the cube law’s unravelling. Changing voting patterns intensified the North–South divide, producing more safe Labour and Conservative seats in the North and South, respectively, while resurgent minor parties made many safe seats safer by taking votes from the second-placed major party. First-past-the-post’s exaggerative quality even looked in danger of disappearing completely ( Curtice and Steed, 1982, p. 217). But more marginals arose in 1992 and 1997, reversing the previous decline ( Curtice and Steed, 1998). There are two ways of charting the changing exaggerative index, using Curtice and Steed’s (1982) focus on marginals, or Blau’s (2002) ‘integrated method’ which examines the whole frequency distribution (see Appendix A). Fig. 2 shows that both methods tell the same story: the exaggerative index fell markedly after 1970, reaching a trough in the 1980s before recovering to around a square law in 1997 and 2001. [pic] [pic](5K) Fig. 2. The cube law’s decline and partial recovery. Notes: both methods exclude Northern Ireland and minor-party seats. Following Curtice and Steed (1982), the starting-point is 1955 — the first election after the redistricting of constituency boundaries earlier that year, and the beginning of the cube law’s decay. [pic] The two methods above are preferable to Taagepera and Shugart’s (1989) deductive model of the cube law. This model excludes contingent factors like electoral geography, and cannot explain the halving of the exaggerative index in a 30-year period. For an extensive critique, see Blau (2002, pp. 110–117). 2.1. The effect on first-past-the-post Has the declining exaggerative index damaged first-past-the-post, by impeding secure government majorities or by reducing accountability' Despite Curtice and Steed’s understandable gloom in the 1980s, single-party governments with workable majorities have usually resulted, because the winning party has mostly had large leads in the vote over its main competitor (Curtice and Steed, 1998). Even when the exaggerative index was at its lowest, in 1983 and 1987, Labour was so unpopular that the Conservatives won easily. Nor have low exaggerative indices apparently hindered first-past-the-post’s accountability function, its ability to help electors ‘throw the rascals out’. However, in recent years the low exaggerative index has produced two unsatisfactory election results: October 1974 and 1992. The government majorities of 4 and 21 seats, respectively, would have been much higher under a cube law (around 30 and 80 seats, respectively). Both governments could then have prevented bills from amendment or defeat, reduced minor-party influence, and avoided minority status. (The Labour government lost its majority in 1976; the Conservative government lost its majority for 6 months in 1994/5 and for a few weeks in 1997. For the governments’ other troubles, see Schwarz, 1980 and Butler and Kavanagh, 1997.) These faltering governments lasted for 4.5 and 5 years, respectively, so together these parliaments constitute over a quarter of the post-1970 era — a worrying and previously unnoticed figure. While the cube law’s decline did not exclusively cause the low majorities, it did exacerbate the situation. The next section charts the possible effects of low exaggerative indices in future elections. Here we should note the difficulty of predicting future changes in the exaggerative index, because many factors affect the proportion of marginals. But without a dramatic shift in electoral geography, or a radically altered party system (such as one of the three main parties splitting or collapsing), something roughly around a square law, give or take a few decimal points, seems likely for the next few elections. 3. The rise in minor-party seats As in Canada and New Zealand, minor parties have grown dramatically in recent years. In 1955, minor parties took only 4% of the UK vote; since the 1980s they have taken 24–30%. Section 2 mentioned that the geography of this increase indirectly weakened first-past-the-post by furthering the cube law’s decline. This section examines a direct effect: minor-party victories. Minor parties won just 8 seats in 1955, but by 2001 this had risen to 80. This is a problem for first-past-the-post because, as I show here, the more seats are won by minor parties, the greater the probability that the winning major party will not have an overall majority. The new method presented here estimates the chances of first-past-the-post failing to meet its defenders’ requirement of single-party government, preferably with a workable parliamentary majority. This innovation helps us quantify how well the British electoral system works. It also shows that Curtice and Steed (1986, p. 216) were mistaken in claiming that the cube law’s decline, not the rise in minor-party seats, was the crucial factor in increasing the likelihood of hung parliaments. Both mattered, but minor-party seats mattered more. The thinking behind this method will be briefly outlined here. (Appendix B provides more detail.) Obviously, a higher exaggerative index makes it easier for the winning party to get a workable Commons majority. Assume, for example, that the winning party needs 53% of the available seats for a secure majority. Under an exaggerative index of 1, the party needs 53% of the two-party vote, but under an exaggerative index of 3, only 51%. Winning 51% of the two-party vote is easier than winning 53%. So, higher exaggerative indices make secure majorities more likely. Fewer minor-party seats also make secure majorities more likely. If there are no minor-party seats, a hung parliament cannot result (unless the election ends in a tie). The more seats are won by minor parties, the greater the chance that they will hold the balance of power, so hung parliaments become more likely. Consider an election where two parties compete for 600 seats. A majority of seats (301 seats) means that the winning party only needs to scrape a majority of the two-party vote, just over 50%. Now consider an election where minor parties take 150 seats, leaving only 450 for the major parties. The winning party still needs 301 seats for an overall majority — but this is now two-thirds of the remaining seats. Under a cube law, the winning party now requires about 56% of the two-party vote for a majority. So when minor parties win more seats, overall majorities become harder to attain. A hung parliament will therefore be avoided if one or other major party can reach a particular number of seats, and once we know the exaggerative index and the proportion of minor-party seats, we can calculate the required share of the two-party vote. But how hard will it be to reach this share' In the above case, for example, how hard would it be for a party to reach 56% of the two-party vote' If past election results suggest that parties rarely do this well, hung parliaments would seem likely. If past election results suggest that parties often do this well, hung parliaments would seem unlikely. So if we can estimate the probability of a party receiving enough votes to win a secure majority, we can quantify how well first-past-the-post performs its crucial function of providing single-party government with a workable Commons majority. Curtice and Steed asked a similar question in the 1980s (though they focused on exaggeration and bias, not exaggeration and minor-party seats). To infer the likelihood of hung parliaments, they used simulations based on past election results (Curtice and Steed, 1982; Curtice and Steed, 1986 and Curtice and Steed, 1986). For example, given electoral geography after the 1979 election, Curtice and Steed estimated that a 3.5-point range in the two-party vote would see hung parliaments, and a further 1.2-point range would see overall majorities of under 20 seats. (Curtice and Steed saw a 20-seat majority as probably large enough to provide a continued majority throughout the parliament even with by-election losses or defections.) Four of the ten elections since 1950 lay in the range of results that would produce hung parliaments, while another two would have given overall majorities of under 20. Curtice and Steed thus concluded that hung parliaments and precarious majorities would be more likely than not. However, Curtice and Steed’s method would not work now: as they recognise, elections have generally become less competitive since their initial test (Curtice and Steed, 1998). At every election between 1950 and 1974, the winning party’s lead in the two-party vote was always under about 8 percentage points (mean 3%), but between 1979 and 2001 the winning party’s lead has always been over 8 points (mean 14%). More elections would thus fit in the current hung parliament ‘danger zone’ than we would now expect, so older election results are less helpful when modelling the future. Curtice and Steed’s approach, although perfectly legitimate at the time, would now overstate first-past-the-post’s failings. Instead, I will assume that in future elections, the major parties could receive anywhere within a 25-point range of the two-party vote, between 37.5 and 62.5%. (The biggest win since 1970 was actually 21 points, in 1983. But inflating the likely range gives more conservative estimates, which is safer if these estimates are being used to question first-past-the-post’s effectiveness.) I also assume that no vote-share within this range is likelier than another. Appendix B details these and several other assumptions. Despite the inevitable simplifications, the method gives a useful first approximation of how well first-past-the-post meets its defenders’ requirement of single-party government with a workable majority. The main finding is that given the current exaggerative index (2.0), and the current proportion of minor-party seats (12%), there is approximately a one-in-four chance of a hung parliament, and a one-in-three chance of a government with a majority of under 25 seats., 2 In other words, assuming that Labour loses its huge advantage over the Conservatives and that some elections become more competitive, there is a distinct risk of a hung parliament, or a small government majority that could create legislative problems and may not last the full parliament. (Problems with small majorities from 1974–9 and 1992–7 were discussed in section 2.) The low exaggerative index and the high proportion of minor-party seats have profoundly affected the translation of votes to seats in British general elections. By contrast, with an exaggerative index of 2.7, and 2% of seats won by minor parties (as seen before the February 1974 election), the probability of a hung parliament would only be about 1-in-30, and the probability of a majority of under 25 seats about 1-in-10. The difference is striking. Changes since the 1970s have clearly endangered the prospects of clear legislative majorities for the winning party. Moreover, if elections enter a consistently more competitive phase (in other words, if it seemed likely that the winning party’s lead in votes would generally be quite low), the probability of hung parliaments and small majorities would increase further. This new method also allows us to assess the claim that the cube law’s decline was the key factor reducing the likelihood of majority government (Curtice and Steed, 1986, p. 216), by controlling for the effect of each factor. Compare two different scenarios: (a) the high exaggerative index from 1970 and the low proportion of minor-party seats from 2001, versus (b) the low exaggerative index from 2001 and the high proportion of minor-party seats seen in 1970. If the cube law’s decline had more effect on majority government than the rise in minor-party seats over the same period, the probability of a hung parliament would be greater under (b) than (a). This is not so: under scenario (a) the probability of a hung parliament is around 1-in-5, but under scenario (b) it is under 1-in-20. Therefore, contrary to Curtice and Steed’s assumption, the increase in minor-party seats has had more direct effect than the cube law’s decline. Of course, we cannot entirely separate the two factors: they work simultaneously, and partly reflect the same causes. But given that both affect the probability of hung parliaments, the above simulation suggests that since 1970, the rise in minor-party seats has increased the risk of a hung parliament more than the cube law’s decline has. This means, of course, that if the proportion of minor-party seats falls in future elections, the chance of majority governments would rise. For example, if minor-party seats dropped to 1980s and early 1990s levels (around 7%, or 45 seats), then the risk of a hung parliament would drop from around 1-in-4 to around 1-in-6. (If a cube law were also restored, this figure would only be around 1-in-10.) But minor-party seats are unlikely to fall much beyond this level. The Liberal Democrats are concentrating their vote, securing more safe seats (Curtice, 2001). Even in the improbable event that the Liberal Democrats split or disappear, the other important minor parties in Northern Ireland, Scotland and Wales are probably entrenched for the moment, especially with devolution giving them greater prominence. Given this method’s inevitable simplifications, the above estimates must be treated cautiously. And obviously they do not predict any individual election. Still, they provide ballpark indications of how well first-past-the-post meets its defenders’ requirement of single-party government with a workable majority. The test also distinguishes the relative effect of the exaggerative index and the level of minor-party seats (inasmuch as they are separable), suggesting that the latter is more important than the former — another notable finding, since commentators who discuss the cube law’s decline (such as Curtice and Steed) largely discount the impact of increased minor-party victories. First-past-the-post’s defenders and critics should take both findings seriously. 4. Decreased cohesiveness of government backbenchers We have seen that hung parliaments and governments with low majorities are likelier than before. Now I shall suggest that low majorities are even more dangerous, because governing parties have become less cohesive, making government harder. Therefore, the electoral system’s declining ability to facilitate secure majorities becomes even more disturbing for its defenders. This intensifies the effect of the first two factors. Defences of first-past-the-post frequently assume a certain view of legislative power: a party enjoying an overall majority can govern effectively without making deals with other parties. But focusing on the governing party’s relationship with other parties may overlook relationships between the governing party’s frontbench and backbench (King, 1976). Tensions within the governing party may therefore affect the majority required to control the Commons securely. If parties were absolutely cohesive, a 10-seat majority would be as robust as a 100-seat majority. This is not the case. So, the less cohesive the governing party, the greater the majority needed for stable government — and the more worrying it becomes if the cube law’s decline and the rise in minor-party seats hamper secure majorities. Standard defences of first-past-the-post would thus be further weakened if small majorities have become not only likelier but also riskier. Indeed, government cross-voting (defined as one or more government MPs voting against the whips’ instructions on a whipped division) has increased in frequency and size since the 1970s as more backbenchers have rebelled more often, and once backbenchers realised that governments would only resign if they lost a vote of confidence, reducing the disincentive to rebel on ordinary votes (Norton, 1978 and Cowley, 2002). Governments with low majorities have thus found life harder since 1970. The 1950, 1951 and 1964 parliaments saw low majorities (6, 17 and 4 seats, respectively), but few backbench rebellions (in 2, 1 and 0.25% of divisions, respectively). Although the 1951 and 1964 parliaments saw backbenchers force occasional policy changes behind the scenes (Seldon, 1987 and Wilson, 1971), these governments mostly got their way. By contrast, in the February 1974 parliament (minority of 34 seats) and in the October 1974 and 1992 parliaments (initial majorities of 4 and 21 seats, respectively), backbench rebellions were much more frequent (in 19, 7 and 13% of divisions, respectively). Not only were changes in government policy behind the scenes more common, but so were legislative defeats, especially in the two 1974 parliaments ( Schwarz, 1980 and Butler and Kavanagh, 1997). Granted, other factors, like increased discord over Europe, are also relevant. But changing parliamentary behaviour since 1970 has certainly influenced backbench dissent, creating particular problems for governments with low majorities. Of course, a future government with a low majority will not necessarily endure these difficulties. But it is distinctly possible. Other things being equal, if the government backbenches are now less cohesive, then the frontbench will face more hardship, and it becomes particularly important to avoid low Commons majorities. Yet we have just seen that low Commons majorities are increasingly likely, due to the cube law’s decline and the rise in minor-party seats. First-past-the-post’s frailties intensify. This deepens our understanding of first-past-the-post’s contingency on external factors. Curtice and Steed repeatedly and rightly argue that the translation of votes to seats in SMP elections is not a fixed, structural outcome, but is contingent on electoral geography and electoral behaviour, changes in which have weakened first-past-the-post. I would now add that we must consider not only the translation of votes to seats, but also the translation of seats to power. This is contingent on parliamentary behaviour, and changes here have also weakened first-past-the-post. First-past-the-post is somewhat culpable for this development. In countries where most citizens vote on party lines, SMP systems facilitate backbench unity: rebellious factions or individuals are unlikely to leave the party, as their electoral prospects are slim, so their blackmail potential over party leaders is weaker and party cohesion is stronger. But these incentives can be overcome (consider America), and in Britain their influence has clearly weakened. Admittedly, first-past-the-post is not often supported for promoting party cohesion. One prominent adherent even praises increased backbench rebelliousness and attacks party dominance in some countries with PR elections, like Germany (Norton, 1992). But perhaps this argument does not recognise how dissenting backbenchers undermine other defences of first-past-the-post. If more government backbenchers are now more willing to rebel, then other things being equal, governments with smaller majorities — which are becoming more likely — will have more problems. In effect, the Commons majority necessary for stable and productive government has increased. Increased backbench rebelliousness may not be directly attributable to first-past-the-post, but has nonetheless aggravated its other difficulties. Standard defences of first-past-the-post therefore make assumptions about the translation of votes to seats and the translation of seats to power. The former is contingent on the number and strength of parties, electoral geography and electoral behaviour; the latter is contingent on legislative behaviour. Changes in the former have made small majorities more likely; changes in the latter have made small majorities more hazardous. The pressure on first-past-the-post mounts. 5. The increase in pro-labour bias Bias has been widely discussed by analysts of first-past-the-post, especially since the 1992 election (see in particular Curtice and Steed, 1992 and Johnston et al., 2001). This section extends their analysis by applying an alternative method for investigating bias, supplementing their focus on hypothetical outcomes by examining actual results. I also query some overly pessimistic forecasts about the Conservatives’ disadvantage in future elections. 5.1. What is bias and how much is there' Bias means that one major party wins more seats than we would expect given its share of the vote, while the other party wins less seats. At its simplest, this means that both parties should win the same number of seats when they have the same number of votes — so if both parties have the same vote but one party wins more seats than the other, there is a bias in its favour. The level of exaggeration determines the number of seats that we would expect a party to win. So roughly, bias adds to or subtracts from the level of exaggeration, irrespective of whether the advantaged party is the winner or loser. For example, with a cube law and no bias, whichever party has a 1% lead in votes would have a 3% lead in seats. But assuming a cube law with a 1% bias to Labour, a 1% lead in the Conservative–Labour vote would become a four percent lead in seats if Labour won, but only a two percent lead if the Conservatives won. Recent election results have been biased to Labour. This helps to explain why the Conservatives’ 7.5 percentage point lead in votes in 1992 only gave a 21-seat overall majority in the House of Commons, while Labour’s 9-point lead in 2001 gave a 167-seat majority. Another example of pro-Labour bias was in February 1974, when Labour finished first in seats despite coming second in votes. Bias derives from any of three main sources. (As will be shown shortly, understanding the different sources of bias is normatively important.) Size bias arises if one party’s votes are concentrated in seats with fewer voters. If party A and party B both have 100 votes, but party A wins 100 seats each containing 1 voter, while party B’s 100 voters are crammed into 1 seat, then both parties will have the same number of votes but party A will have far more seats. Distribution bias arises if one party’s vote is distributed more efficiently. For example, if the two parties have the same number of votes but party A wins many marginals while party B amasses unnecessarily large majorities in a few safe seats and wastes many votes in party A’s marginals, then party A will again win more seats, other things being equal. Minor-party wins bias arises if one party loses fewer seats to minor parties than the other party does. For example, if parties A and B have the same number of votes, but party B comes second in many seats won by minor parties, these votes are wasted, so party A will win more seats, other things being equal. Size bias can in turn be decomposed into four further components. Each reflects a different way in which a party can concentrate its vote in seats with fewer voters. (In other words, if all seats had been the same size, we assume that the party would have won fewer seats.) National constituency bias arises if one party’s vote is concentrated in the smaller seats in Scotland and Wales. In 2001, for example, Boundary Commission rules meant that the average Scottish seat had 55 000 registered electors while the average English seat had 70 000 . If one party is relatively strong in Scotland, then other things being equal it will win more seats than if Scottish seats were larger and thus less numerous. Subnational constituency bias arises if, after these national quotas have been taken into account, one party’s vote is still concentrated in smaller seats, like urban seats (which over time generally lose voters to the suburbs and beyond). Turnout bias arises if one party’s vote is concentrated in seats with lower turnout. Minor-party votes bias arises if one party’s vote is concentrated in seats with more minor-party voters. Johnston et al. (2001) analyse these different biases comprehensively. Bias may worry first-past-the-post’s defenders because a level playing-field is apparently denied to the two major parties. This may be viewed as inherently unfair, but more likely what matters are the results: bias can produce unworkably small majorities, obscure accountability, or let the second-placed party come first in seats. This weakens the ‘removal van’ theory of democracy underlying two recent defences of first-past-the-post (Popper, 1988 and Pinto-Duschinsky, 1999a). If first-past-the-post facilitates accountable governments which can be removed if enough voters want them out, then both major parties should be equally accountable. How bias is evaluated depends partly on how it is calculated. The most common method is to apply a ‘uniform swing’ from one major party to the other in all constituencies until both parties have the same number of votes nationally (Butler, 1951). If one party would then win more seats than the other, there is a bias in its favour (see especially Curtice, 2001 and Johnston et al., 2001). Analysts using uniform swing rarely test its validity in this context. I shall test its validity by seeing how well it predicts the outcome of past elections, using uniform two-party swing., 3 We know the winning party’s actual majority over the losing party, and can compare this to the majority that would have resulted if swing from the previous election had been uniform. Between 1959 and 1970, the mean absolute error in predicting the correct outcome is only 11 seats, and uniform swing was within 10 seats of the actual majority at three of the four elections. Since 1974, however, such accuracy has been achieved only once and the mean error is almost 30 seats. At its worst, in 1997, uniform swing underestimated Labour’s majority by some 70 seats. (For total-vote swing, the same accuracy is found for elections up to 1970, the mean error since 1974 is 23 seats, and the worst error is 54 seats in 1997.) So, the applicability of uniform swing has weakened in recent years. Nonetheless, it is usually a satisfactory assumption, and renders an easily calculable indication of the amount of bias that might have occurred if the two major parties finished on level terms. The integrated method, discussed in section 2 (and detailed in Appendix A), has one advantage over uniform swing: it focuses on actual not hypothetical results. This helps when assessing bias at past elections: we uncover what actually happened at the election, rather than what might have happened if the election had finished with the two parties level in votes. Both methods give different information; both are useful. Note that adherents of uniform swing sometimes make inferences from what bias might have been like under hypothetical conditions to what bias actually was at the election (for example, Rossiter et al., 1999, pp. 146-7). Strictly speaking, this inference is invalid — only the integrated method provides such information — but in practice the difference is usually not very important. Estimates of bias from the uniform swing and integrated methods are depicted in Fig. 3. Both methods tell essentially the same story: Labour suffered from bias in the 1950s and 1960s but later broke through, and was significantly and increasingly advantaged at the 1992, 1997 and 2001 elections. [pic] [pic](6K) Fig. 3. Increasingly pro-Labour bias. Notes: uniform-swing figures are based on total-vote (‘Butler’) swing, the most commonly used form of swing in this context. [pic] One difference between the two methods stands out. Uniform swing suggests that Labour’s (hypothetical) advantage has existed since 1992. Curtice (2001, p. 243) thus writes that ‘the 2001 election was the third in a row at which the electoral system has displayed a significant … bias in Labour’s favour.’ But the integrated method suggests that the 2001 election was the sixth such election in a row, and that whetever the hypothetical situation, Labour has actually benefited from bias at eight of the nine elections since 1970. This may worry first-past-the-post’s defenders more than the orthodox picture created by uniform swing. Table 1 charts the integrated method’s estimates of the total amount of bias, and its main components. Numbers refer to the advantage in one party’s majority over the other. For example, in 2001 Labour had a 246-seat majority over the Conservatives, but without bias this figure would have been around 166 seats. [pic] Table 1. Estimates of bias at past elections [pic] Notes: Negative figures indicate bias to Labour, positive to the Conservatives. Estimates of distribution and size bias are based on British seats contested by both major parties and won by a major party. Estimates of minor-party wins bias are based on British seats won by minor parties. Figures for the components of size bias may not add up to the total amount of size bias because of interactions between the components (not reported). [pic] Labour now benefits because it has overcome its traditional weakness of amassing inefficiently large majorities in safe seats (distribution bias), because it tends to win seats with fewer major-party voters (size bias), and because Liberal Democrat victories have disproportionately come at the Conservatives’ expense (minor-party wins bias). Interestingly, then, the Liberal Democrats are disrupting first-past-the-post in three ways: they have weakened the cube law (section 2), they are mainly responsible for the rise in minor-party seats ( section 3), and they contribute to the increasingly pro-Labour bias. First-past-the-post’s future, it seems, depends partly on the Liberal Democrats. 5.2. How has bias affected past election results' Three past elections in the period studied are potential problems for first-past-the-post’s adherents: February 1974, where the ‘wrong’ party won, and October 1974 and 1992, where the governments’ majorities were smaller than they would have been without bias, exacerbating governmental weakness. In February 1974, Labour took 37% of the total vote and 301 seats, while the Conservatives took 38% of the vote but only 297 seats. This is often called a ‘perverse’ result, as the party with most votes came second in seats. About one in five SMP elections worldwide are perverse (Powell, 2000, p. 130). We might in fact excuse the February 1974 result: tactical voting by Labour supporters helped several Liberal candidates oust Conservative MPs, artificially reducing Labour’s vote by actions intended to help Labour win the election (Steed, 1974). Even some fierce critics of first-past-the-post thus refuse to attack this result ( Bogdanor, 1981, p. 179). Similarly, when the Conservatives won perversely in 1951, uncontested constituencies artificially lowered their vote without affecting their seat tally ( Curtice and Steed, 1998). Whether or not one sees the 1951 and February 1974 results as perverse, the two small majorities are certainly awkward for first-past-the-post’s adherents. In October 1974, a pro-Conservative bias, primarily through inefficiently distributed Labour votes, reduced Labour’s majority to just 4 seats, despite a 5-point lead in the two-party vote. This slender majority soon disappeared, and the government never fully controlled the Commons. In 1992, the Conservatives’ majority could have been as high as 70 without bias, other things being equal. According to John Major, such a majority would have made his life ‘immeasurably easier’, not least because he could have passed the immensely troublesome Maastricht Bill in ten weeks (Major, 2000). ‘With more seats, I would have had the ability to act with force against those in the party who chose to rebel’ ( Major, 1999, p. 307). Again, note the double whammy: bias was exacerbating the rebelliousness of Conservative MPs discussed in section 4. The detrimental impact of bias has been observed for 1992 (Curtice, 2001) but not for October 1974. These two elections, which also suffered from insufficient exaggeration ( section 2), are problematic for first-past-the-post’s defenders. Of course, bias was not the only offender: Major’s tactics, for instance, were hardly faultless. But the system clearly did not work as desired, and its supporters have stayed strangely silent. 5.3. How might bias affect future election results' To evaluate bias fully we must address the pessimistic claims frequently made about the Conservatives’ prospects of overcoming this bias. Assuming uniform swing, the Conservatives would be a staggering 130–140 seats behind Labour when the two parties have the same vote (Norris, 2001; Curtice & Steed, 2002; Johnston et al., 2002 and Rallings and Thrasher, 2002). But as demonstrated above, uniform swing does not always predict election results reliably. I would therefore suggest that uniform swing simulations should act mainly as benchmarks, giving a baseline estimate which can be amended depending on possible non-uniformities in swing. For example, after the 1997 election Curtice and Steed suggested that the Conservative vote might recover faster in its safer areas than its weaker areas, such that uniform swing could overestimate the likely bias to Labour (Curtice and Steed, 1997, pp. 315–8). Unfortunately, aside from Curtice and Steed, the other scholars just cited do not use uniform swing as a benchmark. Even Ron Johnston and co-workers, geographers who rightly uphold the importance of geography in elections, have not examined how bias might vary under different geographies of swing. And some scholars actually use uniform swing without saying so. One writes: ‘To achieve an overall majority, the Conservatives require at least a ten-point lead in votes’ ( Harrop, 2001, p. 311). In fact the Conservatives only require a ten-point lead assuming uniform swing. There are, indeed, good reasons to think that the hypothetical bias of 130–140 seats is too pessimistic. National and subnational constituency biases should fall after the next reapportionment and when Scottish over-representation is removed (Curtice and Steed, 2002). Minor-party wins bias, currently the largest source of anti-Conservative bias, should also fall if a Conservative recovery takes seats from the Liberal Democrats ( Russell et al., 2001). However, if the Liberal Democrats are becoming harder to beat ( Curtice, 2001), the effectiveness of Conservative campaigning in these seats could be crucial. Hypothetical estimates of distribution bias may also be excessive. Assuming uniform swing, Labour’s advantage from distribution bias, currently around 30 seats, would actually increase as the Tories near Labour’s vote, to 75 seats (Johnston et al., 2002). But a uniform Conservative recovery is hardly certain, as just noted. Only a striking change in electoral geography will completely overturn Labour’s distributional advantage, but the 75-seat estimate may well be too large. So, uniform swing predictions of a bias of 130–140 seats to Labour in a future tied election could be overly gloomy. However, Patrick Dunleavy was probably too optimistic in suggesting, after the 1997 election, that pro-Labour bias ‘could vanish as suddenly as it appeared’ (Dunleavy, 2000, p. 138). Table 1, produced using a method not then available to Dunleavy, suggests that Labour’s advantage arose gradually, not suddenly. Dunleavy also falls into the common trap of considering bias as a whole, but the above discussion shows that some components are unlikely to disappear. In summary, while several psephologists are probably too pessimistic about Labour’s future advantage from bias, it is unlikely to disappear over the next decade without dramatic changes in the party system. First-past-the-post’s adherents must therefore accept that over the next decade or two, the Conservatives could win one or even two elections in votes but still come second in seats. Perhaps these perverse results could be explained away, as noted earlier. But in practice such arguments would convince few. The popular and media outcry might virtually force politicians to implement electoral reform — maybe as an election-winning manifesto commitment by the Conservatives. So, even if some of the alleged bias is more apparent than actual, the public and political pressure for electoral reform may be real enough to floor first-past-the-post. 6. Conclusion This paper has examined four challenges facing the British electoral system: the decline of the cube law, the increase in minor-party seats, the decreased cohesiveness of legislative majorities, and the rise in pro-Labour bias. SMP electoral systems do not have fixed outcomes: the creation of workable legislative majorities depends strongly on external factors like electoral geography, electoral behaviour and legislative behaviour. Since at least the 1970s, these outside influences have made first-past-the-post less likely to do what its defenders seek. This has sometimes been masked by the main opposition party’s poor performance, but closer elections should again accentuate the system’s negatives more clearly. First-past-the-post’s defenders may justly claim that it still works more often than not. Quite simply, the British electoral system has done its job at most elections. Its critics should acknowledge this. But its supporters must be more honest about its infirmities. Inconclusive elections are more likely (and have already been seen in the 1970s), minor parties have had more influence, and governments have lacked firm legislative control for over a quarter of the time-period since 1970. First-past-the-post’s supporters often discuss such outcomes as weaknesses of PR elections; evidently first-past-the-post is not immune either. Although all four developments discussed in this paper are exogenous to first-past-the-post, it is still partly culpable. SMP systems are often defended for promoting desirable outcomes: for example, Duverger’s (1959) psychological factor restricts minor parties, helping accountability. But first-past-the-post’s self-sustaining incentives are working less well today. True, all electoral systems are contingent on exogenous factors: for example, some PR elections in newly democratised countries were less proportional than hoped, because so many parties stood. But SMP seems particularly dependent on exogenous factors. The geography of voting behind partisan bias, for example, would not weaken the Dutch or Israeli PR systems. Certain problems for first-past-the-post are, indirectly, its own fault. Ultimately, the most important conclusion is that regardless of the intellectual arguments for first-past-the-post, its gradual decay makes electoral reform more likely in practice. Although the future rarely resembles the past, the four developments described here raise the possibility of two triple whammies for first-past-the-post and one quadruple whammy. One triple whammy sees the decline of the cube law and the rise in minor-party seats interacting to produce a government whose small majority is then destabilized by rebellious government backbenchers. Another triple whammy sees the rise in pro-Labour bias leaving the Conservatives first in votes but second in seats; the decline of the cube law and the rise in minor-party seats could produce a hung parliament with the Liberal Democrats holding the balance of power and insisting on PR as the price for supporting the government. Popular and media outcry against the electoral system might impel the Conservative and/or Labour party leaders to choose electoral reform. The quadruple whammy for first-past-the-post — the nightmare scenario for its defenders — would be the first triple whammy followed by the second: a government with a small majority struggling in the Commons for 5 years, followed by an indecisive election with the ‘wrong’ party coming first and the Liberal Democrats holding the balance of power in a hung parliament. Pressure for electoral reform might then be hard to resist. So, whether or not the theoretical case for first-past-the-post has been damaged by the developments discussed above, first-past-the-post’s burgeoning weaknesses make it practically more likely