In many democracies, political trust is at a low ebb. Echoing concerns stretching back to the 1970s (Miller Reference Miller1974), there have been large falls in political trust across almost all advanced industrial democracies over the last fifty years (Citrin and Stoker Reference Citrin and Stoker2018). For example, in the United Kingdom (UK) trust has been in decline since the mid-twentieth century (Jennings et al. Reference Jennings2017). Why has this happened? One explanation focuses on the prevalence of intensively reported political scandals that implicate voters' constituency representatives and reduce wider levels of trust (Rose and Wessels Reference Rose and Wessels2019). Unsurprisingly, when a politician is implicated in a scandal, support for that politician tends to fall (Ferraz and Finan Reference Ferraz and Finan2008). Many have argued that there is also a ‘spillover effect’. That is, constituents of an implicated politician also reduce their diffuse support for all politicians and national institutions (Bowler and Karp Reference Bowler and Karp2004; van Elsas et al. Reference van Elsas2020; von Sikorski et al. Reference von Sikorski2020). In essence, a voter's judgement of the specific misconduct of their representative becomes a judgement on the system as a whole.
In this article, we test that spillover effect. We investigate the three necessary steps that link misconduct by a voter's own representative to diffuse political trust. First, are voters aware that their representative is implicated? Second, do the constituents of scandal-implicated representatives become less supportive of their representative? Third, do voters' specific attitudes about an implicated representative spill over and reduce their diffuse political trust?
We argue that while there is evidence for the first two steps, there is little support for the third step of spillover. Rather than relying on experimental evidence, which limits partisan biases and exposes voters to (often hypothetical) information which they would otherwise not receive, we use a real scandal: the 2009 Parliamentary expenses scandal in the UK (Heath Reference Heath, Allen and Bartle2011; van Heerde-Hudson Reference van Heerde-Hudson and Wring2011; Vivyan et al. Reference Vivyan2012). Moreover, unlike existing work that operationalizes scandal implication with voter perceptions (Lee Reference Lee2018), self-reported scandal news consumption (von Sikorski et al. Reference von Sikorski2020) or simple yes/no measures (Bowler and Karp Reference Bowler and Karp2004), we precisely quantify the degree to which each representative was financially implicated in the scandal and the degree to which that implication was visible in the media. We use this variation in representative implication to accurately assess the consequences of scandals for political trust. Building on existing work that considers a single form of political trust (von Sikorski et al. Reference von Sikorski2020), we are also able to differentiate between the effects on trust in different institutions and groups.
The Parliamentary expenses scandal is a case in which one would expect large spillovers onto political trust. It was a financial scandal in which politicians from all the major parties were implicated and the media coverage was largely free of partisan bias (Larcinese and Sircar Reference Larcinese and Sircar2017). Many voters could identify whether their particular MP was implicated, reflecting the importance of MPs' constituency ties for British voters (Vivyan et al. Reference Vivyan2012; Vivyan and Wagner Reference Vivyan and Wagner2015). We might, therefore, expect that voters generalize the behaviour of their representative onto diffuse groups and institutions. Yet, in practice, we detect little evidence of this. Rather, using British Election Study survey data and an MP-level implication dataset, we find that while constituents were aware of their representative's implication, and this affected their attitudes toward their representative, these specific judgements did not measurably spill over onto their diffuse political trust of institutions, politicians, or democracy itself.
In what follows, we first explain why we might expect a spillover effect. We then discuss the data we use to measure voter attitudes, MP behaviour and media coverage before detailing the results of our analyses. We conclude by suggesting that our findings cast doubt on the practical applicability of the spillover mechanism. Our evidence suggests that, at least for the type of scandal we consider, scandalous behaviour by a voter's local representative is unlikely to affect their generalized political trust.
Scandals and Political Trust
How could misconduct by a local representative affect political trust? Von Sikorski et al. (Reference von Sikorski2020) differentiate between the eroding and spillover effects of scandals. The former is the effect upon my specific support for my local representative and the latter is the effect on my diffuse political trust. This follows Easton's (Reference Easton1975) division between specific support for individual political actors and diffuse support for institutions and principles.
Intuitively, a scandal should affect specific support for implicated representatives: the eroding effect. It is almost a truism that scandals are associated with negative candidate evaluations and lower vote shares for implicated politicians (von Sikorski et al. Reference von Sikorski2020; Welch and Hibbing Reference Welch and Hibbing1997; although see Keele Reference Keele2007). The media is crucial, however. For representative misconduct to influence a voter's view of the representative, the voter must know about the politician's involvement in the scandal. Voters should, therefore, respond more to scandals that are intensively covered and we tend to see this in studies of electoral accountability. For example, Ferraz and Finan (Reference Ferraz and Finan2008) find that the effect of random corruption audits on anti-incumbent voting is strengthened in communities with more radio coverage.
Overall, there is clear evidence that financial scandals negatively affect support for politicians caught with their hands in the till. Why might this translate to a generalized distrust of politicians, institutions and democracy? Norris (Reference Norris2011) suggests that different types of political trust are arranged hierarchically from the most diffuse (trust in democracy) to the most specific (trust in my particular representative) and voters draw upon their specific attitudes when forming more diffuse attitudes. This means that a voter's negative evaluation of a corrupt politician will ‘spill over’ and reduce the voter's diffuse political trust (Schwarz and Bles Reference Schwarz and Bles1992): the bad apple infects the rest of the barrel. And the effect of that bad apple could be substantial. Since voters prioritize integrity when deciding to trust (Valgarðsson et al. Reference Valgarðsson2021), and corruption undermines both procedural fairness and integrity, spillover effects could be large.
In support of this argument, experiments using single-politician scandals find spillovers onto trust in politicians (Maier Reference Maier2011), and observational studies find that people who receive information about a single-politician scandal subsequently reduce their trust in politicians and institutions (van Elsas et al. Reference von Sikorski2020; von Sikorski et al. Reference von Sikorski2018). In addition, Bowler and Karp (Reference Bowler and Karp2004) find that during two multi-politician scandals (the House Banking Scandal in the USA and the series of ‘sleaze’ scandals involving UK Conservative MPs in the 1990s) those who thought their representative was implicated exhibited lower political trust.
The spillover by assimilation mechanism is straightforward, but perhaps not completely realistic. There is another option for voters when their local representative is involved in a scandal. People may revise their opinion of the politician, but not revise their opinion of politicians or institutions as a whole: their politician is a bad apple, but the rest of the barrel remains untainted. Under this interpretation, people's diffuse political trust is less responsive to local events than their opinions about specific politicians. And while there is evidence of spillovers, as just discussed, there are problems with that evidence.
These problems fall into three categories. First is the nature of the scandal itself. Existing work focuses on single-politician scandals (von Sikorski Reference von Sikorski2018). This limits the sample size and complicates comparisons across cases. Equally, for many scandals, there are no objective measures of implication. This means that existing research often operationalizes scandal implication using either constituents' perceptions of their representative's involvement or self-reported scandal news consumption. Since these are endogenous to voters' underlying political trust and cynicism, the causal relationship remains unclear. The second problem is that the scandals investigated are often not real. Many studies are experimental, precisely to get around the problem of endogenous trust and perceptions of implication. This brings its own problems, as researchers are exposing voters to information about (often hypothetical) scandals that they would not usually receive. This inevitably limits the activation of cognitive biases that may dampen scandal effects and leads to potentially greater spillover effects than would be seen in reality. Third, there is the measurement of the spillover itself. Most observational research measures only very short-term effects immediately after the scandal, and the operationalization of political trust is typically done using a single indicator (even though voters are likely to differentiate their trust between different institutions and categories (van Elsas et al. Reference van Elsas2020)).
We improve upon this existing evidence by testing whether there was spillover onto different types of diffuse political trust in a most likely case of a spillover scandal: the UK Parliamentary expenses scandal of 2009. This case is particularly suitable because it was a financial and non-partisan scandal (see Appendix A for an overview). In addition, expenses misconduct was overwhelmingly seen as extremely serious by both the public and the media. The initial revelations in The Daily Telegraph prompted a rolling barrage of negative coverage from all major outlets (Larcinese and Sircar Reference Larcinese, Sircar and van Heerde Hudson2014), with The Times describing the scandal as ‘Parliament's darkest day’ (Webster Reference Webster2009). In addition, voters almost uniformly believed the scandal to be important and damaging. In the 2010 British Election Study Internet Panel, 90 per cent of respondents agreed that ‘The way some MPs have claimed expenses makes me very angry’, while only 7 per cent of respondents agreed that ‘Despite all the press coverage, what MPs claim for expenses really isn't that important’. Given the depth of public anger, we would expect a strong spillover onto diffuse political attitudes.
Using this most likely case of scandal spillover, we comprehensively test for spillovers by using both financial and media measures of scandal implication and by using multiple indicators of diffuse political trust. We also precisely track each stage of the spillover process: whether voters perceived their representative to be implicated, whether they then changed their attitudes about their MP, and then whether they changed their diffuse political trust.
Hypotheses
Our hypotheses are based on the expectations built into the spillover model. First, we hypothesize that the more heavily implicated the representative, by their conduct or the media, the more likely voters are to think that that representative is involved in the scandal.
H1: Scandal implication influences attribution of misconduct.
Second, we hypothesize that scandal implication erodes a voter's support for their own representative. The more heavily implicated the representative, whether by their conduct or the media, the greater this erosion of support for the representative.
H2: Scandal implication influences specific political trust (eroding effect).
Finally, we hypothesize that the voter's evaluations of their own representative will spill over onto diffuse political trust if the spillover model is correct. Thus, the more heavily implicated the representative, again by their conduct or the media, the more likely voters are to assimilate this specific case into their general view of politicians, political institutions, and democracy.
H3: Scandal implication influences diffuse political trust (spillover effect).
Data and measures
We use an individual-level British Election Study (BES) survey and an MP-level implication dataset. The former is the BES 2010 general election Rolling Thunder Campaign Panel. The survey is representative and gives us a much larger sample than previous studies of spillover effects, that either use small experimental samples (von Sikorski et al. Reference von Sikorski2020) or much smaller panels (Bowler and Karp Reference Bowler and Karp2004). Our data covers 16,429 survey respondents who live in 608 of the 646 parliamentary constituencies. The dependent variable for Hypothesis 1 is constituents' scandal attribution to their own MP. For Hypothesis 2, the dependent variable is constituents' specific trust in their MP, which is operationalized using approval of their MP. While this is not exactly the same as trust in the MP, the two tend to be very highly correlated. For example, in waves 4 and 16 of the 2014 to 2023 BES panel, both MP approval and MP trust are included simultaneously: the correlation between the two is 0.78. For Hypothesis 3, we measure spillover onto diffuse political trust in four ways: trust in the UK parliament; trust in politicians as a group; satisfaction with democracy; and agreement that ‘the reports on MPs’ expense claims prove that most MPs are corrupt’. In our main analysis, we recode our dependent variables to run from 0-1.
Several existing datasets measure MP-level scandal implication. We use a supplemented version of the Graffin et al. (Reference Graffin2013) data and distinguish between media and financial implication. This distinction is important since repayments made ‘due to technical matters which did not seem to concern the public’ (Eggers and Fisher Reference Eggers and Fisher2011, 13) may have received little coverage. For financial implication, we compare illegitimate, technically legitimate, and legitimate claims. Illegitimate claims are the repayments demanded by the Legg Report. Technically legitimate claims are second home claims (the practice of ‘flipping’ second home claims was a key source of public anger). Legitimate claims are the residual claims made in the previous two parliamentary sittings.Footnote 1 We measure media implication using Eggers' (Reference Eggers2014) formula for coverage intensity and include controls at the MP and voter levels. In our main analysis, we standardize our MP implication measures in terms of standard deviation changes. Our effects can, therefore, be interpreted as the effect of a one standard deviation change in MP implication upon each 0–1 outcome variable. See Appendices B and C for a detailed discussion of dataset choice, question wordings, control variables and descriptive statistics.
Analysis
To test Hypothesis 1, we use multilevel regression models with random effects in the MP-level intercepts to predict whether people agreed that their MP was implicated in the scandal. This allows us to account for both MP- and voter-level variation while estimating the effect of the MP-level independent variables (Gelman and Hill Reference Gelman and Hill2006). We show the results of multilevel linear models with the continuous dependent variable recording the extent to which respondents believed their MP had been implicated. Figure 1 plots the standardized coefficient estimates for these models along with their 95 per cent confidence intervals. The two key measures of implication are statistically significantly associated with voters' awareness of implication and are invariant to the inclusion of controls at the MP or voter level. Moreover, these effects are quite large, especially for the media score. An average constituent of the heavily scandal implicated Sir Peter Viggers, whose media score was 2.95 standard deviations above the mean, is predicted to score 0.30 on the 0-1 implication scale. An average constituent of the more frugal Ann Widdecombe, with a media score of 0.11, is predicted to score only 0.12. This effect is clearly quite large and supports Vivyan et al. (Reference Vivyan2012), who use a more permissive operationalization of implication. We thus find strong support for Hypothesis 1.
Voters' scandal attributions reflect financial and media implications.
Note: Linear estimates and 95 per cent confidence intervals from Appendix Table D1. The coefficients are the population average fixed effects, measured on a 0-1 scale, for a one standard deviation change in the independent variable.
We next consider whether MP-level implication influenced voters' specific support for their MP. Figure 2 plots the standardized coefficient estimates and 95 per cent confidence intervals from multilevel linear models predicting specific MP support. Both media implication and financial implication exert a statistically significant and negative effect on constituents' approval of their MP. Again, these effects are not altered by the inclusion of any control variables.Footnote 2 And, again, the effects of media implication are the largest. A one standard deviation increase in media implication is associated with a reduction of 0.02 on the 0-1 scale for specific support: moving from the least implicated to the most implicated MP decreases support by 0.10. Voters not only took notice of their MP's behaviour, but implication matters, at least a little, for their support of their local MP. This supports Hypothesis 2 and is consistent with the negative electoral effect for implicated MPs of around 1.5 per cent (Eggers and Fisher Reference Eggers and Fisher2011). Although the effect is small, it is similar to the estimated incumbency advantage in Westminster elections of around 2 per cent (Gaines Reference Gaines1998).
Scandal implication influenced support for specific representatives.
Note: Linear estimates and 95 per cent confidence intervals from Appendix Table D2. The coefficients are the population average fixed effects, measured on a 0-1 scale, for a one standard deviation change in the independent variable.
Finally, we consider Hypothesis 3. We know that media and financial implication affected whether voters identified their MPs as involved in the scandal and media and financial implication affected voters' approval of their MP, but did any of this spill over? We use multilevel linear models to predict all four of our dependent variables.Footnote 3 Figure 3 reports the standardized coefficient estimates from these models with all controls and their 95 per cent confidence intervals. None of the implication measures exert a statistically significant effect upon any measure of diffuse political trust. Indeed, while the spillover model would suggest a negative effect of implication on trust, the coefficients are inconsistently signed, and in all cases, they are very close to zero. Despite having the strongest effect on inclusion attributions and specific support, a one standard deviation increase in media implication can, at most, be associated with a decrease of 0.007 in diffuse political trust on a 0-1 scale, and its estimated effect on trust in politicians and parliament is positive. Meanwhile, the estimated effect of Legg repayments is positive for satisfaction with democracy, and its largest plausible negative effect on diffuse political trust is 0.006 on the 0-1 scale.Footnote 4 We, therefore, find no support for Hypothesis 3. If my representative is involved in a scandal, I am likely to be aware of this, but I do not become less trusting of politicians, institutions, or democracy.
Their own MP's implication did not influence voters' political trust.
Note: Linear estimates and 95 per cent confidence intervals from Appendix Table D3. The coefficients are the population average fixed effects, measured on a 0-1 scale, for a one standard deviation change in the independent variable.
Robustness
We report robustness tests in Appendix F and show that our results are robust in several different ways. This includes ruling out spillovers among those with the highest political knowledge and running separate models for each implication measure. We also account for don't know responses, scandal fatigue, potentially endogenous controls and boundary changes between the 2005 and 2010 general elections.
One specific concern we might have is that because many MPs were implicated in the expenses scandal, people's diffuse support may have responded only to the national scandal context. That is, trust might have fallen so low that people had no room to further update their trust in response to their local MP's behaviour. But this does not seem to have been the case. While cross-sectional surveys show that the national average of political trust fell after the scandal, the drop was not especially large and it occurred in a context of long-term decline. Accordingly, Curtice and Park (Reference Curtice and Park2010, 17) summarize the public response to the scandal by stating that ‘trust was clearly in decline long before the MPs’ expenses scandal broke … nevertheless, the expenses scandal appears to have helped erode trust yet further’. Although the proportion of people who thought that British governments of any party could be trusted ‘most of the time’ fell by 13 points between 2007 and 2009, nearly 60 per cent of those surveyed in 2009 still thought that the British government could be trusted at least ‘some of the time’ (Curtice and Park Reference Curtice and Park2010). It, therefore, appears unlikely that the post-scandal drop in trust was so large as to prevent people from also considering their local MP when forming their political trust.
To illustrate this, in Fig. 4, we track satisfaction with democracyFootnote 5 using the 2005 to 2010 BES panel. By using panel data, we ensure that diffuse support levels are calculated using responses from the same people in each wave, and we can compare the change in diffuse support following the scandal among those who believed that their MP was implicated with that, among those who did not.Footnote 6 We plot mean levels of satisfaction over time for those two groups. This allows us to contextualize any constituency-level spillover within the broader national scandal by comparing the change in diffuse support among those who believed their MP was implicated compared with those who did not.
The effect of the national scandal context on diffuse political support is small.
Note: Sample means with 95 per cent confidence intervals from the BES 2005–2010 nine-wave panel survey. Respondents split based on their perception of whether their MP was involved in the expenses scandal when asked in the 2010 pre-campaign wave.
The change between 2008 and 2009 for both groups is essentially zero and there is little difference in the change between those who believed their MP was implicated and those whose MP was not implicated. In fact, those who believed their MP to have been involved in the scandal had long been less satisfied with British democracy, suggesting that diffuse support is to some extent dispositional. Overall, therefore, we find little evidence here that the national average in political trust fell after the expenses scandal, and even if it did in fact fall, there appears plenty of room for people to incorporate their views about their local MP's involvement into their trust.
Conclusion
The expenses scandal is a strong non-experimental test of whether scandalous behaviour by a voter's constituency representative affects their diffuse political trust. What did we find? First, voters' attributions of MPs' involvement respond to both media and financial implication, and implication affects voters' specific support for their MP. Second, media implication is much more strongly associated with voters' attitudinal responses than financial implication. This suggests that the media substantially shapes voters' perceptions of political scandals. Third, and most importantly, for the type of scandal considered in this study, scandalous behaviour by a voter's own representative did not affect their diffuse trust. This null finding is robust to various operationalizations of trust and implication and we also find little evidence of a national-level effect.
Overall, our findings suggest that the spillover mechanism did not act at the constituency level during the expenses scandal. This is despite voters being aware of their local MP's behaviour. One explanation for this might be that voters anticipated being able to punish highly implicated MPs, as indeed they later did (Eggers Reference Eggers2014).Footnote 7 Future research could, therefore, test whether misconduct by one's constituency representative has a larger effect on diffuse trust in electoral systems with little direct constituency representation, in which voters may be less confident in their ability to punish their representative. In addition, our results speak to a financial scandal implicating the majority of the political class. While we believe that the expenses scandal is a most likely case for spillover, further research could test whether our results also hold for other types of scandal.
Another explanation for the lack of spillover is that any effect might be concentrated among the most politically knowledgeable (von Sikorski et al. Reference von Sikorski2020). Although we found no evidence of spillovers onto diffuse political trust among the highly knowledgeable, informed partisan opponents of implicated MPs may assimilate perceived malpractice into their evaluation of the opposing party rather than diffuse categories such as ‘politicians’. It is, therefore, possible that spillover effects are concentrated in partisan spillover about opposing political parties even during a relatively nonpartisan scandal. Further research could build on this by testing whether constituency-level scandal spillovers are minimized by partisan blame-shifting and assimilation among those with sufficient knowledge.
These thoughts aside, our findings imply that while voters may be aware of their MP's bad behaviour, and this affects voters' attitudes toward their MP (especially if their MP is implicated in a scandal by the media), the type of scandalous behaviour considered in this study did not influence voters' diffuse political trust. It appears, therefore, that if similar scandals affect political trust, this is unlikely to occur via voters' judgements about their own representatives.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0007123424000383.
Data availability statement
Replication data for this article can be found in Harvard Dataverse at: https://doi.org/10.7910/DVN/MJDFG8.
Acknowledgements
We are grateful to Andy Eggers and Scott Graffin for sharing their data. We would also like to thank Karina Cendon Boveda, Nick Vivyan, and Markus Wagner for their comments on earlier versions of this work. Finally, we are grateful to the anonymous reviewers and the editorial team for their constructive feedback and guidance. Any remaining errors are our own.
Financial support
None.
Competing interests
None.
Open access
