Committee decision making is a central feature of many political and economic organisations, including legislative bodies, government agencies, central banks, law courts, and private and public companies. The issues confronted by committees are typically multi-faceted and complex, often involving a variety of conflicts and private interests. Committee members are usually motivated by the desire to advance their own careers and care about being perceived as high-ability decision-makers. Furthermore, they often have different competencies or information, so that it is not unusual to observe situations where some of them abstain when unable to form a firm conviction. 

In response to a trend towards increased procedural transparency in central banking a number of studies have focused on the effects of the transparency of voting procedures on decision-making in committees (e.g. Armstrong et al. 2015, Barwell and Chadha 2014). Indeed, while a number of committees operate under full transparency, secret voting procedures are also common, where only the final voting outcome or the aggregate number of votes for each alternative are observed. Gersbach and Hahn (2008, 2013) and Levy (2007) show that secret voting may reduce distortions arising from signalling competence. Visser and Swank (2007) show that career-concerns create an incentive for committees to conceal internal disagreements and show a united front in public. From an empirical perspective, Hansen et al. (2018), Meade and Stasavage (2008a, 2008b), and Swank et al. (2008) exploit the decision of the Federal Open Market Committee (FOMC) to make the transcripts of its meetings publicly available to show that transparency changed the nature of deliberation in the committee, decreasing significantly dissent among members. In an earlier Vox column, Horvath et al. (2011) discuss under what conditions the publishing of voting records improve the transparency of monetary policy.

There is ample anecdotal evidence showing that committee members and legislators do behave sharply differently under public and secret voting. For example, in November 2013, the Brazilian Congress approved a constitutional amendment that changed the procedure to be employed in cases concerning the expulsion of congressmen from secret to public voting. The shift from secret to public voting was associated with a major change in voting behaviour with the average number of votes in favour of an expulsion almost doubling. In June 2017, the Italian Parliament voted on a proposal to change the electoral law of one of its districts. The vote was supposed to be secret but, due to a technical error, all individual votes were shown on a large electronic display for a few seconds as they were being cast. The mistake was realised by the chairman of the House within six seconds of the beginning of the vote and, after a few more seconds, the vote was suspended. An inspection of the video recording of the session shows that at least 62 members (15% of the total number of votes) switched their votes in a span of less than eight seconds. 

In our recent work (Mattozzi and Nakaguma 2022), we study both theoretically and empirically a committee decision-making problem which combines all elements described above. Specifically, we examine an environment where committee members are biased towards different alternatives, they are heterogeneous in their level of competence, they care about their reputation for competence, and they may vote or abstain. In such a context, we investigate how the choice between secret and public voting affects the voting behaviour of members and the quality of committee decisions. 

A simple framework

Consider a collective decision-making model where a committee must take a decision over a binary agenda by simple majority, and members can vote for either alternative or abstain. The payoff of a committee member depends on three components: (1) a common value i.e. whether the committee takes the ‘correct’ decision; (2) a private value i.e. whether the decision matches the member’s bias; and (3) a career-concern, i.e. the ex-post perceived competence of the member. Competence and bias are private information. Furthermore, under public voting the individual votes of all committee members are observed, while under secret voting only the aggregate number of votes for each alternative are observed.1

In this framework, competent members are informed, they know what the correct decision is, so that transparency creates an incentive for them to vote for it instead of following their private interests. Conversely, incompetent members are uniformed about which alternative is correct, so that transparency simply creates an incentive for them to vote, either for their biases or for the ex-ante more likely alternative, even though they would have otherwise preferred to abstain. In fact, in the absence of career concerns and provided that the common value is sufficiently large, it is optimal for incompetent members to abstain, since by doing so they delegate the decision to the competent members, as in the ‘swing voter’s curse’ of Feddersen and Pesendorfer (1996). In the presence of career-concerns, however, such behaviour affects perceived competence negatively, since abstentions can be interpreted as a sign of incompetence in equilibrium. 

Theoretical results

The model suggests that voting should be public in committees where members are highly influenced by ideological or self-interested motives such as congressional committees. Conversely, voting should be kept secret when the dissent among members due to individual biases is relatively small, as it is perhaps the case of committees of experts and top bureaucrats responsible for technical decisions. In a version of the model where we allow for a behind closed-doors deliberation prior to the voting stage, we show that the level of transparency induces a trade-off between the quality of information aggregation at the deliberation stage and the quality of decisions at the voting stage. 

Given that the choice between secret and public voting is often endogenous to the composition of the committee as well as to the type of decision being taken, it is difficult to isolate the impact of transparency on voting outcomes using non-experimental data. A lab experiment allows us to collect information on individual voting behaviour and compare the quality of the decisions made under public and secret voting while controlling for the level of competence and the biases of committee members. 

Experimental design 

We consider committees of three members with a uniform prior and a symmetric distribution of biases and competence. We amend our basic setup by assuming that the career-concern rewards associated with a correct vote are exogenously given under both public and secret voting. This simplification maintains all basic features of the original model, except that we are now modelling career-concerns in a reduced form fashion. By doing so, we limit attention to testing individual and collective voting behaviour, taking the process of establishment of reputation as given, instead of having human subjects involved in this computation. We implement a two-by-two design where we vary the magnitude of the bias (low or high) and the size of the career-concern rewards associated with a correct vote, which is low under secret voting and high under public voting.2 Our study focuses on a parametrisation such that a change in the level of transparency may lead to a change in equilibrium voting behaviour. Furthermore, since the theoretical model features multiple equilibria with different information aggregation properties, a laboratory experiment can help informing whether subjects coordinate on a particular class of equilibrium. 

Experimental results

Table 1 reports the proportion of correct decisions under each treatment for both the full sample and a subsample considering only the last five rounds (repetitions) of each experimental session. In the last column, we also report the fraction of correct decisions predicted to hold in equilibrium under each treatment according to our model. Consistently with the comparative static predictions of the model, under ‘low bias’ treatments the quality of decisions is larger under 'secret voting’ than 'public voting’, whereas under 'high bias’ treatments the fraction of correct decisions is significantly larger under ‘public voting’ than 'secret voting’.

Table 1

Table 1

Notes: This table reports the proportion of correct decisions observed under each treatment in both the full sample and in a subsample considering only the last five rounds of each session. The last column reports the fraction of correct decisions predicted to hold in equilibrium according to the model.

Figure 1 summarises the individual behaviour of informed and uninformed subjects (i.e. competent and incompetent members in the theoretical framework). Consistently with the predictions of the model, we find that: (i) under ‘high bias’ treatments, the proportion of informed subjects who vote correctly is significantly larger under ‘public voting’ (panel A), while (ii) under ‘low bias’ treatments, the proportion of uninformed subjects who abstain is significantly larger under ‘secret voting’ (panel B). Furthermore, in treatments where there are multiple equilibria, a substantial fraction of subjects coordinates on the efficient equilibrium. 

Figure 1

Figure 1


Our analysis identifies both theoretically and experimentally a novel trade-off: transparency of individual votes attenuates the pre-existing biases of competent members and exacerbates the biases of incompetent members. Furthermore, public voting leads to better decisions when the magnitude of the bias is large, while secret voting performs better otherwise.


Armstrong, A,  F Caselli, J Chadha and W den Haan (2015), “Transparency and the effectiveness of monetary policy”,, 17 March. 

Barwell, R and J Chadha (2014), “Publish or be damned – or why central banks need to say more about the path of their policy rates”,, 31 August.

Feddersen, T and W Pesendorfer (1996), “The Swing Voter’s Curse”, American Economic Review 86(3): 408-24.

Fehrler, S and N Hughes (2018), “How Transparency Kills Information Aggregation: Theory and Experiments”, American Economic Journal: Microeconomics 10(1): 181-209.

Gersbach, H and V Hahn (2008), “Should the Individual Voting Records of Central Bankers be Published?”, Social Choice and Welfare 30: 655-83.

Gersbach, H and V Hahn (2013), “Should ECB minutes be published?”,, 7 October.

Hansen, S, M McMahon and A Prat (2018), “Transparency and Deliberation Within the FOMC: A Computational Linguistics Approach”, Quarterly Journal of Economics 133(2): 801–870.

Horváth, R, J Zapal and K Šmídková (2011), “Central banks’ voting records and future policy”,, 13 November. 

Levy, G (2007), “Decision Making in Committees: Transparency, Reputation and Voting Rules”, American Economic Review 97(1): 150-68.

Mattozzi, A and M Nakaguma (2022), “Public versus Secret Voting in Committees”, accepted for the Journal of the European Economic Association.

Meade, E and D Stasavage (2008a), “The cost of central bank transparency: Closed doors, open minds?”,, 26 June.

Meade, E and D Stasavage (2008b), “The Dangers of Increased Transparency in Monetary Policymaking”, Economic Journal 118(528): 695-717.

Meloso, D, S Nunnari and M Ottaviani (2022), “Looking into Crystal Balls: A Laboratory Experiment on Reputational Cheap Talk”, Mimeo.

Swank, J, H Swank and B Visser (2008), “How Committees of Experts Interact with the Outside World: Some Theory and Evidence from the FOMC”, Journal of the European Economic Association 6(2): 478-86.

Swank, O and B Visser (2007), “On Committees of Experts”, Quarterly Journal of Economics 122(1): 337-72.


1 We extend the model to consider an alternative setting where only the final decision is observed under secret voting. In this case, career-concerns become directly tied to the quality of committee’s decisions, so that incompetent members have more incentive to abstain.

2 Both Fehrler and Hughes (2018) and Meloso et al. (2022) find that experimental subjects have a hard time updating beliefs correctly in the lab.

210 Reads