The voting behavior of members of the British House of Commons is principally determined by their

Elect Stud. 2012 Dec; 31(4): 750–763.

Abstract

This paper examines electoral accountability after the 2009–10 UK expenses scandal. Existing research shows that Members of Parliament (MPs) implicated in the scandal fared only marginally worse in the election than non-implicated colleagues. This lack of electoral accountability for misconduct could have arisen either because voters did not know about their representative's wrongdoing or because they chose not to electorally sanction them. We combine panel survey data with new measures of MP implication in the expenses scandal to test where electoral accountability failed. We find that MP implication influenced voter perceptions of wrongdoing more than expected. In contrast, constituents were only marginally less likely to vote for MPs who were implicated in the scandal. Electoral accountability may therefore be constrained even when information about representative misconduct is easily available and clearly influences voter perceptions.

Keywords: Accountability, Expenses scandal, Information, Members of Parliament, Voting behaviour

Highlights

► We examine voters' beliefs about their MP's involvement in the expenses scandal. ► Voters' perceptions were surprisingly responsive to public information. ► Yet voter ignorance was considerable, while partisan bias shaped perceptions. ► Voters mostly failed to punish incumbents based on these beliefs. ► Accountability can fail even when information is available and voters are informed.

1. Introduction

In a representative democracy, regular elections provide a potential mechanism for voters to hold politicians accountable for their performance by enabling the removal of office-holders that are under-performing (Key, 1966; Ferejohn, 1986; Besley, 2006). Such electoral accountability has been shown to exist for a variety of aspects of representatives' record: for instance, their handling of the economy (see e.g. Powell and Whitten, 1993; Lewis-Beck and Stegmaier, 2000; Samuels and Hellwig, 2010) or their policy stances as expressed in votes in the legislature (Canes-Wrone et al., 2002; Ansolabehere and Jones, 2010; Vivyan and Wagner, 2012). Other scholars have asked whether voters also hold representatives electorally accountable for unethical or corrupt behaviour (Rundquist et al., 1977; Peters and Welch, 1980; Chang et al., 2010). To do this, voters need to, first, update perceptions about the politician's conduct if he or she is revealed to have engaged in misconduct, and second, take electoral choices based on these perceptions (Ferejohn, 1986; Przeworski et al., 1999; Besley, 2006).

This paper uses the case of the 2009–10 UK expenses scandal to conduct a unique test of the functioning of these two necessary steps for electoral accountability for misconduct. During the expenses scandal the media and official investigations exposed many Members of Parliament (MPs) as having made improper claims under parliamentary allowances schemes. Thus, a large number of sitting legislators were publicly implicated for similar types of misconduct in a relatively short space of time. In other words, we can exploit natural variation in whether or not legislators were implicated while holding the type of scandal and the national political circumstances relatively constant. The UK expenses scandal therefore provides a particularly advantageous setting to test the functioning of electoral accountability for misconduct.

Yet this scandal also presents us with a puzzle: even though voters felt strongly about expenses-related misconduct by their MPs, electoral accountability for this misconduct was severely limited. On the one hand, survey evidence suggests that the scandal was highly salient to voters in Britain. According to the 2010 British Election Study (BES) pre-election survey carried out nine months after the scandal first erupted, 93 per cent of British voters had heard of the scandal, while over 90 per cent still agreed or strongly agreed with the statement that the scandal made them very angry. Moreover, 80 per cent of voters agreed or strongly agreed that MPs implicated in the scandal should resign.2 On the other hand, aggregate evidence indicates that at the ballot box voters largely did not punish MPs who had been publicly implicated in the scandal. Among sitting MPs who stood for re-election in 2010, the electoral cost of implication in the scandal has so far been estimated at around 1.5 per cent (Curtice et al., 2010; Eggers and Fischer, 2011; Pattie and Johnston, in press).

Why, in a case such as this, is electoral accountability for misconduct so limited? One potential explanation is that the MPs exposed as having made the most egregious expenses claims anticipated electoral punishment and therefore declined to re-stand for election at the 2010 general election. Yet existing analysis shows that even among those MPs most severely publicly implicated in the expenses scandal, over 50 per cent chose to stand for re-election in 2010 (Pattie and Johnston, in press). Thus, strategic retirement cannot entirely explain the apparent lack of electoral accountability for expenses-related misconduct in 2010. Instead, we have to consider why the accountability link may function poorly even when representatives whose wrongdoing has been exposed decide to run for office again. We argue that this can occur for two reasons: either voters do not form or update their beliefs about representatives based on new public information concerning their conduct, or citizens do not cast their vote in such a way that politicians they believe are guilty of misconduct are less likely to retain office. Put simply, electoral accountability for misconduct can fail at either of either the perception or the sanction step. In this paper, we test these two explanations empirically.

In doing so, we address key theoretical debates in the literature. Several existing theories would predict that accountability fails at the perceptions step. Specifically, voters may either not know about their representative's behaviour (Delli Carpini and Keeter 1996; Holmberg, 2009) or develop incorrect perceptions based on personal or partisan biases (Campbell et al., 1960; Fenno, 1978; Peters and Welch, 1980; Ferejohn, 1990). Other accounts would emphasise that, even if they have the correct beliefs, voters often have little incentive to act on these beliefs at the ballot box. This can happen when voters are not willing or able to use their vote to punish representatives for wrongdoing, for example if they think the balance of power in parliament is more important (Rundquist et al., 1977). Of course, institutional features of a political system also affect the extent to which voters have to balance different considerations in determining their vote choice. For example, electoral systems can influence the relative importance of candidate versus party performance in voter decisions (Carey and Shugart, 1995; Persson and Tabellini, 2000; Persson et al., 2003; Kunicová and Rose-Ackerman, 2005; Chang and Golden, 2006).

Testing the separate functioning of the perceptions and sanctioning steps of electoral accountability for misconduct is generally difficult, but our empirical strategy with regard to the UK expenses scandal allows us to overcome many of these difficulties. First, as noted above we can exploit natural variation in the implication of politicians in the scandal while holding relatively constant the timing and nature of the scandal. To our knowledge, this feature of a scandal is shared only by the US House Banking Scandal (Dimock and Jacobsen, 1995; Ahuja et al., 1994; Banducci and Karp, 1994; Clarke et al., 1999).

Second, voter surveys only rarely ask detailed questions about voter perceptions of wrongdoing. Fortunately, the BES measured voter beliefs regarding their MP's expenses claims shortly after the publication of the official report investigating the scandal. We supplement these voter-level survey responses with new measures of the actual public implication of respondents' MPs in the expenses scandal. As a result, we are able to separately and directly test whether constituent perceptions of MP misconduct respond to public information about that MP, and then whether constituents electorally sanction MPs based on these perceptions.

Third, cross-sectional surveys measure voter affect towards the MP and his or her party and voter perceptions of MP involvement in the scandal at the same point in time. This makes it difficult to establish to which extent affect influences perceptions and vice versa. By using the 2005–10 BES panel study, we can estimate and control for the influence of respondent's prior affect toward MPs and their party using measures of these dispositions taken from pre-scandal waves, which are not affected by respondent exposure to the scandal. A further advantage of the panel survey data is that we can use direct measures of vote choice from post-election waves in 2005 and 2010.

Our key results are as follows. First, in terms of the perceptions step of electoral accountability, constituent perceptions of MP expenses-related misconduct were clearly responsive to publicly available information about MP behaviour. British voters were substantially more likely to think that their MP overclaimed on expenses when that MP was publicly implicated in the expenses scandal. For example, we estimate a typical voter to be 22 percentage points more likely to think that their MP overclaimed on expenses if their MP was, in fact, publicly implicated. Nevertheless, we also find, as expected, that levels of ignorance (i.e., the numbers of voters who simply “don't know” whether or not their MP overclaimed) were considerable and that partisan biases played an important role in influencing perceptions. Our findings are robust to various alternative specifications and measures of MP implication in the expenses scandal. Thus the lack of electoral accountability for expenses-related misconduct in 2010 cannot be attributed solely to a failure in the functioning of the perceptions step of electoral accountability.

Second, we find evidence that the sanction step of electoral accountability functioned only weakly. Regression analysis of vote choices shows that, controlling for their political predispositions, constituents were only marginally less likely to vote for their incumbent MP when they perceived that MP to have overclaimed on expenses. In terms of predicted probabilities, a typical voter who thinks that their MP overclaimed on expenses was five per cent less likely to vote for him or her than a voter who did not think their MP overclaimed. While the magnitude of this effect of perceptions on vote choice is certainly non-negligible, it is less strong than the effect on perceptions of public information about MP misconduct. Put simply, if an MP was publicly implicated in the expenses scandal, people were reasonably likely to change their perceptions of the MP's conduct because of this new information, but were less likely to change their vote because of their perceptions.

These findings imply that voter perceptions of their representatives' misconduct are not completely dominated by ignorance of misbehaviour or by biases based on partisan or MP-specific predispositions. They also suggest that the problem of limited accountability can stem from sources other than a lack of information: few voters in the UK chose to punish their MPs for wrongdoing in a situation where this wrongdoing was highly salient and information about it was publicly available and clearly influenced voter perceptions.

This paper is structured as follows. In the next section, we expand on the two steps necessary for electoral accountability for misconduct and discuss factors that may aid or hinder their operation in practice. Then we describe our new measure of whether or not each MP was publicly revealed to have made improper expenses claims. Following this, we empirically examine the influence of public information on perceptions of misconduct, before turning to the consequences of perceptions of misconduct for electoral choices. We conclude by summarizing our findings and considering their implications.

2. Electoral accountability for misconduct: perceptions and sanction

Before electoral accountability for misconduct can occur, the wrongdoing by representatives must be exposed. Often, the media play an important role in investigating, uncovering and publicizing misbehaviour. In the case of the expenses scandal, the Telegraph newspaper had exclusive access to leaked details of MPs' expenses claims, and beginning in May 2009, it ran stories on excessive or questionable claims for several weeks. The scandal quickly engulfed a large number of MPs and received widespread media attention (for more details, see Kelso, 2009).

In some cases, public information about representatives' misconduct may stem from an official investigation or governmental body, not just investigative media reporting. For example, the 1992 US House Banking Scandal was triggered when the General Accounting Office issued a report detailing the misbehaviour of members (Dimock and Jacobsen, 1995). In the case of the expenses scandal, the House of Commons reacted to relentless media revelations by commissioning Sir Thomas Legg to investigate all MPs' claims in detail. Legg's final report, published in February 2010 (House of Commons Members Estimate Committee, 2010), focused on overly excessive claims made within the Additional Costs Allowance, a scheme that allowed MPs to claim money to pay for and maintain a second home, either in their constituency or in London.3 Nevertheless, it is of course unlikely that many voters consult the official report into the scandal directly. Rather, awareness of misconduct exposed by an official investigation will tend to spread via the national or local media, campaign groups, local competing candidates and the politicians themselves (Arnold, 1993).

2.1. Perceptions of misconduct

For accountability to exist, voters need to be aware of publicly available information about their representatives' conduct, but the literature suggests two main reasons why such awareness may be low. First, voters can simply be ignorant of their representative's wrongdoing. Gathering and processing such information is generally costly (Ferejohn, 1990). Moreover, institutional settings vary in the extent they provide voters with incentives to become informed (Stevenson and Vonnahme, 2009; Holmberg, 2009). In Britain, these incentives are low and costs high: the political system is characterized by strong party control of representatives in the legislature and low levels of local campaign spending and local media coverage (Benedetto and Hix, 2007; Cain et al., 1987; Kam, 2009; Pattie et al., 1994; Pattie and Johnston, 2004). In addition to a lack of knowledge, voters may also have biased perceptions. To reduce the costs associated with information-gathering, individuals make use of heuristics when forming an opinion about their representative (Ferejohn, 1990). A simple decision rule allows voters to make judgments about their representative even while remaining relatively ignorant about the representative's actual behaviour. Two such rules of thumb may be particularly influential in the British context: affect towards an MP and affect towards the MP's party. Voters may be less likely to think that their MP misbehaved if they are positively predisposed towards him or her. This positive predisposition may exist because of the MP's constituency and parliamentary work (Cain et al., 1987) or because of their attachment to the MP's party. Recent findings from the UK (e.g. Marsh and Tilley, 2010; Tilley and Hobolt, 2011) indicate that theories of partisan perceptual bias (Campbell et al., 1960) travel well.

2.2. Sanctioning of misconduct

Despite the problems of ignorance and bias, some voters may nevertheless update perceptions of their representative's conduct in light of public revelations. But voters also then have to decide whether this perceived misconduct should influence their vote choice. Some voters will believe that other concerns, such as the balance of power in parliament, are more important than the misbehaviour of their representative (Rundquist et al., 1977). Other voters will decide to punish a representative who they believe has engaged in wrongdoing. In the latter case, we can speak of electoral accountability for misconduct: publicly available information about an MP's misconduct influences voter perceptions, which in turn influence electoral choices.

The electoral effects of malfeasance have been studied extensively, though mostly in the US and only to a lesser extent in other countries. In the US, studies have found that congressional incumbents charged with corruption have lost, on average, between six and eleven percentage points at the polls (Peters and Welch, 1980; Welch and Hibbing, 1997). Furthermore, the impact of the aforementioned US House Banking Scandal on the vote share of Members of Congress has been studied extensively (Ahuja et al., 1994; Banducci and Karp, 1994; Clarke et al., 1999). Perhaps most notably, Dimock and Jacobsen (1995) found evidence that incumbents were electorally punished for their misbehaviour. Outside of the US, in a study of Italian legislators across Italian post-war elections, Chang et al. (2010) show that judicial accusations of wrongdoing affect re-election rates, but only when there is sufficient national media attention on such wrongdoing.

In the UK, Farrell et al. (1998) study the relationship between corruption and vote loss and find that candidates involved in a scandal lose upwards of five percentage points, a magnitude similar to that in the US. The specific electoral effects of the 2009–10 expenses scandal have been studied using aggregate constituency-level data (Curtice et al., 2010; Eggers and Fischer, 2011; Pattie and Johnston, in press) and survey data with a pre- and post-election panel (Pattie and Johnston, in press). These studies find that involvement in the scandal had a small negative effect – around 1.5 percentage points – on the vote share of an incumbent MP. We add to the understanding of the impact of the expenses scandal by using voter-level panel data to explicitly and separately test the link between MP involvement and constituent perceptions on the one hand and that between constituent perceptions and vote choice on the other.

3. MP involvement in the expenses scandal

Our measure of whether a British4 MP who sat in the 2005–10 Parliament was publicly implicated in the expenses scandal is constructed based on two sources: the Legg report together with the Daily Telegraph and Sunday Telegraph.5 We coded an MP as being ‘implicated’ by Legg if they were formally asked to repay money in his report. We coded an MP as being ‘implicated’ by the Telegraph if the specifics of the MP's claim were discussed in a Telegraph article (MPs were not coded as ‘implicated’ by the Telegraph if their claims were only mentioned in a general article on the scandal).6 We focus on the Telegraph as a media source because it had sole access to the expenses records and acted as a gatekeeper on public information; prior to the publication of the Legg report, any MP publicly implicated was therefore first mentioned by the Telegraph.

Our final implicated variable equals one if an MP was coded as being implicated by either the Telegraph or the Legg report, and zero otherwise. This might be considered a blunt measure of implication, but we choose this as our primary measure because it realistically captures whether or not an MP was publicly exposed as having made dubious claims. Both the Legg report and the Telegraph exposed misconduct, but who was implicated in each source differed. The Legg report did not encompass all potential types of expenses-related misconduct. For instance, Geoffrey Clifton Brown, who was the Conservative Shadow Minister for International Development, allegedly ‘flipped’ (i.e. changed) his second home designation from London to his Gloucestershire home and, at the same time, bought another countryside home for almost three million pounds. His case garnered tremendous media attention and ridicule, but he was not asked to repay any money by Legg because his claims were within the rules. In contrast, Alun Michael's claims were not detailed in the Telegraph, but he was ordered in the Legg report to repay over £19,000 for invalid mortgage claims. These types of cases illustrate the importance of using both sources to capture public implication in the scandal. Though we believe that our combined measure best captures public implication, we also present robustness checks using alternative measures.

Overall we have 587 MP-level observations on our implicated variable.7 Of these, just over two thirds (418) are coded as implicated and just under a third (169) are not. This illustrates a useful feature of the expenses scandal for empirically examining electoral accountability for representative misconduct: the fact that a large (though not overwhelming) proportion of sitting MPs was publicly exposed for the same types of misconduct in a short time window.

4. Constituent perceptions of misconduct: data and analysis

To test the first step necessary for electoral accountability, we combine our measure of MP involvement in the expenses scandal with BES panel data. This allows us to examine whether constituent perceptions of MP conduct are influenced by public information concerning this conduct.

4.1. Data

We use survey data from four waves of the BES 2005–10 internet panel study8: the first wave, carried out in March and April 2005, before that year's general election; the third wave, from May 2005, after that year's election; the fifth wave, from June 2008; and the seventh wave, from late March/early April 2010, before the 2010 election. Based on their location, defined in terms of 2005 general election boundaries, each survey respondent was matched to an MP from the 2005–10 Parliament and assigned the corresponding implicated score for that MP.9 Our sample consists of any survey respondent located in one of the 587 constituencies for which there is a non-missing value on our MP implicated variable. Our resulting data set contains observations on respondents from 579 of the 587 constituencies for which we measure our implicated variable, with an average of 5 respondents per constituency. Of the 3218 respondents in this data, almost 70 per cent (2245) were represented in the 2005–10 Parliament by an MP who was implicated in the scandal by the Telegraph or by Sir Thomas Legg. The remaining 30 per cent (973) had an MP who was not implicated in the expenses scandal.

To measure respondent perceptions regarding the propriety of the expenses claims made by their MP, we utilize their responses to a question from the seventh wave (March/April 2010). Respondents were asked whether they had heard or read about the expenses scandal. Those answering ‘yes’ were then presented with the following question:

Now, thinking about the MP in your local constituency, has he or she claimed expense money to which they are not entitled?

The response options were ‘yes’, ‘no’ and ‘don't know’. These responses are our main dependent variable, which measures whether or not a respondent perceives their MP to have been guilty of overclaiming on expenses.10 The item was fielded just before the 2010 election campaign got underway but after the February 2010 Legg report. As a result, it measures constituent beliefs about MP's expenses claims shortly after the last public revelations concerning the scandal.

4.2. Descriptive analysis

Table 1 breaks down respondent perceptions of their MP's expenses claims according to whether or not the MP was publicly implicated in the scandal or not.11 The first notable feature of this table is that almost half of the 3218 respondents (44.9 per cent) ‘don't know’ whether or not their MP overclaimed. That such a high proportion of respondents feel unable to form a clear opinion on the claims of their MP would appear to bear out expectations that a large proportion of British voters are relatively uninformed the conduct of their local MP. This of course places an immediate limit on the operation of the perceptions step necessary for accountability for misconduct.

Table 1

Voter perceptions of MP behaviour.

It also appears from Table 1 that voters who do form an opinion (i.e. do not answer ‘don't know’) have a tendency to attribute innocence to their own MP, a finding that echoes Fenno's (1978) well-known paradox that American voters tend to disapprove of Congress but trust their district's representative. The bottom row shows that of the 1772 respondents who state an opinion, over half (55 per cent) think their MP did not overclaim.

But despite this, there is also clear evidence of an association between public implication of an MP in the scandal and respondents' perceptions. Of respondents whose MP was not implicated, 40.8 per cent think their MP did not overclaim, compared to 12.6 per cent who think their MP did overclaim. In contrast, among those respondents whose MP was implicated, the corresponding proportions are 25.7 per cent and 30.1 per cent, respectively. This still means that many people whose MP was implicated still believed he or she was innocent, but respondents were nevertheless more likely, by a magnitude of 17.5 percentage points, to think that their MP had overclaimed when the MP was publicly implicated in the scandal.

4.3. Regression analysis

Regression analysis can provide firmer evidence for the finding that constituent perceptions of MP conduct differ depending on whether the MP was publicly implicated in the scandal. We model our main dependent variable using multinomial logit since it has three response options (‘overclaimed’, ‘did not overclaim’ and ‘don't know’).12 Standard errors are clustered by constituency to account for the fact that respondents are nested within parliamentary seats.13

Our key explanatory variable is our measure of whether or not a respondent's MP was implicated in the expenses scandal. We also control for a number of other variables that might plausibly influence respondent perceptions. The most important control variables measure partisan and MP-related voter predispositions. Here, the existence of the panel survey proves very useful: we can use respondent answers to items on previous waves to measure important political attitudes before the scandal broke, so these attitudes cannot have been influenced by the expenses scandal itself.

Our first control measures a respondent's previous electoral support for their 2005–10 incumbent MP. It is coded as one if (in the 2005 post-election wave of the BES panel) the respondent reported voting for this MP at the 2005 general election and zero otherwise. This variable acts as a proxy for the respondent's prior support towards the MP's party as well as towards the MP him- or herself.

Second, as a more direct measure of affect toward an MP's party, we control for whether a constituent identifies with the party of his or her 2005–10 incumbent MP. We code this binary measure based on a respondent's answers to standard party ID items in the 2008 wave of the BES panel, the last wave before the scandal broke. This variable equals one if a respondent identifies with the party of their MP, and zero otherwise.14

Aside from controlling for respondent predispositions toward their MP, we also control for more general respondent attributes that may influence their responses. We include variables that measure attention paid to politics and political knowledge, since higher levels of each should decrease ‘don't know’ responses. We also include measures for general trust and political efficacy, since those who do not trust people in general and feel politically powerless might be more inclined to think their MP overclaimed.

We include two constituency-level variables. First, in order to control for the possibility that perceptions differ for representatives of each party, we control for the party affiliation of a respondent's MP with three dummy variables (Conservative, Liberal Democrat and other parties, respectively, with Labour as the reference category).15 We also include a control for the notional size of the majority the winning MP enjoyed in the respondent's constituency in 2005 (Norris, 2010).16 The tighter the race in the seat, the more incentive a constituent may have to become informed about the behaviour of their MP, and the more challenger candidates might try to insinuate wrong-doing on the part of an incumbent MP.

Finally, we also include a dummy variable that equals one when a respondent's 2005 constituency matches his or her 2010 constituency, and zero otherwise.17 The variable controls for the possibility that the respondent moved between 2005 and 2010 or that he or she was assigned a new incumbent MP due to boundary changes. While our dependent variable is based on a survey question fielded before the 2010 election, and therefore refers to a respondent's incumbent under the 2005 electoral boundaries, a respondent whose constituency is due to change under the re-drawn 2010 boundaries might be less sure of their MP's identity and therefore more likely to answer ‘don't know’.18

Our key results are presented in Table 2. Model 1 includes only our main explanatory variable. Models 2 and 3 then add the series of controls described above, but vary in the measures of constituent predisposition toward an MP included: in Model 2, we control for whether the constituent voted for the MP in the 2005 election, whereas in Model 3 we also control for identification with the party of the MP.

Table 2

Predicting voter perceptions of MP behaviour.

These results provide strong evidence that constituents' perceptions regarding the propriety of their MP's expenses claims do respond to publicly available information. The negative and significant coefficients on the implicated variable across Models 1 to 3 indicate that a constituent whose MP was publicly implicated in the expenses scandal is more likely to think that his or her MP overclaimed (relative to either the ‘did not overclaim’ and to ‘don't know’ response options), and that this result is robust to the inclusion of control variables.

The control variables also largely have the expected effects. For example, increasing general trust and political efficacy make a constituent more likely to think that their MP ‘did not overclaim’ (relative to thinking the MP did overclaim), while increasing self-reported attention to politics and political knowledge make a ‘don't know’ response less likely.19 Interestingly, the coefficients also indicate that an increase in self-reported political attention also significantly increases a constituent's probability of perceiving their MP to have ‘overclaimed’ relative to ‘not overclaimed’. Finally, the coefficients on the MP party affiliation suggest that, all else equal, in 2010 Liberal Democrats were perceived as more honest with regard to their expenses claims than other MPs.20

To interpret the magnitude of the effect of MP implication in the expenses scandal on constituent perceptions, in Fig. 1 we plot predicted probabilities based on Model 3. We hold all continuous variables at their mean and categorical variables at their mode. The top panel presents predicted probabilities for a constituent who did not vote for their MP in 2005 and did not identify with the MP's party in 2008, and the bottom panel for a constituent who did both. The effect of a constituent's MP being implicated in the scandal is surprisingly strong. For example, for voters in the top panel and whose MP is not implicated in the scandal, the probability of an ‘MP did not overclaim’ perception is around 0.33 and that of an ‘MP did overclaim’ perception is 0.14. In contrast, for such voters with an MP implicated in the scandal, the equivalent numbers are 0.20 and 0.35, respectively. This is a clear reversal of relative perceptions among those with an opinion on their MP's behaviour. The change in the predicted probabilities is very similar for voters who voted for their MP in 2005 and who identified with their MP's party in 2008 (bottom panel). Given the general belief that voters in Britain know little about their representatives, the effect of publicly available information is much stronger than we would have expected.

Voter perceptions of whether MPs claimed excessive expenses. a. Did not vote for MP and does not identify with MP's party. b. Did vote for MP and does identify with MP's party. Note: Based on Model 3, Table 2. Panel (a) holds 2005 vote for MP and party identification with MP's party at ‘no’, panel (b) at ‘yes’. All other variables held at their means (continuous variables) or modes (indicator variables).

Yet, it is not the case that constituent perceptions of MP misconduct are overwhelmingly accurate. First, Table 1 showed that a large proportion of voters simply do not know whether their MP was involved in the expenses scandal or not, or at the very least do not feel certain in their beliefs. Second, voter predispositions toward their MP are also very important in determining voter perceptions of MP misbehaviour. The estimates for Model 3 (Table 2) show that having voted for the MP and identifying with the MP's party increase the probability of thinking that one's MP did not overclaim. The two panels of Fig. 1 help us to understand the predicted differences. For example, where an MP is not implicated in the scandal, a typical constituent who voted for that MP and identifies with the MP's party is estimated to think that the MP ‘did not overclaim’ with probability 0.56, whereas the equivalent estimate is 0.33 for a constituent who did not vote for the MP and does not identify with the MP's party. Similarly, where an MP was implicated in the scandal, a constituent who voted for the MP and identifies with the MP's party is 20 points more likely to maintain that their MP did not overclaim than a constituent who neither voted for the MP nor identifies with the MP's party. In sum, voter predispositions toward MPs are as influential in determining perceptions of misconduct as whether or not an MP was publicly implicated in the scandal.

One concern with these results is that they may be an artefact of the particular way in which we measure public implication of an MP in the expenses scandal. To show that this is not the case, in Table 3 we report an extra series of models of voter perceptions of MP overclaiming that check the robustness of our findings to alternative measures of MP implication. The estimates for Models 4, 5 and 6 show the regression results for when we break down our implicated measure into its two constituent parts and then either include these in separate regressions (Models 4 and 5) or together in the same regression (Model 6). The Table shows that both measures are significantly positively associated with the probability that a voter thinks their MP overclaimed. In addition, Models 7 and 8 proxy the severity of an MP's expense overclaiming by measuring the amount an MP was asked to repay by Legg (measured continuously in Model 7 and categorized in Model 8 to allow for nonlinearities). Estimates for both of these models show that voters were significantly more likely to think their MP overclaimed the more the MP was asked to repay. Note that these models include all of the controls used in Model 3.21

Table 3

Robustness checks for predicting perceptions of MP expenses-related behaviour.

Across all models in Table 3, the coefficient estimates for these controls are very similar to the corresponding coefficient estimates in Table 2. Further analyses (included in the Supplemental materials) also show that voter perceptions of how much MPs overclaimed are related to the actual amount they were ordered to repay in the Legg report. Thus, our central finding, that public information about an MP's expenses claims has a substantive influence on constituent perceptions of those claims, is robust to alternative measures of public implication.

5. Perceptions of misconduct and constituent vote choice

Do constituents then sanction MPs they perceive to have overclaimed? To answer this question, we test whether a constituent is less likely to vote for an incumbent MP at the 2010 general election if they believe that he or she overclaimed in the expenses scandal.22 Of course, not all 2005–10 incumbent MPs stood for re-election in 2010, and boundary changes introduced at the 2010 election further complicate the picture. Because we are interested in whether constituents use their votes to hold their incumbent representatives electorally accountable for perceived misconduct, for this part of the analysis we restrict our sample to those respondents who were located in a 2010-defined constituency where a sitting MP ran for re-election in 2010, and who had been represented by that MP between 2005–10.23

Are voters who think their MP overclaimed more likely to vote for an opposing candidate?24 Our dependent variable here is a binary measure that equals one if the respondent voted for the incumbent MP at the 2010 general election, and zero if he or she voted for another candidate. This variable is coded based on a respondent's self-reported vote choice in the 2010 post-election wave of the BES panel. Since our dependent variable is dichotomous, we use binary logistic regression, again clustering standard errors by constituency. Because of the attrition that arises due to the use of an extra panel wave, and the restriction of the sample to respondents with an incumbent MP re-running in 2010 who also voted in the election, our overall sample size is here reduced to 1134.

Our main explanatory variable of interest is the categorical measure of whether or not a respondent believed that their incumbent MP was guilty of overclaiming on expenses, or whether they did not know; we use the ‘did not overclaim’ perception as the baseline.25 A potential complication here is that, as we have already shown, respondents who are predisposed toward their incumbent MP (either for MP-specific or partisan reasons) are less likely to perceive that MP to have overclaimed on expenses, but these predispositions are also likely to influence vote choice. To deal with this, we again take advantage of the panel nature of our data and control for a respondent's predisposition toward their MP as measured prior to our expenses perceptions variable. Thus, in Model 9 we control for whether, at the 2005 election, a respondent voted for their 2005–10 incumbent MP. This captures the prior electoral behaviour of the voter. Furthermore, in Model 10 we include a further control for whether the respondent identified with the party of their incumbent MP when surveyed in 2008. In both models we also account for differential partisan swings at the 2010 election by including dummies indicating the party affiliation of a respondent's MP.26

As the estimates for Models 9 and 10 show (Table 4), both of our measures of respondent predisposition toward an MP are strongly positively associated with voting for an incumbent MP.27 Nevertheless, while the coefficient on the ‘don't know’ dummy variable is not significant, the coefficient on the ‘overclaimed’ dummy variable is both significant and negative. This provides strong evidence that when a constituent perceived their MP to have made improper expenses claims, they were less likely to support that MP at election time.

Table 4

Constituent perceptions and vote choice.

To gauge the magnitude of this effect, we present predicted probabilities based on Model 10 for two types of respondents in Fig. 2. The left panel shows the probability of voting for the incumbent MP for a respondent who did not vote for that MP in 2005 and who does not identify with the MP's party. For such voters, a perception of guilt reduces the probability of voting for the incumbent by five points, from 0.14 to 0.09. The right panel shows the same probabilities for voters who identify with the MP's party and voted for him or her in 2005. The overall predicted probabilities are much higher, but the decline due to a perception of MP overclaiming is a similar 7 points (from 0.84 to 0.77). Furthermore, ninety-five per cent confidence intervals (not graphed here) indicate that the marginal effects on vote choice of moving from a ‘did not overclaim’ to an ‘overclaimed’ perception are significant for both types of voters.28

Perceptions of misconduct and vote choice. Note: Based on Model 10, Table 4. Panel (a) holds 2005 vote for MP and party identification with MP's party at ‘no’, panel (b) at ‘yes’. All other variables held at their means (continuous variables) or modes (indicator variables).

As a further check on our results, we also estimate the association between the amount of money a respondent believed their MP had overclaimed and the probability of voting for the MP. To measure perceptions of the severity of an MP's overclaiming, we use answers to a BES item that asked respondents to state how much expense money they thought their MP had overclaimed, on a scale of 0–10 (with higher scores corresponding to greater amounts). This item was asked as a follow-up question only to those respondents who believed their MP had overclaimed. We deal with this by coding the variable in two ways: in Model 11, we use the raw scores, treating as missing those respondents who did not think their MP overclaimed; in Model 12, we include all respondents who thought their MP either did or did not overclaim by coding the latter respondents as −1. The coefficient sizes for the perceived extent of overclaiming are similar in both Models. In terms of marginal effects, in Model 12 a one standard-deviation change in the perceived severity of overclaiming is predicted to lead to about a 6 per cent reduction in the probability of voting for the incumbent. This effect is similar in magnitude to the impact of thinking an MP had overclaimed at all in the Model 10.29 This suggests that our estimate of the magnitude of the effect of perceptions of misconduct on vote choice is not an artefact of the way we code perceptions.

In sum, there is evidence that at the 2010 general election British voters to some extent acted to sanction MPs electorally for perceived overclaiming on expenses by voting for an opponent.30 Considering that perceived MP implication in the scandal is just one of many possible considerations at a general election, a five per cent reduction in the probability of voting for a candidate due to perceived wrongdoing is not trivial. Nevertheless, effects of this size do not suggest that perceptions of MP misconduct were strong determinants of voting behaviour.

The predicted influence of perceptions on vote choice can be combined with the predicted influence of implication on perceptions in order to gauge how implication indirectly affects vote choice via perceptions. To do this, we consider an average voter with all controls held at their mean value in the data. First, we calculated the predicted probability for average voters of thinking their MP was implicated, depending on whether or not the MP was in fact implicated. Then, we calculated the predicted probability for that average respondent of voting for the incumbent based on these perceptions, again separately for implicated and non-implicated MPs. For non-implicated MPs, the predicted probability that an average voter will vote for them is 38.7 per cent. This drops to 36.8 per cent for implicated MPs. This indicative calculation suggests that implication in the scandal reduced the predicted probability of voting for the incumbent by 1.9 per cent. This overall effect of MP implication on the vote choice of an average voter implies an aggregate effect of MP implication on MP vote share that is similar in magnitude to that obtained in existing aggregate-level analyses of the expenses scandal (Curtice et al., 2010; Eggers and Fischer, 2011; Pattie and Johnston, in press).

6. Conclusion

Using the uniquely advantageous case of the UK expenses scandal, we have provided new evidence as to why electoral accountability for representative misconduct may function imperfectly. With regard to voter perceptions of their MP's conduct, whilst almost half of voters do not know whether their MP was involved in the scandal and whilst perceptions are biased by political predispositions, our findings nevertheless demonstrate that voter beliefs about the expenses-related conduct of their MP are clearly and strongly influenced by public revelations concerning that conduct. Given past research and the institutional setting of the UK, we are surprised that the responsiveness of perceptions to publicly available information is as high as it is.

Turning to sanctions, we find that constituent perceptions do not translate strongly into electoral punishment: few voters punish their MP for perceived misbehaviour. A perception that an incumbent MP had overclaimed on expenses was associated with a predicted decline of around 5 per cent in the probability of voting for him or her. Thus, constituents to some extent hold their MPs accountable for perceived misconduct, but the link between perceptions and vote choice is weak compared to that between publicly available information and perceptions.

When the two steps are considered jointly, we would predict an average voter to be 1.9 percentage points less likely to vote for an implicated incumbent MP than for a non-implicated MP. This small effect is consistent with previous studies of the scandal (Curtice et al., 2010; Eggers and Fischer, 2011; Pattie and Johnston, in press).

Apart from the availability of panel survey data and good measures of voter perceptions of misconduct, a key advantage of our research design was the fact that we could hold the nature and timing of the scandal constant while exploiting natural variation in whether a survey respondent's MP was implicated in the scandal. Of course, this means that we could only examine one type of scandal, the misuse of parliamentary expenses. We do not claim that the pattern we find need apply to all cases of scandal. For example, in situations where the wrongdoing was more severely unethical or more clearly isolated in media reporting, even voters in the UK may be likely to overwhelmingly punish their representative. One example is the case of Neil Hamilton, a Conservative MP who lost his seat as a direct result of his involvement in the so-called cash-for-questions affair in 1997.

Notwithstanding this qualification, our findings have three important implications. First, while ignorance and bias about representative misconduct exist, these do not dominate voter perceptions, and information about representatives clearly influences constituent perceptions. This is the case even in a system as dominated by party – rather than candidate – competition such as the UK, where the institutional setting works against voters learning about the conduct of their specific MP (Cain et al., 1987; Carey and Shugart, 1995; Pattie and Johnston, 2004; Kam, 2009). Second, in contrast to other studies which emphasise the critical role of publicly available information about the misconduct of representatives in assuring electoral accountability for representative misconduct (Chang et al., 2010), our findings point to the fact that electoral accountability can fail even in situations where a lack of information is not the main problem. Third, if electoral accountability for misconduct is our goal, it is important to ensure that voters can use the information they have to sanction their representatives. The British-style party-centric political system works against voters conditioning their vote choice on MP behaviour and may explain why even voters who think their representative engaged in wrongdoing decide to cast their vote based on other considerations. These incentives are likely to be significantly different under other institutional arrangements, such as open lists or primaries, which allow voters to separate their vote for their MP from their vote for a party. So, while it is absolutely necessary that voters are provided with information about misconduct and that they update beliefs based on this information, we also need to choose institutional frameworks that allow voters to use that information to guide their vote choice.

Footnotes

☆Previous versions of this paper were presented at the 2011 EPSA General Conference and the 2011 EPOP Annual Conference. We would like to thank the anonymous reviewers as well as Andrew Eggers, Emily Clough, Torun Dewan, Martin Ejnar Hansen, Olli Hellmann, Simon Hix, Shanto Iyengar and Ron Johnston for comments on an earlier version of this paper. Markus Wagner's research for this paper was supported by the Austrian Science Fund (S10903-G11).

2BES 2010 pre-election internet cross-section survey of 15,660 British voters, weighted using the standard BES weight.

3See Besley and Larcinese (2011) for a more detailed discussion of the different types of expenses allowances available to British MPs.

4We exclude the 18 Northern Ireland MPs from our analysis, since the British Election Study did not survey Northern Ireland residents.

5We refer to the combined source of the Daily Telegraph and the Sunday Telegraph as ‘the Telegraph’ throughout the paper.

6We searched The Daily Telegraph and Sunday Telegraph articles published between 1 January 2009 and 26 March 2010, via NexisUK (an electronic media search engine). A general article refers to those pieces wherein the Telegraph listed the claims of five or more MPs without specific details of, for example, the amount an MP claimed per year.

7Though there were 628 sitting British MPs in the 2005-10 Parliament, we exclude four groups of MPs when coding our implicated variable: MPs representing inner-London constituencies, as they were not allowed to claim expenses under the Additional Costs Allowance and were therefore never likely to be implicated in the scandal; the three MPs involved in criminal proceedings as a result of their expenses claims, as their misbehaviour was qualitatively far different from that of other MPs; the MPs elected in by-elections between 2005 and 2010, because the survey respondents did not vote for these MPs in 2005 and because they often had less chance to be involved in the scandal; and an MP who died in March 2010, so just before the period the survey was fielded.

8The BES ran parallel internet and face-to-face surveys in 2005, and detailed inspection has shown that the samples are almost indistinguishable (Sanders et al., 2007). Sanders et al. (2011) also report that the 2005–2010 panel continued to be representative of the population in terms of their reported vote and their vote intention.

9The BES matches respondents to seats using the postcodes provided by the participants.

10We add to the ‘don't knows’ the very small number of respondents who in the previous question indicated they had not heard of the scandal at all.

11The distribution of responses to this question is very similar in the BES 2010 cross-sectional pre-campaign internet survey, which has the advantages of a larger number of respondents. The main difference is that the number of ‘don't knows’ increases slightly, probably a consequence of panel attrition.

12Multinomial logit models assume independence of irrelevant alternatives (IIA). But we do not believe the IIA assumption is problematic here since the three response options are not substantively similar (Kennedy, 2008). Moreover, IIA is important mainly when we are concerned with the effect of one of the options being added or removed (Glasgow, 2001). While this can easily occur with parties and candidates, this is of less relevance in our case. Nevertheless, we ran the Hausman–McFadden test for each model, and results failed to indicate that the IIA assumption was a problem.

13We also tried accounting for the hierarchical nature of our data by analysing data aggregated to the constituency level (Angrist and Pischke, 2009, p.322) and also by running hierarchical binary logits on pairs of response options using the original individual-level data. No substantively important differences to the results of the multinomial logit model were found; see supplemental material for details.

14We also ran our models using a stricter measure of identification with the MP's party; respondents were coded as 1 only if they claimed to identify with the party of their MP in two 2005 waves, the 2006 wave and the 2008 wave, 0 otherwise (Sanders et al., 2011). (The pre-campaign 2005 wave is excluded from this measure as it did not include the standard party identification question in the whole sample.) We also tried two further measures that might indicate voter attachment to the MP and his or her party. We thus substituted party identification with party affect (a 0-10 like-dislike scale) and with an indicator of whether the respondent voted for the MP in question in the 2005 election. The results are robust to all three modifications of the measure of voter-MP attachment.

15Other survey evidence points to the fact that Liberal Democrat MPs were seen as less implicated by the scandal (YouGov, 2009).

16We use the notional 2005 majority size based on the 2010 boundaries since this is more likely to reflect the competitiveness of the constituency in 2010 than the actual 2005 majority size using the old boundaries.

17This variable (‘seat change’) was created by manually pairing up the 2005 and 2010 constituencies. In cases of boundary changes, 2005 constituencies were matched to successor constituencies. In most cases a successor constituency was identified when the incumbent ran again in a newly-formed constituency that contained part of his or her old constituency. In the remaining cases, we tried to match 2010 to 2005 constituencies based on the similarity of electorate. The 15 2010-defined seats that we did not match a 2005 constituency to were either entirely new seats or correspond to 2005-defined seats where a by-election occurred between 2005 and 2010.

18We also ran all models excluding all respondents who are in different seats in 2005 and 2010. All key findings are robust to this reduction of the sample.

19We also tested whether the effects of MP implication on voter perceptions were greater among respondents that are more knowledgeable about politics and pay more attention to it. We did so by including the appropriate interaction term. The results for the political knowledge–implication interaction offer no evidence that political knowledge has a statistically significant moderating effect on the link between MP implication and voter perceptions of their MP's expenses-related conduct. For the attention–implication interaction, the results are not straightforward: our findings indicate a statistically significant interaction, with voters with lower levels of attention slightly more likely to have a perception that matches the MP's implication. The explanation for this counter-intuitive finding is unclear, but it may be respondents who are less willing to admit they pay little attention to politics are also less likely to admit they do not know about their MP's involvement in the scandal, and thus more likely to guess whether or not their MP over-claimed (thus weakening the link between implication and a perception of over-claiming).

20We also ran models with appropriate interaction terms to check whether the effect of wrongdoing on perceptions thereof depended on the party affiliation of the MP or on the marginality of the constituency. We find no evidence for either type of conditionality.

21A further concern is that panel attrition and/or the initial make-up of the internet panel may have led to an over-representation of well-informed and politically involved respondents, which might in turn lead us to over-estimate the responsiveness of constituent perceptions to local MP implication. While we cannot completely allay this concern with the available data, we do control for respondents' political knowledge and self-reported political attention in our models. Furthermore, inspection of the distribution of these variables in our data indicates that there is at least a sizeable number of respondents with low scores on these two variables (for example, there are 79 respondents with the lowest possible score on the attention measure).

22Our focus is on incumbent MPs as we are interested in accountability for misbehaviour. However, even new replacement candidates could face negative electoral consequences if voters decide to sanction them for their predecessor's wrongdoing or if their predecessor's behaviour also shapes perceptions of his or her replacement.

23We would be underestimating voter sanctioning of misconduct if the worst offenders had resigned before the election. However, as noted above even among the most consistently implicated MPs, more than half chose to run again (Pattie and Johnston, in press), so focusing on re-standing MPs should not make our estimates severely unrepresentative.

24It may also be that voters who think their MP overclaimed would be more likely to abstain, especially if they would prefer to support the party of that MP. However, we cannot carry out useful analyses of turnout decisions with our data because self-reported turnout is very high, partly a consequence of our focus on panel data. Unfortunately, validated vote data is only available for the BES pre-/post-election face-to-face survey data.

25There may also be an effect of MP wrongdoing on voting behaviour that does not operate via constituent perceptions, for example if the MP's misbehaviour led to increased campaign effort among opponents and decreased effort by the MP. We tested this by using misbehaviour itself as a predictor variable instead of perceptions, and actual misbehaviour has no effect on vote choice.

26We also tried including three constituency-level variables: the length of time the MP has been in parliament; whether the MP sat on the government or opposition front bench; and the (notional) majority won by the MP in 2005. Since none of these variables proved to be significant predictors, we leave them out of the models we present.

27Our sample size changes between Models 9 and 10 because we include different predictors in each model and these predictors have different numbers of missing observations.

28We also checked whether the link between perceptions and vote choice depends on party identification or MP affect. To do so, we ran models with an interaction between respondent perceptions of misconduct, identification with MP's party in 2005, and respondent vote in 2005. None of the interaction term coefficients were significant. However, the direction of the moderating effect indicates that perceptions of wrongdoing may influence vote choice only among voters those who did not identify with their MP's party, which would mean that party identifiers are more resistant to changing their vote choice as a result of their beliefs regarding their MP's behaviour.

29The marginal effect is calculated for a case where the respondent has a Labour MP, voted for his or her incumbent and identifies with the MP's party.

30As Curtice et al. (2010) show, there is evidence of this at the aggregate level as well.

Appendix ASupplementary material associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.electstud.2012.06.010.

Appendix A. Supplementary material

References

Ahuja S., Beavers S.L., Berreau C., Dodson A., Hourigan P., Showalter S., Walz J., Hibbing J.R. Modern congressional election-theory meets the 1992 house elections. Political Research Quarterly. 1994;47:909–921. [Google Scholar]

Angrist J.D., Pischke J.-S. Princeton University Press; Princeton, NJ: 2009. Mostly Harmless Econometrics: An Empiricist's Companion. [Google Scholar]

Ansolabehere S., Jones P.E. Constituents' responses to congressional roll-call voting. American Journal of Political Science. 2010;54(3):583–597. [Google Scholar]

Arnold R.D. Can inattentive citizens control their elected representatives? In: Dodd L.C., Oppenheimer B.I., editors. Congress Reconsidered. Congressional Quarterly Press; Washington, DC: 1993. [Google Scholar]

Banducci S.A., Karp J.A. The electoral consequences of scandal and reapportionment in the 1992 house elections. American Politics Quarterly. 1994;22:3–26. [Google Scholar]

Benedetto G., Hix S. The rejected, the ejected and the dejected: explaining government rebels in the 2001–2005 British House of Commons. Comparative Political Studies. 2007;40(7):755–781. [Google Scholar]

Besley T. Oxford University Press; Oxford, UK: 2006. Principled Agents? The Political Economy of Good Government. [Google Scholar]

Besley T., Larcinese V. Working or Shirking? Expenses and attendance in the UK Parliament. Public Choice. 2011;147:291–317. [Google Scholar]

Cain B., Ferejohn J., Fiorina M. Harvard University Press; Cambridge, MA: 1987. The Personal Vote: Constituency Service and Electoral Independence. [Google Scholar]

Campbell A., Converse P.E., Miller W.E., Stokes D.E. Wiley; New York: 1960. The American Voter. [Google Scholar]

Canes-Wrone B., Brady D.W., Cogan J.F. Out of step, out of office: electoral accountability and house members' voting. American Political Science Review. 2002;96(1):127–140. [Google Scholar]

Carey J.M., Shugart M.S. Incentives to cultivate a personal vote: a rank ordering of electoral formulas. Electoral Studies. 1995;14(4):417–439. [Google Scholar]

Chang E.C.C., Golden M.A. Electoral systems, district magnitude and corruption. British Journal of Political Science. 2006;37:115–137. [Google Scholar]

Chang E.C.C., Golden M.A., Hill S.J. Legislative malfeasance and political accountability. World Politics. 2010;62(2):177–220. [Google Scholar]

Clarke H.D., Feigert F.B., Seldon B.J., Stewart M.C. More time with my money: leaving the house and going home in 1992 and 1994. Political Research Quarterly. 1999;62:67–85. [Google Scholar]

Curtice J., Fisher S.D., Ford R. Appendix 2: an analysis of the results. In: Kavanagh D., Cowley P., editors. The British General Election of 2010. Palgrave Macmillan; Basingstoke: 2010. [Google Scholar]

Delli Carpini M.X., Keeter S. Yale University Press; New Haven, CT: 1996. What Americans Know About Politics and Why It Matters. [Google Scholar]

Dimock M.A., Jacobsen G.C. Checks and choices: the house Bank Scandal's impact on voters in 1992. The Journal of Politics. 1995;57(4):1143–1159. [Google Scholar]

Farrell D.M., McAllister I., Studlar D.T. Sex, money and politics: sleaze and conservative incumbency in the 1997 British election. In: Denver D., editor. vol. 8. Frank Cass; London: 1998. (British Elections and Parties Review). [Google Scholar]

Fenno R. Home Style: House Members in the Districts. Little Brown; Boston: 1978. F. [Google Scholar]

Ferejohn J.A. Incumbent performance and electoral control. Public Choice. 1986;50:5–20. [Google Scholar]

Ferejohn J.A. Information and the electoral process. In: Ferejohn J.A., Kuklinski J.H., editors. Information and the Democratic Process. Illinois University Press; Urbana, IL: 1990. pp. 1–22. [Google Scholar]

Glasgow G. Mixed logit models for multiparty elections. Political Analysis. 2001;9(2):116–136. [Google Scholar]

Holmberg S. Candidate recognition in different electoral systems. In: Klingemann H.-D., editor. The Comparative Study of Electoral Systems. Oxford University Press; Oxford: 2009. pp. 158–170. [Google Scholar]

Kam C.J. Cambridge University Press; Cambridge: 2009. Party Discipline and Parliamentary Politics. [Google Scholar]

Kelso A. Parliament on its knees: MPs' expenses and the crisis of transparency at Westminster. The Political Quarterly. 2009;80(3):329–338. [Google Scholar]

Kennedy P. Wiley–Blackwell; Oxford: 2008. A Guide to Econometrics. [Google Scholar]

Key V.O. Vintage; New York: 1966. The Responsible Electorate. [Google Scholar]

Kunicová J., Rose-Ackerman S. Electoral rules and constitutional structures as constraints on corruption. British Journal of Political Science. 2005;35(4):573–606. [Google Scholar]

Lewis-Beck M.S., Stegmaier M. Economic determinants of electoral outcomes. Annual Review of Political Science. 2000;3:183–219. [Google Scholar]

Marsh M., Tilley J. The attribution of credit and blame to governments and its impact on vote choice. British Journal of Political Science. 2010;40(1):115–134. [Google Scholar]

Pattie C., Johnston R. Party knowledge and candidate knowledge: constituency campaigning and voting and the 1997 British general election. Electoral Studies. 2004;23(4):795–819. [Google Scholar]

Pattie, C., Johnston, R. The electoral impact of the UK 2009 MPs' expenses scandal. Political Studies, in press.

Pattie C., Fieldhouse E., Johnston R. The price of conscience: the electoral correlates and consequences of free votes and rebellions in the British House of Commons, 1987–92. British Journal of Political Science. 1994;24(3):359–380. [Google Scholar]

Persson T., Tabellini G. MIT Press; Cambridge, MA: 2000. Political Economics: Explaining Economic Policy. [Google Scholar]

Persson T.G., Tabellini, Trebbi F. Electoral rules and corruption. Journal of the European Economic Association. 2003;1(4):958–989. [Google Scholar]

Peters J.G., Welch S. The effect of charges of corruption on voting behaviour in congressional elections. American Political Science Review. 1980;74:697–708. [Google Scholar]

Powell G.B., Whitten G.D. A cross-national analysis of economic voting: taking account of the political context. American Journal of Political Science. 1993;37:391–414. [Google Scholar]

Przeworski A., Stokes S.C., Manin B. Elections and representation. In: Przeworski A., Stokes S.C., Manin B., editors. Democracy, Accountability and Representation. Cambridge University Press; Cambridge: 1999. pp. 29–54. [Google Scholar]

Rundquist B.S., Strom G.S., Peters J.G. Corrupt politicians and their electoral support: some experimental observations. American Political Science Review. 1977;71:954–963. [Google Scholar]

Samuels D., Hellwig T. Elections and accountability for the economy: a conceptual and empirical reassessment. Journal of Elections, Public Opinion and Parties. 2010;20(4):393–419. [Google Scholar]

Sanders D., Clarke H.D., Stewart M.C., Whiteley P. Does mode matter for modeling political choice? Evidence from the 2005 British election study. Political Analysis. 2007;15(3):257–285. [Google Scholar]

Sanders D., Clarke H.D., Stewart M.C., Whiteley P. Downs, stokes and the dynamics of electoral choice. British Journal of Political Science. 2011;41:287–314. [Google Scholar]

Stevenson, R. and G. Vonnahme (2009) Executive Selection and the Informed Electorate: How the Rules and Norms Governing Cabinet Formation Impact Citizens' Knowledge of Politics, paper presented at the Annual Meeting of the Midwest Political Science Association, Chicago, 2–5 April 2009.

Tilley J., Hobolt S. Is the government to blame? An experimental test of how partisanship shapes perceptions of performance and responsibility. Journal of Politics. 2011;73(2):316–330. [Google Scholar]

Vivyan N., Wagner M. Do voters reward rebellion? The electoral accountability of MPs in Britain. European Journal of Political Research. 2012;51(2):235–264. [Google Scholar]

Welch S., Hibbing J.R. The effects of charges of corruption on voting behavior in congressional elections, 1982–1990. Journal of Politics. 1997;59:226–239. [Google Scholar]