Introduction
Political discussion requires that, beyond stating the positions they hold, people articulate reasons for their policy preferences. Reason-giving is central to contemporary accounts of liberal theory (Habermas Reference Habermas2015; Rawls Reference Rawls1997; Chambers Reference Chambers2010) and deliberative democrats have argued that the public exchange of reasons between individuals can ‘change minds and transform opinions’ (Chambers Reference Chambers2003, 318; see also Dryzek Reference Dryzek2002; Cohen Reference Cohen2005; Thompson Reference Thompson2008; Mutz Reference Mutz2008; Gutmann and Thompson Reference Gutmann and Thompson2009). In addition to the attitudinal effects of interpersonal deliberation, scholars have also speculated that justifying one’s political attitudes may also induce a greater degree of ‘internal-reflective’ (Goodin Reference Goodin2000) deliberation and introspection which, in turn, might affect the content of political attitudes. For instance, deliberative processes in which the weighing of reasons ‘take[s] place within the head of each individual’ (Goodin Reference Goodin2000, 82) are thought to affect ‘how we decide what position to take’ (Goodin and Niemeyer Reference Goodin, Simon and Niemeyer2003, 629). Similarly, Cohen (Reference Cohen2007, 228) suggests that ‘the practice of defending proposals with reasons may change my preferences’. However, while voters are able to provide substantive reasons in defence of their preferences (Colombo Reference Colombo2018, Reference Colombo2021), very little existing research evaluates whether and how introspective reason-giving affects the attitudes that voters express. This is a significant omission given that reason-giving is considered to be the ‘first and most important characteristic’ of deliberative democracy (Gutmann and Thompson Reference Gutmann and Thompson2009, 24).
How might reason-giving affect attitudes? On the one hand, models from political behaviour hold that voters’ reasons play a causal role in the construction of their attitudes (for example, Zaller Reference Zaller1992; Zaller and Feldman Reference Zaller and Feldman1992). Reasons-as-causes models of this form highlight that, by encouraging them to introspect about their preferences, reason-giving might change the set of reasons that voters consider, and thereby affect the attitudes they express. This model of attitude formation leads to three specific expectations. First, reason-giving might increase the temporal stability of voters’ attitudes by reducing judgments made on the basis of idiosyncratic, ‘top of the head’ considerations. Second, reason-giving might increase the ideological constraint of voters’ attitudes by highlighting substantive connections across issues. Third, reason-giving might reduce issue-based polarization if deeper introspection reduces the variance of voters’ expressed attitudes, or if it encourages voters to consider arguments that support positions other than their own, thereby reducing voters’ attitudinal extremity. On the other hand, models from social and political psychology suggest that reasons are used only to justify attitudes after they have been adopted (for example, Lodge and Taber Reference Lodge and Taber2013; Mercier and Sperber Reference Mercier and Sperber2018; Haidt Reference Haidt2001). These reasons-as-rationalizations models imply that, if reasons are used to rationalize (rather than cause) political beliefs, there are few mechanisms through which introspecting about reasons might lead to attitude change. As a consequence, these models predict much more limited effects of reason-giving on political attitudes.
Theoretical disagreement about the predicted effects of reason-giving suggests a productive opportunity for empirical work. I implement an experimental design in which survey respondents report their preferences on a set of political issues. While half of the respondents provide only their policy preferences, the other half first provide the reasons that underpin their policy positions via an open-ended response, a treatment designed to increase cognitive effort and introspection. I evaluate the effects of reason-giving by measuring differences between treatment and control with respect to constraint (the correlation between respondents’ positions on different issues), stability (the correlation of respondents’ positions across survey waves), and polarization (the disagreement between respondents’ positions on a given issue).
Fielding this pre-registered experiment in a new two-wave panel survey of more than 4,000 UK citizens, I find that there are very limited effects of reason-giving on political attitudes. Despite some heterogeneity at the level of individual issues, reason-giving has precisely estimated null average effects on both the polarization and stability of voters’ attitudes, a finding that replicates across two survey samples. For constraint, attitudes are marginally more highly correlated across issues for the reason-giving respondents than for the control group in one sample of respondents but this finding does not replicate in a second sample of respondents. For all three outcomes, I show that the null average effects do not mask significant heterogeneity between different subgroups of voters and that these nulls are unlikely to be driven by a weak treatment. Taken together, the results demonstrate that providing justifications for one’s political attitudes has no appreciable effects on the stability, constraint, or polarization of public opinion.
These findings have important implications for our understanding of both deliberative democracy and the quality of voters’ political opinions. Deliberative democrats have invoked a wide variety of requirements for successful deliberation, including civility, face-to-face exchange, and equality of participation, in addition to reason-giving. While a number of studies demonstrate the broader effects of deliberation on voters’ attitudes (for example, Gastil and Dillard Reference Gastil and Dillard1999; Sturgis, Roberts and Allum Reference Sturgis, Roberts and Allum2005; Fishkin et al. 2020; Farrar et al. Reference Farrar, Fishkin, Green, List, Luskin and Paluck2010; List et al. Reference List, Luskin, Fishkin and McLean2013; Minozzi et al. 2023), the deliberative experiences that form the basis of these studies include highly compound treatments, where reason-giving is bundled together with many other features of deliberation. By contrast, this paper helps to open up the ‘black box of deliberation’ (Mutz Reference Mutz2008, 531) by demonstrating that one particularly important component of deliberative practice – the articulation of reasons – has essentially no effect on the attitudes that voters express. As others have noted, developing empirical evidence on the consequences of specific components of deliberation is an important endeavour, which ‘greatly enhance[s] the capacity of deliberative theory to contribute to democratic society’ (Mutz Reference Mutz2008, 531).
In particular, the results here speak to the relative efficacy of public versus private deliberation. For many deliberative democrats, interpersonal exchange is critical for realizing the benefits of deliberation (for example, Dryzek Reference Dryzek2002; Rawls Reference Rawls1997; Gutmann and Thompson Reference Gutmann and Thompson2009; Cohen Reference Cohen2005). For others, by contrast, deliberation between people is just one mechanism by which voters might be encouraged to engage in ‘internal-reflective’ deliberation and it is in this introspective reasoning that the true value of deliberation resides (Goodin Reference Goodin2000; Goodin and Niemeyer 2003). One motivating factor for studying the internal-reflective form of deliberation is that interpersonal exchanges of the type that occur in citizen assemblies, citizen juries, and other deliberative experiences are hard to scale to large populations. If the types of attitudinal change associated with interpersonal deliberative experiences could be achieved by people deliberating alone, then the benefits of deliberation might be more easily scaled to more people. As Goodin (Reference Goodin2000, 84) suggests, internal-reflective deliberation may therefore ‘relieve many of the burdens plaguing external-collective deliberation in modern mass societies’. Empirical studies of the effects of internal-reflective deliberation are, however, rare.Footnote 1 My results suggest that – at least with respect to political attitudes – solitary reason-giving does not have effects equivalent to public deliberation. As a consequence, the benefits ascribed to broader deliberative practices must therefore be generated by components of deliberation other than the private, internal form of reason-giving studied here.
Finally, testing the expectations generated by the reasons-as-causes model is important because of the optimistic view that model implies about voters’ capacity to hold well-structured political preferences. Decades of survey research have painted a pessimistic picture of the stability, constraint, and polarization of voters’ attitudes (for example, Converse 1964; Achen Reference Achen1975; Zaller and Feldman Reference Zaller and Feldman1992; Ansolabehere, Rodden and Snyder Reference Ansolabehere, Rodden and Snyder2008; Freeder, Lenz and Turney Reference Freeder, Lenz and Turney2019; Abramowitz and Saunders Reference Abramowitz and Saunders2008; Fiorina and Abrams Reference Fiorina and Abrams2008; Mason Reference Mason2015). Deficiencies in these dimensions pose an obvious normative threat: if voters hold unstable, incoherent and extreme policy views, then their ability to hold politicians accountable is consequently diminished (see, for example, Achen and Bartels Reference Achen and Bartels2017, 306). The hopeful suggestion generated by the reason-as-causes model is that the quality of voters’ attitudes might increase if only voters could be induced to ‘think harder’ about their political opinions, a suggestion buttressed by evidence from social psychology that shows greater introspection can indeed lead to more stable, more coherent, and less polarized attitudes in other domains (for example, Petty and Briñol Reference Petty and Pablo Briñol2011; Tesser Reference Tesser1978; Wilson et al. Reference Wilson, Lisle, Schooler, Hodges, Klaaren and LaFleur1993; Wilson, Kraft and Dunn Reference Wilson, Kraft and Dunn1989; Dijksterhuis Reference Dijksterhuis2004). However, I show that increased cognitive effort does not result in such salutary effects for three important measures of political attitude quality (Price and Neijens, Reference Price and Neijens1997). Notwithstanding any intrinsic value of reason-giving, the results here therefore imply that more introspective and reason-based processing of political issues is unlikely to act as a panacea to the problem of low-quality democratic attitudes.
Reason-Giving and Political Attitudes
Reason-Giving and Normative Democratic Theory
In liberal democratic theory, reason-giving is typically seen as a mechanism through which political legitimacy is achieved and the ideals of mutual respect and the equality of persons are manifested (Rawls Reference Rawls1997; Chambers Reference Chambers2010; Habermas Reference Habermas2015). The centrality of the ‘reason-giving requirement’ (Gutmann and Thompson Reference Gutmann and Thompson2009, 24) in deliberative democracy, for instance, stems from the idea that presenting and responding to public reasons is the ‘primary conceptual criterion for [political] legitimacy’ (Thompson Reference Thompson2008, 504). In addition to this intrinsic virtue, however, deliberation is also thought to ‘change minds and transform opinions’ (Chambers Reference Chambers2003, 318), and reason-giving is seen as a central mechanism through which such effects might operate.
For many scholars, deliberation is thought to affect preferences through the public and social exchange of views between different people (for example, Dryzek Reference Dryzek2002; Cohen Reference Cohen2005). However, other scholars have focused on the ‘internal-reflective’ nature of deliberation in which the weighing of reasons ‘ultimately must take place within the head of each individual’ (Goodin Reference Goodin2000, 81). From this perspective, any process in which voters are induced to be more reflective about their political positions can therefore be considered deliberative, regardless of whether such processes include the public exchange of reasons (Goodin and Niemeyer Reference Goodin, Simon and Niemeyer2003, 629). Public reason-giving might be one way of inducing internal reflection, but it is not the only way. As Goodin (Reference Goodin2000, 95) argues, ‘sometimes ‘answering to oneself’ might suffice’. The key insight from this work is that solitary reason-giving might affect preferences via the internal process of introspection it engenders in voters, rather than through a public process in which voters exchange reasons with one another.Footnote 2
However, beyond the broad hypothesis that reason-giving might affect attitudes, work in normative theory provides few operationalizable predictions about the effects of reason-giving on specific attitudinal outcomes. In part, this is because these accounts do not (and were not designed to) clearly articulate the psychological mechanisms through which reason-giving might lead to attitude change. In the next section, I contrast two models of attitude formation which take different perspectives on the role that reasons play when voters think about politics and which generate different expectations for the effects of reason-giving on political attitudes.
Expected Effects of Reason-Giving on Political Attitudes
Reasons-as-causes models assume that voters form attitudes by averaging over a set of reasons relevant to a given issue and that reported attitudes are determined by those reasons. Crucially, in this perspective, reasons play a causal role in opinion formation: if the set of reasons that a voter considers on a given issue changes, then the voter’s opinion on that issue may also change. The idea that attitudes are causally determined by aggregating across reasons is shared by many accounts,Footnote 3 but the most prominent example of such an argument comes from Zaller (Reference Zaller1992).Footnote 4 Zaller (Reference Zaller1992) suggests that voters have in their heads a distribution of potentially competing ‘considerations’ from which they sample stochastically when prompted to express their political opinions on a given subject. Attitude reports do not therefore represent the considered opinions of voters on particular issues, but rather reflect the outcome of a process in which voters average over those sampled considerations and make choices ‘in great haste – typically on the basis of the one or perhaps two considerations that happen to be at the ‘top of the head’ at the moment of response’ (Zaller Reference Zaller1992, 36). The critical assumption here is the idea that voters draw a sample of reasons each time they are required to produce a political opinion, and it is from this sample they then construct their attitudes. This assumption drives many of the predictions derived below, as the additional cognitive effort that reason-giving induces is expected to affect attitudes by changing the sample of reasons that voters consider.
By contrast, reasons-as-rationalizations models see political attitudes as deriving from fast and intuitive processing in which explicit reasoning plays a very limited role. Lodge and Taber (Reference Lodge and Taber2013), for instance, argue that voters do not consider and evaluate political arguments and justifications in order to form preferences, but rather that voters’ attitudes arise from spontaneous and affect-driven processes, which are entirely unrelated to the evaluation of specific reasons. The idea that people will provide evaluations without engaging in a cognitive reasoning process is also common in both social (for example, Mercier and Sperber Reference Mercier and Sperber2018) and moral (for example, Haidt Reference Haidt2001) psychology. Even when voters have the time, motivation, and opportunity to engage in deliberative reasoning, these perspectives suggest that the process of reasoning will itself be biased by the valence of the initial affect towards a given issue. In these models, then, reasons are used by voters to rationalize their intuitively formed attitudes. As Mercier and Sperber (Reference Mercier and Sperber2018, 112) suggest, reasons do not ‘motivate or guide us in reaching conclusions’ but rather ‘justify after the fact the conclusions we have reached’. Reasons-as-rationalizations models therefore differ sharply from reasons-as-causes models, as the causal path connecting reasons to attitudes runs in reverse: people produce reasons to support the attitudes they intuitively adopt, rather than constructing their attitudes from the reasons they hold.
In general, both approaches understand attitude formation as a fast and constructive process in which attitudes are generated at the moment of response, rather than existing as a fixed point in voters’ minds. In this sense, both approaches suggest that the slow, deliberative, and conscious evaluation of reasons (‘System-2’ thinking) is likely to be rare, with most attitudes forming as a result of fast, automatic, and unconscious processes (‘System-1’ thinking). Where the approaches disagree, however, is in the mechanisms by which the process of attitude construction occurs. The critical distinction is that while reason-as-causes models suggest a cognitive process in which reasons are aggregated to form attitudes, reasons-as-rationalizations models suggest an affect-based process in which attitudes are adopted spontaneously without any evaluation of specific reasons. These differences in perspective about the internal workings of the attitude formation process are relevant because they imply very different predictions for the distribution of attitudes when voters are induced to engage in slower, more effortful contemplation.
What do these models predict for the effects of reason-giving on political attitudes? First, reason-as-causes models suggest that reason-giving might affect the stability of voters’ attitudes. If, following Zaller, voters form attitudes by sampling from a population of reasons, then the variance of voters’ attitudes will be lower when the voter draws a larger sample of considerations (Zaller, Reference Zaller1992, 86).Footnote 5 As a consequence, we should expect attitudinal instability – the degree to which voters’ attitudes change over time – to be lower in contexts where they are induced to think about a wider range of considerations related to a given policy. Zaller argues that the key to increasing the number of considerations used in forming attitudes is increased engagement or ‘extra thought’ (Zaller Reference Zaller1992, 86) about a given issue. A similar argument can be found in the ‘elaboration likelihood model’ of attitude change (for example, Petty and Briñol Reference Petty and Pablo Briñol2011), in which attitude strength, stability, and coherence are seen as a function of the amount of thought that people devote to a given attitude object. Therefore, if reason-giving provokes voters to ‘slow down and re-examine his or her line of thought’, Mansbridge (Reference Mansbridge2007, 262), then we should expect justification-providing voters to express more stable attitudes than voters who are not asked to provide reasons for their attitudes.
Second, reason-giving might also increase the correlation between attitudes on different political issues – a quantity typically referred to as attitude constraint (Converse 1964). One key mechanism driving this prediction is again that the sampling variation of attitudes will be related to the effort exerted in searching for reasons. The correlation between voters’ expressed attitudes on different issues will be biased towards zero when the variance of those attitudes is high. Therefore, if reason-giving induces voters to consider a larger number of reasons when constructing attitudes, their expressed attitudes will be less variable, and the correlation of their attitudes across issues will increase.
A second, more substantive, mechanism linking reason-giving to constraint is that explicitly stating justifications might also make voters aware of conceptual links across different issues, thus inducing them to express more correlated attitudes (for example, Keating and Bergan Reference Keating and Bergan2017). For instance, if a voter believes that ‘the poor don’t have enough to get by’ is an important justification for their support for a higher tax rate on high-income individuals, then the articulation of that belief might encourage them to recognize the potential validity of the same justification when considering a subsequent question about unemployment benefits. Similarly, if a voter believes that ‘individuals should be free to make their own choices’ is a valid defence of their views on free speech, articulating that justification might make it a more prominent feature in determining their attitudes towards transgender rights. If voters who think about reasons are more likely to make connections between issues that have common underpinnings, they may, therefore, be more likely to express correlated views on those topics.
Finally, reason-giving might also affect the polarization of voters’ attitudes. I conceptualize polarization as the extent of disagreement between voters’ issue positions on a given issue. Decreases in polarization might result from different mechanisms. First, reason-giving could – à la Zaller – increase the number of sampled considerations and reduce the variance of expressed attitudes which would, in expectation, result in less polarized attitudes across voters on a given issue. This moderating effect occurs purely as a result of the reduced variability in attitudes that comes from averaging over a larger set of considerations. Second, engaging in reason-giving might also induce voters to consider the arguments on the other side of the issue more carefully, thus encouraging them to take a more moderate position on the issue. This idea is central to many ‘perspective-taking’ accounts of political moderation, which suggest that understanding the experiences and perspectives of political opponents can durably reduce political polarization (Kalla and Broockman Reference Kalla and Broockman2022; Kalla and Broockman Reference Kalla and Broockman2020; Broockman and Kalla Reference Broockman and Kalla2016).
The common logic underpinning the expectations from the reasons-as-causes model is that reason-giving might change the set of considerations that voters use to construct their attitudes. These expectations are substantively important because they imply that three key properties of attitude quality might be improved simply by voters exerting a greater degree of cognitive effort. What, then, does the reasons-as-rationalizations model predict for the effects of reason-giving on attitudes? For the most part, this perspective suggests that reason-giving should have little or no effect on expressed attitudes. If attitudes are determined by affective, intuitive, and unconscious responses to external stimuli, and reasons are used only to post-hoc justify spontaneously generated feelings, then this considerably weakens the mechanism through which thinking about and articulating those reasons can lead to attitude change. Crucially, for these accounts, any cognitive reasoning process about an object will be biased by the initial affective response to that object, which reduces the probability that introspection about reasons will shift attitudes. As a result, we should expect introspective reason-giving to have very limited effects on attitudes.
Nevertheless, the reason-as-rationalization perspective has nuanced predictions for the effect of reason-giving on polarization. If people reason in a biased manner, their initial affective reactions might be further reinforced by the accumulation of reasons that align with that response. As a consequence, such voters might develop greater confidence in the attitudes they express as the reasons drawn to mind could justify and validate their initial intuitive responses. This type of biased processing may also lead voters to adopt more extreme views. For instance, if the reasons a voter recalls all align with the particular side of a debate to which the voter is intuitively attracted, considering those reasons might prompt them to reconsider their initial response as being too moderate and encourage them to take a more extreme position on that issue (Tesser Reference Tesser1978, 310). Under the reasons-as-rationalizations model, then, reason-giving should be expected to have either null effects on aggregate polarization (if reasoning only increases attitudinal confidence), or positive effects on polarization (if reasoning increases attitudinal extremity). Importantly, both of these predictions differ from those derived from the reasons-as-causes model, which implies that introspective reason-giving will reduce attitudinal polarization.
In the context of the typical survey response, both models suggest that voters make fast, and largely unconscious, judgements. For reasons-as-causes models, these judgements arise via a fast and shallow sampling of reasons, which are then used to determine their choices, while for reasons-as-rationalizations models, they stem from intuitive, affect-based, and spontaneous responses. As a consequence, both models are consistent with commonly observed response patterns in many political surveys in which voters’ attitudes are marked by low levels of constraint and stability and high levels of polarization. However, contrasting predictions of these models arise when considering the expected effects of increased effortful thinking: where the reasons-as-causes approach assumes that such effort will change the set of considerations brought to mind and, therefore, the resulting attitudes that voters express, the reasons-as-rationalizations approach assumes that effortful thinking will produce reasons that justify the initial affective response of the voter and will have few consequences for expressed attitudes.
Empirical Evidence on the Effects of Reason-Giving
Existing evidence from social and cognitive psychology suggests that engaging in processes of reasoning can affect the attitudes people endorse (for example, Tesser Reference Tesser1978). In particular, introspecting about reasons appears to affect the decisions that people make and the satisfaction they subsequently feel from those decisions (Wilson, Kraft and Dunn Reference Wilson, Kraft and Dunn1989; Wilson and Schooler 1991; Wilson et al. Reference Wilson, Lisle, Schooler, Hodges, Klaaren and LaFleur1993; Dijksterhuis Reference Dijksterhuis2004; Simonson Reference Simonson1989; Hsee Reference Hsee1999). The broad conclusion of this literature is that ‘people who reason more act differently from those who reason less or not at all’ (Mercier and Sperber Reference Mercier and Sperber2018, 253). However, these studies do not directly address reason-giving as a specific mechanism for attitude change. Moreover, many of these papers focus on consumers’ choices, which limits the degree to which they are informative about political attitudes.Footnote 6
In political science, voters participating in interpersonal deliberative forums develop attitudes that are more ideologically constrained (Sturgis, Roberts and Allum Reference Sturgis, Roberts and Allum2005; Gastil and Dillard Reference Gastil and Dillard1999) and less polarized (Fishkin et al. 2020), and also have preferences that come closer to demonstrating properties of single-peakedness (Farrar et al. Reference Farrar, Fishkin, Green, List, Luskin and Paluck2010; List et al. Reference List, Luskin, Fishkin and McLean2013) than voters who did not participate in those forums. However, the deliberative settings that underpin these studies represent highly compound treatments, as – in addition to reason-giving – participants also receive a great deal of policy-relevant information, engage in group-based discussion, cast votes for preferred outcomes, and so on. Therefore, while these studies are helpful for determining whether deliberation as a whole affects attitudes, they are not informative about the effects of individual elements of deliberation, such as reason-giving. If reason-giving is thought to affect attitudes in particular ways, the appropriate test is one which compares the views of those who engage in reason-giving to those who do not. As Mutz (Reference Mutz2008, 530) suggests, to understand the mechanisms that drive the effects of deliberation, we need to ‘identify which characteristics of deliberative practice produce which kinds of desirable outcomes’, a sentiment shared by many other scholars (for example, Gastil and Dillard Reference Gastil and Dillard1999, 21; Thompson Reference Thompson2008, 500–501).
In a recent study, Minozzi et al. (2023) focused specifically on evaluating the separate effects of public and private deliberation on a range of outcomes, such as knowledge gains, emotional reactions, and civic attitudes. Consistent with the results I present below, they found only limited effects of individual deliberation. However, the treatment that Minozzi et al. (2023) employed differs in important ways from the treatment I introduce below, most notably in that it did not require participants to engage in reason-giving. This, in addition to the fact that Minozzi et al. (2023) do not study the effects of individual deliberation on the quality of voters’ attitudes, suggests that further research into the specific effects of introspective reason-giving is warranted.Footnote 7
The study that comes closest to evaluating the effects of reason-giving on attitudes is by Zaller and Feldman (Reference Zaller and Feldman1992)Footnote 8 who randomly assigned some survey respondents to answer a ‘stop-and-think’ question which required them to report some relevant considerations before providing their views on a given issue. Consistent with the discussion above, Zaller and Feldman expected respondents in the stop-and-think condition to report attitudes that were more stable across survey waves and more highly correlated across issues. However, stopping-and-thinking increased ideological constraint only for respondents with high levels of political sophistication, while attitude stability was (insignificantly) lower in the stop-and-think condition than in the control condition (Zaller and Feldman Reference Zaller and Feldman1992, 605).
However, the experiment reported by Zaller and Feldman (Reference Zaller and Feldman1992) represents an incomplete test of the effects of reason-giving. First, the treatment administered by Zaller and Feldman (Reference Zaller and Feldman1992) was a thought-listing exercise,Footnote 9 which is conceptually distinct both from the treatment described below and from reason-giving as understood in the literature on deliberation. Second, the experiment was fielded as a part of the 1987 ANES pilot study to a very small sample of respondents (only 450 respondents in the first wave and 357 in the second), making the null results somewhat difficult to interpret. Third, their analysis focused on only three issues, which limits the generalisability of the findings. Finally, the response options available to respondents differed between the treatment and control groups, a decision that reintroduces the possibility of selection bias. As a result of these issues, Zaller (Reference Zaller1992, 91) concluded that the predictions of his model that relate to the effects of reasoning on constraint and stability ‘cannot be said to have been adequately tested’.
Experimental Design
In this section, I describe the design of a two-wave online panel survey which was fielded to UK respondents by Opinium in early 2022. All analyses described below were pre-registered with the Evidence in Governance and Politics (EGAP) registry (Blumenau Reference Blumenau2022).
Sample and Randomization
The first survey wave – fielded in January 2022 – consisted of 3,010 respondents, who were selected using nationally representative quotas for gender, age, and vote in the 2019 UK General election and political attention. In the first survey wave, respondents were randomly assigned into two groups with equal probability. Respondents in each group were asked to report their positions on four issues (sampled at random from a set of six issues, described below) in current UK politics. Respondents in the control group were only asked to provide their preferred policy option on each issue. Respondents in the treatment group were asked, before giving their policy preferences, to provide the reasons for their positions on each issue (prompt described below). After providing their reasons, treatment-group respondents then answered the same set of policy questions as the control group. I refer to results from the first sample of respondents in the first wave of the survey as ‘Sample One, Wave One’ results.
From the first wave, 2,545 respondents were successfully recontacted in the second survey wave, fielded in May and June 2022. These respondents were asked to provide their preferences (and, if in the treatment group, reasons) for the same set of political issues that they considered in wave one. The treatment assignment persisted across the two waves of the survey such that reason-giving respondents in wave one also provided reasons for their positions in wave two. This allows me to assess the extent to which repeated treatment exposure affects expressed attitudes. I refer to results from this set of respondents as ‘Sample One, Wave Two’ results.
In addition, the second wave also included 1,438 new respondents who did not appear in the first wave. These newly added respondents in wave two were also randomized into treatment and control groups with equal probability and followed the same survey as other wave two respondents (with the four issues sampled at random). This allows me to replicate two of the analyses (for constraint and polarization) on a fresh sample. I refer to results from this second sample of respondents as ‘Sample Two’ results.
Policy Areas
The six policies included in the experiment included a mix of high- and low-salience issues, including four broadly related to the economic ‘left-right’ dimension of UK politics (‘Unemployment Support’, ‘Higher Rate of Tax’, ‘Minimum Wage’, and ‘Zero hours contracts’) and two related to the social ‘liberal-conservative’ dimension (‘Transgender Rights’ and ‘Offensive Speech’). These issues also span a range of ‘easy’ (symbolic and easily communicable) issues and ‘hard’ (technical and complex) issues, attitudes on which are thought to be structured by different types of cognitive processes (Carmines and Stimson Reference Carmines and Stimson1980). Several of the policies were drawn from those used in Hanretty, Lauderdale and Vivyan (Reference Hanretty, Lauderdale and Nick Vivyan2020), while others were written to cover more recent topical issues in UK politics. Each respondent answered questions relating to four out of the six issues. Each issue was paired with a thematically similar issue (discussed below), and sampling was conducted at the issue-pair level, such that for each respondent two issue-pairs were sampled and respondents provided responses to all four issues.
The design is only sufficiently powered to detect relatively large treatment effects at the level of individual issues (see the Appendix, section B). An alternative design would have been to select a smaller number of issues and gather a larger number of responses for each of them. However, that approach would be subject to generalizability concerns, as any inferences would be limited to the specific issues included. Instead, I use a larger number of policy areas but focus on the average effect of the treatment across issues. Using a large set of policy issues maximizes the external validity of the experimental results while targeting the average effect of the treatment effect maximizes the power of the design (Blumenau and Lauderdale Reference Blumenau and Lauderdale2022).
Survey Prompts
Figure 1 provides an example of the open-ended reason-giving prompt displayed to respondents in the treatment group for the ‘Higher Rate of Tax’ issue. After a short introduction, respondents were asked to provide the reasons that supported their view on whether the government should increase or decrease the rate of income tax for high-income individuals. This prompt was designed to reflect how reason-giving is conceived in the theoretical literature and to provoke the type of introspection that the reasons-as-causes model predicts will be consequential. First, consistent with Mansbridge (Reference Mansbridge2007, 261), who argues that reason-giving ‘can include any statement that sincerely answers the ‘why’ question’, the prompt instructs voters to provide the reasons that they see as supporting their own position on the issue. Second, by asking respondents to ‘think very carefully’ about their own reasons, it provides a plausible inducement for respondents to engage in the type of ‘internal-reflective process’ that many scholars believe is a key mechanism linking deliberation to attitude change (Goodin Reference Goodin2000, 95; see also Bortolotti Reference Bortolotti2009; Cohen Reference Cohen2005; Goodin and Niemeyer Reference Goodin, Simon and Niemeyer2003). Finally, the prompt emphasizes that respondents should ‘explain as many reasons as possible for your view’, a phrase which directly attempts to manipulate the number of considerations that respondents draw into their minds at the point of attitude formation, something that is central to many of the predictions of the reasons-as-causes model (Zaller Reference Zaller1992; Zaller and Feldman Reference Zaller and Feldman1992).
Reason-giving prompt.
After providing justifications, the treatment group were asked to select the position closest to their own from five logically ordered alternatives (plus a ‘Don’t know’ response option). Figure 2 provides an example of the ‘Higher Rate of Tax’ issue. In this case, respondents could select a taxation rate for yearly incomes above £150,000, with options ranging from ten percentage points below to fifteen percentage points above the current status quo (45 per cent).
Issue position prompt.
Control-group respondents, by contrast, saw only the introduction to the policy issue (the blue text visible in Figure 1) and the issue-position prompt in Figure 2 but were not asked to provide reasons supporting their attitudes. The full text of both prompts for each of the six issues included in the experiment is given in Appendix A.
Measuring Constraint, Stability, and Polarization
To assess the effects of reason-giving on attitudes, I analyze the correlation between responses on different issue items (constraint), the correlation on the same issue items across survey waves (stability), and the dispersion of responses across respondents on each item (polarization). As declared in the pre-registration plan, I 1) conduct all analyses using survey weights; 2) recode the policy item variables such that higher scores indicate more left-wing or more socially liberal positions; and 3) remove ‘Don’t know’ responses for any of the policy questions.Footnote 10
Constraint
To investigate the effects of reason-giving on ideological constraint, I measure the degree to which correlations between issue stances are higher in the reason-giving treatment group than in the control group. In particular, I calculate the weighted polychoric correlation between each pair of policy items for each group, where, because all policy items are recoded to indicate more left-wing responses, higher correlations indicate a greater degree of ideological consistency across items. The differences in these correlations for each issue-pair (for example,
$\rho _{{\rm{HighTax}},{\rm{MinWage}}}^{D = 1} - \rho _{{\rm{HighTax}},{\rm{MinWage}}}^{D = 0}$
) reflect the extent to which the reason-giving treatment induces more highly correlated attitudes on a given pair of issues relative to the control condition. However, as noted above, the design is well-powered to detect only large treatment effects at the individual issue level, and so for each group, I also calculate the average correlation across the fifteen issue-pairs. The main inferential quantity of interest is therefore the difference in these average correlations between treatment and control groups (that is,
$\overline \rho _{{\rm{Constraint}}}^{D = 1} - \overline \rho _{{\rm{Constraint}}}^{D = 0}$
). When this difference is positive, it suggests that reason-giving respondents report attitudes that are more consistently left- or right-wing across issues compared to control-group respondents.
In addition, the theoretical discussion revealed that we should expect the effects of reason-giving to differ across different issue pairs as reason-giving might affect constraint by making respondents aware of common justifications that apply across related political issues. For instance, common reasons might support a respondent’s views on both the ‘minimum wage’ and ‘zero hours contracts’ issues, but it is less likely that common reasons would apply to the ‘higher rate of tax’ and ‘transgender rights’ issues. Evidence for this mechanism therefore requires categorizing the pairs of issues that plausibly have common substantive underpinnings. Before fielding the experiment, I selected three pairs of issues that I expected to ‘hang together’ in terms of their underlying ideological stance. These pairings were as follows:
-
1. Increase Unemployment Support/Increase Higher Rate of Tax
-
2. Increase Minimum Wage/Restrict Zero Hours Contracts
-
3. Expand Transgender rights/Limit Offensive Speech
These pairings reflect an expectation that attitudes on issues of this sort could be underpinned by common reasons. If the effects of reason-giving run primarily through an increased appreciation of arguments that are common across policies, we should expect effects to be stronger for these selected pairs of policies than for other issue pairs. I pre-registered this expectation and highlighted estimates from these selected issue-pairs in the results below.
Stability
To measure the stability of voters’ attitudes, I calculate weighted polychoric correlations of the six policy items between survey waves for both treatment and control groups. These correlations capture the degree to which respondents’ answers in the first wave of the survey persisted in the second wave of the survey. The differences in the correlations for each issue (for example,
$\rho _{{\rm{HighTax}}}^{D = 1} - \rho _{{\rm{HighTax}}}^{D = 0}$
) therefore reflect the extent to which respondents in the treatment group (
$D = 1$
) have more or less stable attitudes for a given issue than respondents in the control group (
$D = 0$
). As with the constraint measure, the main quantity of interest is the difference in the average (that is, across issue) correlations (
$\overline \rho _{{\rm{Stability}}}^{D = 1} - \overline \rho _{{\rm{Stability}}}^{D = 0}$
) between treatment and control groups.
Polarization
To measure the polarization of issue-based preferences, I calculate the weighted mean absolute error (MAE) of the responses to each policy item in the treatment (for example,
$MAE_{{\rm{HighTax}}}^{D = 1}$
) and control groups (for example,
$MAE_{{\rm{HighTax}}}^{D = 0}$
).Footnote
11
The MAE is the average of the absolute differences between each survey response and the sample mean, meaning that higher values of the MAE indicate that responses to a given policy item are more polarized. As with the other measures, in addition to reporting issue-level treatment effects, the main inferential focus is on the average difference in MAE across issues between treatment and control groups (
${\overline {MAE} ^{D = 1}} - {\overline {MAE} ^{D = 0}}$
). Positive values for this difference indicate that the average polarization of attitudes is higher in the treatment group and negative values indicate higher average polarization in the control group.
The MAE statistic reflects the conceptualization of polarization as the extent of disagreement between voters’ issue positions on a given issue. I focus on this measure, rather than the proportion of voters who adopt ‘extreme’ issue positions because it is possible for a decrease in issue-based disagreement to occur in the absence of voters adopting uniformly more moderate positions. For instance, if reason-giving were to shift a large group of voters with moderate positions a little to the right, and a small group of very right-wing voters a little to the left, the result would be a decrease in polarization (disagreement between the groups would have declined) but an increase in the average extremity of voters’ attitudes (the median voter would have more right-wing attitudes than previously). Accordingly, I focus on measuring the effects of reason-giving on the degree of disagreement among voters on a given issue, rather than the share of voters who adopt extreme issue positions. However, the supplementary analyses in Appendix Section F I, demonstrate that the results are unaffected by using alternative measurement strategies for polarization.
For all quantities of interest, I evaluate sampling uncertainty via a non-parametric bootstrap. I resample 500 times from the original survey data with replacement, blocking on individual respondents, and I construct the quantities above for each iteration. I summarize the results of this procedure using 95 per cent confidence intervals for all quantities.
Results
Constraint
Figure 3 depicts the estimated treatment effects for all fifteen pairwise correlations between the six issues included in the experiment. The left and centre panels of the figure show the effects for the first sample of respondents, with correlations measured in the first and second waves of the survey, respectively. The right-hand panel shows the effects for the second sample of respondents. Points further to the right indicate that reason-giving respondents had attitudes that were more highly correlated on a given pair of issues than control-group respondents. Points further to the left indicate that the control group responses were more highly correlated. Vertical lines represent the average treatment effects across issues for each sample/survey wave.
Effects of Reason-Giving on Constraint.
The figure indicates that reason-giving results in a small average increase in the correlation of attitudes across issues for the first sample of respondents: the average correlation across issues for respondents in the treatment group was 0.086 [0.013, 0.149] points higher than for those in the control group. The average effect of reason-giving is also roughly the same magnitude after the treatment is repeated in the second wave of the survey, where the estimated difference between treatment and control respondents is 0.100 [0.019, 0.175]. However, this effect does not replicate in the second sample of respondents, where the estimated treatment effect is 0.005 [−0.067, 0.090]. Taken together, these results – which average across the effects on different issue pairs – provide only weak support for the idea that reason-giving induces people to provide more ideologically consistent responses.
In addition, the figure also reveals significant heterogeneity in the effects of reason-giving across the issue-pairs included in the experiment. For instance, for the ‘Sample One, Wave One’ results, the estimated treatment effect for the Trans Rights/Zero Hours issue-pair was 0.334 [0.144, 0.534], which implies that treatment-group responses on these issues were marked by substantially higher correlations than control group responses. By contrast, on the Minimum Wage/High Tax issue-pair, the estimated treatment effect was −0.137 [−0.313, 0.032], implying that those proving reasons for their preferences reported attitudes that were somewhat less correlated than those in the control group.
Notably, the positive effects of justification on constraint do not appear to be driven by the pairs of issues which I expected, a priori, to be more responsive to reason-giving. The grey horizontal bars in Figure 3 indicate the issue pairs that were selected as being thematically related in the pre-analysis plan. If the effects of reason-giving run primarily through an increased appreciation of arguments that are common across policies, then we should expect effects to be stronger for policies that are thematically related. However, as the figure reveals, the effects of reason-giving are actually smaller for these issue pairs than the average treatment effect across all issue pairs. For these three issues, the average effect of reason-giving was indistinguishable from zero for the first sample of respondents in both wave one (−0.010 [−0.083, 0.064]) and wave two (0.026 [−0.069, 0.119]), and negative (though insignificant) for the second sample of respondents (−0.132 [−0.278, 0.032]). Somewhat surprisingly, the largest effects of reason-giving appear for issue pairs that include both the first and second dimensions of British politics. For instance, when voters give reasons for their policy views, attitudes on the two social issues (transgender rights and offensive speech) become more correlated with attitudes on a number of economic issues, such as zero-hours contracts, unemployment support, and the minimum wage. Again, however, these patterns do not replicate in the second sample, making it hard to put a lot of weight on these inferences.
Stability
Figure 4 presents the estimated effects of reason-giving on the stability of public attitudes. I again present estimates for each issue included in the experiment, and the main quantity of interest – the average effect of the treatment across all issues – is depicted with vertical lines and confidence bands. As stability is only measurable for the set of respondents who appear in both waves of the survey, I present only one set of estimates for this outcome variable.
Effects of Reason-Giving on Stability.
As with the constraint analysis, despite some heterogeneity at the issue level, the average effect of the reason-giving treatment on the stability of expressed attitudes is close to zero (0.007 [−0.039, 0.049]). For none of the individual issues is the treatment effect significant and positive, and in one case – the Zero Hours Contracts issue – reason-giving appears to decrease attitude stability relative to the control group. This evidence therefore again fails to conform to the prediction of the reasons-as-causes model that reason-giving will lead to greater attitudinal stability. That does not appear to be the case here, as people engaged in reason-giving have attitudes that demonstrate as much temporal variation as those who do not provide reasons for their attitudes.
Polarization
Finally, figure 5 shows the estimated difference in the mean absolute error between treatment-group and control-group respondents on each of the six issues included in the experiment. Again, vertical lines and error bars indicate the average effects across issues, and I present estimates for the different samples of respondents and the different waves of the survey.
Effects of Reason-Giving on Polarization.
By now, the story is familiar: there is a reasonably large amount of treatment heterogeneity across issues but the average effect of the treatment is very close to zero. For example, reason-giving appears to modestly increase attitude polarization on the unemployment support issue but modestly decreases the polarization of attitudes on the appropriate rate of tax for high-income individuals. More importantly, the average effect of reason-giving on attitude polarization is very close to zero. For respondents in the first sample, the average treatment effect is indistinguishable from zero in both wave one (0.000 [−0.029, 0.033]) and wave two (0.016 [−0.024, 0.051]) of the survey. The same is true for the second sample of respondents where the estimated treatment effect is 0.024 [-0.019, 0.073]. Together, these results again fail to support the idea that reason-giving might have systematic effects on political attitudes.
Heterogeneous Treatment Effects
Do these null average effects mask heterogeneity at the respondent level? One might expect, for instance, that the effects of reason-giving would be more pronounced for voters who typically exert little effort thinking about politics (for example, Zaller Reference Zaller1992, 86–88). For such voters, engaging in reason-giving could have strong effects because it is for these voters that greater introspection might most expand the set of considerations brought to mind. By contrast, for voters who typically pay more attention to politics, reason-giving could have less pronounced effects because such voters are likely to already consult a broad variety of considerations when forming their opinions. To test this expectation, figure 6 presents issue-level treatment effects, conditional on respondents’ self-reported level of political attention.Footnote 12
Conditional Issue-Level Treatment Effects by Political Attention.
There is little evidence that the effects of reason-giving vary systematically by political attention. Although for some issues and issue-pairs there are small differences between the treatment effects for high- and low-attention respondents; in general, there is a high degree of correlation across issues and it is not the case that low-attention respondents are systematically more responsive to the treatment than other respondents. In Appendix Figure A16, in analyses that were not pre-registered, I explore whether the average (that is, across issues) effect of the reason-giving treatment on each outcome varies across different groupings of respondents, determined by age, gender, education, political attention, and past vote in the 2016 Brexit referendum and the 2019 general election. I find very little evidence of systematic heterogeneity across these groups. Taken together, these results suggest that the average effects reported above do not mask highly differential responses to the treatment by different groups of respondents.
Threats to Inference
One potential objection is that if the reason-giving treatment did not provoke respondents to think more deeply about their attitudes, then the null effects reported above might be attributable to the experimental design rather than reflecting properties of the attitude formation process. I provide two pieces of evidence that inconsistent with this ‘weak treatment’ interpretation.
First, there is clear evidence that reason-giving respondents spend more time thinking about a given issue before providing their responses than do control-group respondents. Figure A2 in Appendix Section C shows the amount of time in seconds that respondents spent on the introductory screen for each issue, which they viewed before providing their issue preferences. For control group respondents, who only saw a short introduction to the issue, the median time spent contemplating the issue before providing their preferences was 6.0 [5.0, 7.0] seconds. By contrast, reason-giving respondents – who saw the same introduction to the issue as the control group but then also provided justifications – spent 69.0 [67.1, 70.9] seconds contemplating the issue before stating their preferences. That is, the typical treatment-group respondent spent over a minute longer – a ten-fold increase – thinking about the issue at hand before providing their policy preferences than did the typical control-group respondent. Appendix Section C also includes further analyses which demonstrate that these differences in average engagement are not driven by any particular subset of treatment group respondents and that results are not affected by subsetting to units who engaged with the reason-giving treatment at greater length.
Second, the content of the reasons provided by respondents in the treatment group suggests a high degree of engagement with the underlying issues. The median length of responses to the open-ended reason-giving prompt was between fifteen and twenty-two words, depending on the issue, which provides reassuring face validity that respondents were engaging with the reason-giving task. In addition, in Appendix H, I provide evidence that the reasons treatment respondents provided are substantively related to the issues under consideration and that supporters and opponents of different policy positions use predictably different words in justifying their personal stances (Figure A15). This again suggests that people were following the instructions in the prompt and actively considering the reasons that lie behind their political beliefs. Overall, it is therefore unlikely that the reason-giving prompt failed to compel respondents to canvas their minds for salient considerations.
An additional threat to inference is that the reason-giving treatment requires greater cognitive effort than the control condition, which could cause less motivated respondents in the treatment group to refuse to answer some questions or drop out of some survey waves. Given that the treatment requires respondents to engage in an effortful political-reasoning task, it is plausible that the estimated treatment effects could be upwardly biased, as respondents who remain in the treatment-group sample are those for whom we would expect higher levels of constraint and stability and lower levels of polarization. Given the likely direction of the bias, it is all the more striking that the results here suggest such limited effects of reason-giving. In addition, in Appendix Section D, I replicate the main analyses in the paper using inverse-probability-of-attrition weights (IPAWs) to adjust for differential item and unit non-response (Gerber and Green Reference Gerber and Green2012). I show that the substantive findings reported here are not sensitive to the incorporation of such weights. In Appendix Section E, I also demonstrate that the null results are very unlikely to be attributable to ceiling or floor effects.
Readers may also be concerned that the time between the first and survey waves is longer than is typically the case in survey experiments. This could potentially bias the stability analysis towards a null result as the effect of reason-giving has a reasonably long period of time to dissipate. While definitively ruling out this explanation would require replicating the experiment across shorter time horizons, it is worth noting that the design here differs from survey experiments which seek to measure the effect of an information-provision treatment in wave one on attitudinal responses in wave two. Here, the reason-giving treatment is repeated for all treatment-group respondents in the second wave of the survey. This repeated exposure reinforces the treatment strength and makes it more likely for any stability-inducing effects of reason-giving to manifest, despite the somewhat longer time period between survey waves.
Finally, readers might wonder whether reason-giving has effects on other properties of attitudes. One obvious hypothesis is that reason-giving respondents might provide responses that are systematically further to the left or the right on a given issue. In Appendix G, I show that although some differences appear on individual issues, the magnitude of these differences is very small, and the average effect of reason-giving across all issues is indistinguishable from zero.
Conclusion
The core contribution of this paper is to show that reason-giving does not, in isolation, have the salutary effects on political attitudes predicted by the reasons-as-causes model hoped for by proponents of deliberative democracy. Many scholars are optimistic that deliberation can profoundly affect the quality of the attitudes that voters hold, and recent work has explored the potential for ‘reflective, intrapersonal, and private’ thought (Minozzi et al. 2023, 2) to act as a mechanism for delivering the benefits of deliberation. Indeed, for some, ‘internal-reflective’ deliberation ‘might even be a more important part of the process than the dialogic and discursive element’ of deliberation (Goodin and Niemeyer Reference Goodin, Simon and Niemeyer2003, 628). I argued that the reasons-as-causes model helps to clarify how such introspection, as induced by reason-giving, might lead to specific effects on a series of important measures of attitudinal quality. The normative importance of these expectations is clear: if greater cognitive effort could help to save voters from having ‘vague, uninformed, or incoherent’ (Achen and Bartels Reference Achen and Bartels2017, 108) attitudes, then the prospects of strengthened democratic accountability would be consequently enhanced. The null results presented here, however, suggest that whatever weaknesses exist in the political attitudes of the public, inducing voters to devote more cognitive effort to the reasons that underpin their attitudes is insufficient for improving the quality of those attitudes.
The findings here do not, of course, undermine the claim that deliberation, in toto, might have beneficial effects on democratic attitudes. I took seriously calls to investigate ‘important, specifiable, and falsifiable’ claims in deliberative democratic theory (Mutz Reference Mutz2008, 521) by focusing attention on understanding the specific effects of introspective reason-giving, but there are alternative mechanisms by which deliberation might affect attitudes. First, the treatment employed here aimed to solicit ‘internal-reflective’ reasoning, but it might miss potentially important effects stemming from the public exchange of political reasons (for example, Rawls Reference Rawls1997; Mercier and Sperber Reference Mercier and Sperber2018). Second, deliberation might also expose voters to new information about different policy options, and that information might affect expressed preferences. Future work should therefore investigate whether different types of reason-giving and different elements of deliberation have effects on political attitudes, and under which conditions.
These results also have important implications for existing models of attitude formation. In particular, the results contrast with the expectations generated from the model presented by Zaller (Reference Zaller1992). The critical assumption of that model is that, at the point of attitude construction, voters sample considerations over which they then aggregate to form attitudes. The additional cognitive effort induced by reason-giving is therefore expected to affect attitudes by changing the sample of reasons that voters consider. However, voters’ issue attitudes appear to be largely insensitive to the amount of introspective reasoning in which they engage. These null effects therefore cast doubt on the idea that voters construct attitudes via such a cognitively-based, consideration-sampling process and instead are more consistent with the idea that voters’ attitudes primarily form via instinctive, affect-driven reactions (for example, Lodge and Taber Reference Lodge and Taber2013).
It is important to note, however, that Zaller’s consideration-sampling assumption is analytically separable from the assumption that reasons play a causal role in the construction of voters’ attitudes. For instance, it is possible that voters form attitudes by averaging over attitude-relevant considerations but do not sample different considerations in each instance. If this is the case, then, even though reasons are playing a causal role in attitude formation, we would not expect reason-giving to have large effects, as the reason-averaging process would make use of the same considerations at all points in time. Therefore, while the results here contrast with the predictions of the Zallarian consideration-sampling view, they do not necessarily represent strong evidence against reasons-as-causes models as a whole. Nevertheless, given that Zaller’s (Reference Zaller1992) consideration-sampling logic is used as a foundation for many arguments in political behaviour (for example, Bullock and Lenz Reference Bullock and Lenz2019, 327; Freeder, Lenz and Turney Reference Freeder, Lenz and Turney2019, 288; Baccini and Leemann Reference Baccini and Leemann2021, 471), demonstrating that the predictions of that model are not supported by this experiment is an important contribution to the debate over the psychological mechanisms that underpin the expression of political attitudes.
Stability, constraint, and polarization are all important aspects of voter preferences because of the role they play in strengthening democratic accountability and facilitating political agreement (Price and Neijens Reference Price and Neijens1997). However, these properties do not represent all the potentially relevant outcomes which might be affected by reason-giving. An interesting prediction of the reasons-as-rationalizations model is that by engaging in a process of reason-giving, voters will draw to mind considerations that buttress their intuitively formed attitudes, thus increasing the confidence with which they express those attitudes. One important omission here is therefore the absence of data on the strength of voters’ attitudes, and further research might profitably explore whether reason-giving has such effects on opinion strength. Similarly, another interesting avenue would be to explore whether the exchange of reasons between voters of different political opinions might help to decrease hostility across lines of political disagreement.
Finally, my results contrast with a well-established literature in social psychology which finds that asking people to explain the reasons for their attitudes can change the attitudes that they express (for example, Wilson, Kraft and Dunn Reference Wilson, Kraft and Dunn1989; Wilson and Schooler 1991; Wilson et al. Reference Wilson, Lisle, Schooler, Hodges, Klaaren and LaFleur1993; Dijksterhuis Reference Dijksterhuis2004; Simonson Reference Simonson1989; Hsee Reference Hsee1999). In most cases, this research focuses on reason-giving in non-political settings, which provokes the question of whether there is something distinct about the process of reasoning about politics that prevents introspection from having the effects that are apparent elsewhere. One possibility is that people are less informed or knowledgeable about their political attitudes, and so the quality of their introspective reasoning is lower than for affairs with which they are more familiar. Another possibility is that the affective reactions that people experience when thinking about politics are stronger than in other domains, and thus subsequent reasoning is more likely to be biased in the direction of their initial response. Answering these questions is beyond the scope of this paper, but exploring why there are differences in reason-giving effects across settings would be another interesting direction for future research.
Supplementary material
For supplementary material accompanying this paper visit https://doi.org/10.1017/S0007123424000875
Data availability statement
Replication data for this paper can be found at https://doi.org/10.7910/DVN/YG6354
Acknowledgements
With thanks to Lucy Barnes, Peter Dinesen, Timothy Hicks, J. Scott Matthews, and participants in seminars at University College London, Royal Holloway, Durham University, the University of Manchester, and the 2022 annual conference of the Elections, Public Opinion and Parties specialist group.
Financial support
This research was supported by funding from the British Academy (SRG2021-210655) and the Leverhulme Trust (RF-2021-327).
Competing interests
None.
Open access
