The ethical conduct of political science research is increasingly contested, with researchers, professional associations and regulators having to chart a perilous course between respect for the autonomy and dignity of the subject and the need to carry out independent and innovative studies. Experiments on elites, such as elected officials, pose additional challenges. There should be no distinction between the conduct of these experiments and those with other research subjects, but they differ because politicians are powerful. Politicians can communicate their disapproval to funding bodies and universities, potentially blocking or detering such research. In response, it has been countered that the power of elites should not be a factor in deciding whether to carry out these kinds of studies (McClendon Reference McClendon2012, 17). Researchers should be able to speak truth to power. But it is hard to deny that experiments may be seen as intrusive, especially when they involve identity deception. There is also the need to ensure that elites maintain their trust in research from the academy, which can be undermined by studies that appear to deceive public officials. Partly because of these debates, researchers have moved away from researcher-determined designs to more collaborative forms of endeavour (Butler Reference Butler2019; Loewen and Rubenson Reference Loewen and Rubenson2022). In such partnerships, politicians and other elites can find out about the advantages of research done on themselves, which can assist in their quest for accountability.

Such issues emerged discernibly over the research project led by Rosie Campbell in 2021 (Campbell and Bolet Reference Campbell and Bolet2022). This study randomly assigned e-mails from different kinds of fictitious constituents to UK Members of Parliament (MPs), using a method pioneered in the USA by Butler, Broockman and others (Butler and Broockman Reference Butler and Broockman2011; Butler and Nickerson Reference Butler and Nickerson2011; Butler Reference Butler2014), and widely applied in many other jurisdictions (McClendon Reference McClendon2016; Vries, Dinas, and Solaz Reference Vries, Dinas and Solaz2016; Habel and Birch Reference Habel and Birch2019; Crawfurd and Ramli Reference Crawfurd and Ramli2021). The difference to other studies was the requirement by the university’s Research Ethics Committee to carry out a debriefing after the experiment had taken place. Once MPs found out about the research from the debrief, some bitterly complained about being deceived, also saying that it was a waste of their staff’s time. There was a media and Twitter storm (see: https://www.bbc.co.uk/news/uk-politics-56196967). The Speaker of the House of Commons wrote to the Principal of King’s College London, where Campbell is based, and to the executive chair of the Economic and Social Research Council, which funded the study. The question this case poses is how best to achieve the balance between respect for the subject and freedom to carry out research.

As well as politicians, regulators and researchers, there is another important stakeholder whose views might help resolve the conundrum of agreeing ethical designs on elected officials: citizens. Addressing the views of the wider public has become a strong theme in research on human subjects in the biosciences in what is called ‘empirical ethics’ (see Borry, Schotsmans, and Dierickx Reference Borry, Schotsmans, Dierickx, Widdershoven, McMillan, Hope and van der Scheer2008 and Appendix A). It also matters what citizens think of experiments designed to improve knowledge about elected representatives, given that MPs are accountable to them. As researchers need to make choices about how to carry out elite experiments, the public’s view of the use of identity deception counts, especially when compared to other designs, such as the use of confederates, that is recruiting real constituents who are asked to contact MPs. The public might view the use of deception as not adhering to standards of fairness and fair dealing. On the other hand, the public has become critical of politicians and their motives in recent years (Clarke et al. Reference Clarke, Jennings, Moss and Stoker2018). Beliefs in ‘anti-politics’ and distrust of politicians might encourage citizens to approve of stronger review and audit.

Researchers have already started along this path with Desposato’s (Reference Desposato2018) study of the North American public’s view of correspondence audit experiments, especially when they involve deception, and Naurin and Öhberg’s (Reference Naurin and Öhberg2021) comparison of the views of Swedish voters and officials. Desposato (Reference Desposato2018) surveyed US citizens and scholars about their perspectives on the use of deception (i.e. ex ante consent, identity deception and no consent, and identity deception and debrief to gain post-hoc consent) as well according to the target object group (i.e. politicians, business owners or private homeowners) and purpose of research (i.e. assessing discriminatory behaviour, communications, customer service or constituency service). When asked about the vignette’s acceptability, both types of respondents expressed negative reactions to experiments carried out without consent and to all forms of deception. Naurin and Öhberg (Reference Naurin and Öhberg2021) performed a similar survey experiment in Sweden, investigating politician and citizen perceptions on research ethics and experimentation. Respondents were asked, ‘To what extent would you find the following things to be ethically problematic if you were asked to participate in a survey addressed to you/to you in your capacity as politician?’ and asked to rate certain research practices within a hypothetical survey on a scale of 1 – Yes, very problematic, to 7 – no completely unproblematic. Politicians tended to rate each prompt as more ethically problematic than citizens, which challenges the common assumption of politicians being ‘less sensitive to experimental designs than ordinary citizens because they are used to being scrutinized by the media, voters, opponents, and others’ (Naurin and Öhberg, Reference Naurin and Öhberg2021, pp. 890–891).

The research for this paper extends this line of research to the UK context, evaluating Campbell’s design and possible alternatives. We presented randomly assigned scenarios to a representative sample of UK residents, varying the use of identity deception, debriefing and the recruitment of confederates. Following Desposato, we hypothesised that the public would be more approving of scenarios involving the recruitment of real constituents than those where identity deception is used (H1). We also wanted to capture the idea – implied by Desposato’s research – that the more overt the deception, the greater the public’s disapproval. We thus concluded that the public would find research designs carried out by a researcher from a university more acceptable than an investigation by a journalist using deception (H2). We considered that there was less justification for other forms of deception on politicians than from an independent/official researcher, which have a serious purpose and where safeguards are in place. Also, collaboration with MPs would be seen as preferable to deception, but not approved as much as using confederates (H3). Causing upset among politicians after a debrief would be seen as less acceptable than when there is just identity deception and a debrief (H4). Not debriefing would have less acceptability than deception with debriefing (H5) and the use of confederates with debriefing (H6). There were also exploratory hypotheses based on the anti-politics literature: working class and Conservative voters would be more supportive of measures to hold MPs accountable. The wording of these hypotheses, pre-registered at the Open Science Framework (OSF),Footnote ¹ is reproduced in Appendix B.

Research methods

We developed seven scenarios to test the hypotheses (see Table 1 for the summary of the scenarios and Appendix C for their full wording). One of these (A3) was designed to replicate Campbell’s research design with the debrief and mentioning the furore of the MPs, contrasting with other designs just using identity deception (A1) and one adding the debrief (A2). There were two scenarios with confederates, one without (B1) and the other with the debrief (B2). Scenario C was a collaboration with MPs also with identity deception. To test Hypothesis 2, we included a different kind of scenario: a journalist posing as a constituent to obtain a story from a MP when off-guard. It was loosely based on a real case of the senior Liberal Democrat MP, Vince Cable, who was taped by two journalists posing as constituents (The Daily Telegraph, 22 December 2010). The scenario provided a baseline from which to compare the others (without being a placebo).Footnote ²

Table 1 Summary of Scenarios

We carried out cognitive interviews (Miller et al. Reference Miller, Chepp, Willson and Padilla2014) to test the wording of the scenarios and to ensure that respondents understood them (see Appendix D). As a result, the scenarios were redrafted with more ordinary-sounding language.

Outcome variables are the extent to which the research is regarded as acceptable, whether the respondent would like their own MP in the research, and the extent to which the research is seen to be socially valuable. Covariates, which were either supplied by the survey company or came from the survey itself, are gender, age, ethnicity/race, education, religiosity, region, social grade, vote in 2019 and 2021 and income. We asked standard attitudinal questions on political interest and efficacy, which were used as additional independent variables (see codebook in Appendix F).

The survey was carried out by Deltapoll, a quota sample drawn from its panel of just over 750,000 UK adults. The quotas were age, gender, region, past vote and EU referendum vote. The survey launched on 13 December 2021, yielding 8,040 respondents (for data and supporting files, see John et al Reference John, Kim and Soto Tamayo2022). A summary table of the demographic variables is contained in Appendix G. The sample is representative on the main demographic variables but is slightly short of Conservative-supporting voters. There is a weight that can deal with the lack of filling the quota, but the tables in the paper are unweighted (key weighted tables are produced in Appendix H). Appendix I reports balance tests, which show equivalence across the scenarios and no more significant terms than would have occurred by chance.

Appendix J reports the manipulation checks. Respondents were able to detect differences between the scenarios, with the percentage correct within each treated group ranging from 34.0 to 47.9%. Some respondents wrongly attributed aspects of these scenarios even when they were not presented with these features, varying from 7.8 to 18.3%. In contrast, 61.9% recognised the journalist scenario, with 13.3% of the rest of the sample who were not shown this scenario believing the journalist was part of theirs.Footnote ³

Responses to the qualitative analysis (see Appendix K) give added support to the internal validity of the experiment. Respondents were asked: ‘In more than 10 words tell us how you feel about this study?’ and were given a textbox to write their answers. After developing a code-frame, we coded a random sample of 2,000 responses into 15 categories, which showed respondents engaging with the scenarios. For example, 18.3% were coded to the category ‘Honest response from MP/Holding MPs accountable’, 6.1% to ‘Negative response to deception’ and 4.0% to ‘Interested in potential results of study’. The categories that showed no engagement only took up a small proportion of responses: 3.2% ‘Don’t understand/confused’ and 5.8% ‘Other’ (see Table K2).

Results

Figure 1 displays the levels of approval for the different scenarios, with 95% confidence intervals.

Figure 1 Citizen Views of the Acceptability of Research on Politicians.

With responses to the scenarios averaging at 5.5 on the seven-point scale, we infer that the UK public approves of studies on MPs. There are no significant differences of views of the acceptability between the experimental scenarios, showing that the public neither distinguish between those based on identity deception, confederates and collaboration nor between studies that use debriefing and those that do not. The only scenario showing a strong difference is D, the journalist story. Here, the public were less approving, with a mean of 4.66 on the seven-point scale, 0.97 points (p < 0.0001) below the combined mean of 5.62 of the other groups (see Appendix L). We do not confirm any of the hypotheses except H2. Regression analysis shows that support for these measures is driven by political interest and efficacy, not demographics (see Appendix M).

Table 2 shows the results for the question, ‘Suppose you learned that a study like the one described before had been carried out in your community with your MP. Which of the following best describes how you would feel about the MP being included in the study?’

Table 2 Feelings About Their MP Being Included in the Study

The first row presents frequencies. The second row contains column percentages.

Between 46.4 and 62.8% of respondents would be glad their MP was included in these scenarios, which is quite high given ethical concerns about deception. Between 8.4 and 16.1% would rather their MP did not participate, with between 22.2 and 29.7% not caring either way. Differences between most of the experimental research scenarios are not significant (between A1 and A2, A2 and A3, B1 and B2, and between C and the others). As before, the big difference is between Scenario D and the rest, with 46.4% glad their MP is in the study, compared to 60.4 average for the other scenarios (p > 0.00001). There is, however, a statistically significant 2.7 percentage point difference (p > 0.05) between respondents being glad their MP is in scenarios that use identity deception (A1, A2, A3 and C) and those that do not, the confederate studies (B1, B2).Footnote ⁴ Rather than confirming our expectation that the public would display greater disapproval of scenarios using deception, the relationship is in the opposite direction.

In Appendix N, we explore why respondents respond more positively to the deception conditions. Although we did not propose this estimation in the pre-analysis plan, it is in the spirit of our exploratory Hypothesis 7: lower educated groups will show greater approval of Scenario A1 and Scenario A2 than other groups, with Conservative voters being more willing to appreciate interventions using deception. Overall, the interaction analysis does not provide evidence that an anti-politics mentality among citizens might be a cause of liking more robust measures of holding politicians to account, which include identity deception. Rather, those who are engaged with politics are more likely to be content their MPs are in these studies. The visual presentations of the marginal plots of the interaction between political interest and the use of deception remain the same with no interaction at high levels of interest and interaction at low levels: see Figure 2 (left panel). There is a similar outcome in seeking to evaluate the difference between Scenario D and the rest of the treatments on the main outcome variable of acceptability. Figure 2 (right panel) shows that the journalists’ scenario is less approved than others, and that this approval is higher within those with more interest in politics (see also regression in Appendix O). In addition, being a co-partisan with the respondent’s MP might have caused respondents to be more protective of their MPs being in these studies, but the term is not significant and has the opposite sign (Appendix N, Table N3).

Figure 2 Impact of Political Interest on Feelings About Their MP Being Included in the Study.

The final outcome measure is derived from the question, ‘To what extent do you think that this study is worthwhile to carry out?’, with responses coded from 7 highly worthwhile to not at all worthwhile 1. The mean response is 3.4, which shows respondents are at the mid-point. As before, there were no significant differences between the scenarios, except between D and the others, which scored a mean of 3.82 (p < 0 .0001). We also asked, ‘To what extent do you think the study shows how a MP would be likely to answer real e-mails from local residents?’ with responses coded from 7 very likely to not at all likely 1, recoding a mean of 4.2 indicating likelihood. There are no differences in responses to this question, even for D, which scores 3.35 (p > 0.1).