Scrutinizing Punishment Institutions: A Lab Experiment on Alternative Punishment Regimes

1 Introduction: Institutional Background

This article is an experimental paper in law and economics. It uses incentivized experiments to address a genuine institutional/legal topic: whether (and, if yes, how exactly) alternative procedural regimes entice different levels of costly punishment.

The aim of this paper is threefold. First, the paper provides, in its initial baseline analysis, a very brief contribution to the role of law enforcement by means of “altruistic third party punishment” – namely, by providing an independent observer with the option for costly punishment regarding an underlying “stealing scenario”. In its main second part, the paper contributes to comparative institutionalism by comparing the incidence of this altruistic, third party punishment in the aforementioned “institution-free” baseline punishment-scenario with different institutional (“procedural”) environments. We begin this comparative analysis, thereby building on our previous work, by studying the punishment effects that stem from (i) the introduction of a second (vertical) instance that has the last word regarding punishment, possibly overriding the first instance decision. We then turn to an alternative institutional setting – namely, (ii) the introduction of an obligation for the (potential) punisher to provide justification for her/his decision. The pertinent research question is whether such an obligation to reflect on one’s punishment decision and provide written reasons increases or decreases costly, third party (altruistic) punishment. We proceed by (iii) merging the two aforementioned variants, such that the “first instance” punisher must justify her/his decision, which is subsequently subject to a (possibly overriding) reassessment by the second instance. Third, the paper addresses the influence of the aforementioned various enforcement institutions on the underlying “stealing behaviour” – namely, on their effects to reduce the “incidence of crime”. Both the experimental analysis of a giving-reasons-requirement and its combination with an “appeals-setting” on enforcement levels have not been investigated in the previous literature.

On a more technical level, the paper is “procedural”. There is (and has been) wide consensus that ideal procedures shall be expedient, inexpensive, and just (=fair^[1]). A simultaneous accomplishment of these goals, however, is a difficult task, because the aforementioned goals are partially conflicting.^[2] Because of this trade-off, there is no such thing as the single best procedure. Procedural rules are always a compromise seeking to balance the accomplishment of these different goals in a socially acceptable manner,^[3] whereby some goals may be prioritized over others depending on the situation at hand. In that sense, there are many different (procedural) paths to Rome. In terms of law enforcement, concepts of fairness appear to require that a committed wrong shall not go unpunished. The institutional question, however, is whether this enforcement-task can be best accomplished in an institution-free world by introducing a second vertical instance, a giving-reason-requirement, or a combination of the two. Although these alternative settings obviously differ in terms of cost (with the usage of a second instance naturally increasing costs in relation to a simple giving-reasons-requirement) and the time necessary to reach a final decision, the question is whether they accomplish similar or different enforcement results (in terms of the incidence and amount of punishment).

Historically, the picture is complex. In ancient times, broadly conceived dispute resolution (including “criminal enforcement”) was primarily done within the community and often even in front of the entire community, by specific community members (often by wise elderly men). In such a world, there is no need for a second instance because, on the one hand, all conceivable evidence is already being collected by the decision makers in the course of their basic decision and, on the other hand, there are no other, potentially better decision makers available who could constitute a second instance in a meaningful sense. Whereas, in such a world, the need for reasoned decisions is reduced, early tribal forms of restorative justice did encompass a pertinent communication element in the ultimate “settlement decision”. In turn, giving reasons for a verdict was obsolete to the extent that the procedural system relied on formalistic types of proof (say, in “ordeals”, where the result of the ordeal already decides the trial).^[4]

Still, in more modern times, two fundamental institutional instruments have emerged with the specific potential to improve judicial decision making – namely, the provision of a second instance (“vertical control”) and a requirement to motivate one’s decision (“giving reasons”).^[5] The paper analyzes the effects of these instruments both separately and jointly (i.e. in a procedural world that combines a giving reasons requirement for the first instance with a second instance) in their potential to increase the incidence of costly punishment.

Vertical control is essentially provided by allowing for an appeal against the first-instance-decision.^[6] The introduction of any such remedy serves a twofold goal. First, appeals represent an instrument designed to check and possibly improve the decision by the first instance by correcting conceivable errors.^[7] Second, the institution of appellate review aims as such to improve the quality of first-instance judicial decision making: the mere possibility of such an appeal casts a shadow on the first instance and provides a powerful incentive for the decision maker (i.e. the trial judge) to render a decision such that it will not be overturned. In substance, this means that the first instance court will orient itself towards the presumed position by the higher court on this question – which is generally a good thing.^{[8] , [9]}

A second basic procedural feature, a “giving reasons” requirement for judicial decisions, is considered a fundamental aspect of procedural fairness and decision making under the rule of law.^[10] Requiring the decision maker to motivate her/his decision is a tool to rationalize the decision and improve its quality.^[11]

In this paper, we assess the potential of the two aforementioned “institutional levers” to affect legal decision making (= punishment decisions) from an empirical, namely experimental, viewpoint. To the best of our knowledge, a comparable experimental study has not yet been undertaken. There is, thus far, no systematic study examining the impact of “giving reasons” on legal decision making as such, neither regarding the impact on the decision maker her/himself (compared with no reason being required), nor on the way in which a “second instance” views a reasoned versus non-reasoned decision. Our experiments aim to provide a first view on this set of issues through the lens of costly punishment.^[12]

Whereas the paper analyses – with the focus on “giving reasons” and vertical control – the effects of various procedural settings on punishment and law enforcement (and on the incidence of the underlying “unwanted” behaviour), it does not investigate other relevant procedural issues. In our experiments, there is no ambiguity regarding the underlying deed; the relevant facts (stealing vs non-stealing) are evident.^[13] As we leave out questions of fact-finding, we also do not study judicial effort in terms of concerns for accuracy.^[14] What we are interested in, however, is the punishment-decision of the “judge(s)” in the light of the different institutional environments and the therein embedded incentives. This paper is also not a study on the incentives for bringing an appeal.^[15] In fact, the appeal is “automatic”. In the vertical treatment, each first instance punishment decision is subject to review by the second instance decision maker without a pertinent initiative or action by a player being necessary or even possible to trigger this “vertical control”.

Our aim is to understand whether one of these two mechanisms works better than the other for “enforcement purposes” (in terms of incidence of punishment) as well as whether a combination of the two is preferable. We believe these are relevant questions for at least two reasons.

First, the implementation of these mechanisms entails some costs, both monetary and in terms of time (i.e. both mechanisms increase the monetary cost of a trial and the time required to resolve a dispute). Attempting to understand whether one of these two mechanisms (or a combination of the two) entice higher levels of punishment may be a starting point for a more sophisticated cost/benefit analysis. Second, the experimental analysis allows for the identification of possible shortfalls in enforcement that may result from undesired counter-effects in a “combined scenario”.

Moreover, the use of experiments allows us to perfectly disentangle the effects of the two mechanism and, at the same time, analyse them jointly. The experimental methodology appears to be the most suitable tool to answer our research questions, since it allows us to control the institutional environment – something that is nearly impossible in a real-life context.

Experiments have been widely used to highlight different scenarios and aspects of (third-party) punishment. Starting with the pioneering works by Fehr and Gäechter (2000, 2002) on public good games, the third-party punishment game developed by Fehr and Fischbacher (2004), allowing a third party to sanction a dictator if s/he considers the allocation of the pot unfair, was another milestone of the literature. A refinement of this experiment is the “taking game” (or “gangster game”), initially developed in 1998 by (see Flage 2024 for a meta- analysis), in which the dictator is not the owner of the endowment but may take the money from the other player. Lewisch et al. (2015) combined the taking game with the third-party punishment game. This is the basis of our experiment.

Section 2 describes the experimental design and procedure. Section 3 reports the pertinent payoff-structure. Section 4 defines the expected results, whereas Section 5 focuses on data analysis. Section 6 summarises our findings and provides a conclusion.

2 Experimental Design and Procedure

In our experiment, subjects participate in a game designed to represent a simplified version of a specific, real-life situation, namely costly punishment under different institutional environments. Beginning with a benchmark scenario (which we will call “Baseline”), we manipulate our interest variables (“top-down control” and “giving reasons”) – first, one at a time and, eventually, in a combined format. Thus, the resulting design consists of four treatments: baseline (= ordinary third-party punishment), vertical control, giving reasons, and vertical control with giving reasons. Each of these treatments is implemented with different groups of subjects.^[16]

This is a laboratory experiment in which decisions are recorded by the computer and anonymity is guaranteed even when the experimental session ends.^[17] Participants enter the laboratory and take a seat in front of a computer. They are immediately asked to switch off their mobile phones and refrain from talking to other participants. Participants then read the experiment instructions on their computer screen, while an experimenter simultaneously reads them out loud. A sheet of paper with all information concerning the game is also handed out.

In the B aseline T reatment (BT), a basic, 3-person punishment scenario is implemented. Each session involves n A-players, n B-players and n*3 C-players.^[18] Each player is assigned just one role for the whole session. In each session, players participate in 3 rounds. At the beginning of each round, all players receive their initial endowment (200 tokens each). The first round consists of two stages. In the first stage, A-player and B-player are paired and participate in a Theft Game, in which A-player is offered the option to take away (“steal”) 50 tokens of B-player’s initial endowment. In the second stage, 1/3 of C-players are randomly selected and assigned to a team who played the theft game in the first stage. Each selected C-player must decide whether she/he wants to spend 10 tokens of the endowment to punish A-player^[19] in the case of stealing (i.e. ex ante).^[20] Punishment may be lenient (A-player’s endowment is reduced by 100 tokens) or harsh (A-player’s endowment is reduced by 150 tokens). We introduce these two levels of punishment since they represent two ways of sanctioning the theft. In the lenient case (100 tokens) A-player’s payoff when s/he steals and the payoff of B are equal. In this case, C-player restores the initial situation of equality between A-player and B-player. In the harsh case (150 tokens), A-player becomes poorer than B-player. Thus, C-player presumably believes that A-player deserves to be in lower position than B-player.

C-players who are not selected to participate in the second stage are asked to wait. The second round is exactly the same as the first one, except that a different 1/3 of C-players are selected to make their choice in the second stage. In the third round, A-players and B-players play again, while the remaining 1/3 of C-players participate in the second stage. This implies that A-players and B-players play the game 3 times, whereas each C-player makes her/his choice only once. Even if A-players and B-players play for all the rounds, no feedback is provided until the end of the experimental session. This holds for all treatments.

The BT represents our benchmark scenario. In fact, it represents a standard setting (of so-called “third party altruistic punishment”) that has been frequently examined in experimental studies. Obviously, a “social conflict” emerges if A-player takes (“steals”) money from B-player. Whereas B-player is unable to react, the neutral third party, the observer C-player, has the opportunity to intervene and punish the A-player. This punishment option is costly for the punisher in that, following a standard format, the C-player may “purchase”, by investing some of her/his own endowment, monetary deductions from A-player’s endowment. The punishment deduction from the A-player is a genuine punishment, i.e. it goes to the experimenter and not the punishing third party. The punishment/non-punishment decision by the C-player is the end-step of this social conflict. In some meaningful sense, it terminates the conflict that was originally sparked by A-player’s “theft”. Since the parties are aware of C-player’s punishment option, decisions by the A-player to take money from the B-player are always done in light of C-player’s punishment options and the respective expectations.^[21]

In the other three treatments, we introduce “second-instance control” and “giving reasons”, both separately and jointly, in order to study the influence of these institutional tools on conflict resolution and conflict avoidance.

The V ertical C ontrol T reatment (VCT) is an extension of the baseline model with an “automatic” second-instance control competent of superseding the first-instance punishment decision. The D-player is informed about the underlying decision by C-player.^[22] Technically, s/he has to confirm or not confirm C-player’s choice. In the case of no confirmation, s/he can correct C-player’s choice by selecting one of the other two options available. For example, if C-player decides not to punish the A-player in the case of theft and the D-player does not confirm this choice, s/he must impose a punishment decision and provide a final sentence by choosing between the lenient and harsher sanction. If, say, the C-player punishes but the D-player disagrees as to the harshness of the penalty, s/he may only alter the penalty. Each session involves n A-players, n B-players, n*3 C-players, and n D-players. Again, in each session players participate in 3 rounds. At the beginning of each round, all players receive their initial endowment (200 tokens each). Each round consists of three stages. The first two stages are the same as in the BT. However, in the third stage, the “second instance” comes into play as a new D-player. Each D-player is assigned to a team (A-player, B-player and C-player) and takes a fresh decision on whether or not – and, if yes, to what degree (lenient/harsh) – to punish the A-player. D-player’s punishment decision amounts to a confirmation or a correction of the C-player’s decision (to punish or not to punish). Mirroring the real-life situation that “confirmation” of a decision is easier than “overturning” it, only the decision to supersede the C-player’s decision is costly for the D-player. The decision to overrule the C-player’s punishment decision comes at a cost of 10 tokens for the D-player. Since the decision by the D-player marks the final step in the entire interaction, it ultimately determines the outcomes for both the players and the first instance punisher in the underlying scenario. D-player’s decision^[23] directly determines A-player’s ultimate payoff (meaning that the punishment decision of the first instance is set aside by the appeals decision). Even if the C-player had not punished the A-player, a possible punishment decision by the D-player will lead to A-player’s punishment.^[24] Likewise, if the A-player was punished by the C-player, but D-player does not punish the A-player, the A-player receives her/his initial payoff (without the punishment deduction). The appellate decision (realistically) also affects the “first-instance-punisher”, the C-player, in the sense that modifying decisions by the instance (D-player) lead to a reduction of 20 tokens in C-player’s payoff, whereas confirming decisions leave the C-player’s payoff unchanged.^[25] The second round is exactly the same as the first one, with the exception that another 1/3 of C-players are selected to make their choices in the second stage. In the third round, the remaining 1/3 of C-players participate in the second stage. In the VCT, A-players, B-players, and D-players play the game 3 times, whereas each C-player makes her/his choices only once.

The G iving R easons T reatment (GRT) follows almost exactly the set-up of the BT, with the – however crucial – difference that now, C-players must write down the reason for her/his choice.^[26] This reason is not communicated to players A and B in the group to avoid “words-against-fists-substitution” – i.e. C-players punish by (costless) harsh words, while avoiding costly monetary punishment.^[27] However, A-player and B-player are informed that they are in a scenario in which the potential punisher C-player must motivate her/his punishment decision. The C-player, of course, also knows before making her/his choice that she/he will have to justify her/his decision.

In the “giving reasons” scenario, the fulfillment of this mandatory requirement (which applies to both punishment and non-punishment decisions) does not involve additional pecuniary costs.

The V ertical Control G iving R easons T reatment (VGRT) merges the two aforementioned procedural worlds, namely the giving-reason requirement with the introduction of a “second instance”. This scenario, therefore, matches the VCT, with the notable difference that now each C-player (i.e. the “first instance”) must write down the reasons of her/his choice.^[28] This reasoned decision is communicated to the assigned D-player before she/he makes a choice. As in the GR, the C-player is aware of the requirement to give reasons prior to making her/his choice and also that the reasons will be communicated to the D-player. A-players and B-players are again not informed about the reasons given by the C-player; however, they know that they are in a scenario in which the C-player must motivate her/his decision and that the D-player will be informed about these reasons before making her/his own (final) decision.

At the end of the each experimental session, after filling out a socio-demographic questionnaire, subjects receive their payments in Euro.^[29] Participants are paid according to their role and decisions. The final payment is the sum of the payments received in each round. This point is clear from the beginning of the experimental session. The average duration for each session is 45 min. The average payment is 14 €.

Figure 1 and Table 1 summarize the experimental design. Figures 2 and 3 report the trees of the games.

3 Payoff

First, regarding the BT and the GRT: if A-player does not take money from B-player, the final payoff is the same for each player and is equal to the initial endowment of 200 tokens each.

If A-player takes (“steals”) the 50 tokens from B-player, A-player can obviously enrich herself/himself by the amount taken from B-player. Hence, the payoffs are as follows:

Where:

E = 200 tokens is the endowment each player is given at the beginning of the session.

T _B  = 50 are the tokens that the A-player can steal from the B-player in the first stage.

C = 10 is the cost that the C-player bears if he/she sanctions the A-player. The cost for punishment is a flat fee; therefore, it does not differ according to the severity of the punishment.

S_C is the level of sanction that the C-player imposes on A-player. It may be equal to 0 (if the C-player decides not to punish the A-player for her/his theft), 100, or 150 tokens.

Second, regarding the VCT and VGRT, if the A-player does not take money from the B-player, the final payoff is the same for each player and is equal to the initial endowment of 200 tokens each.

If the A-player steals 50 tokens from the B-player, the payoffs are as follows:

Where:

S_F is again the amount of punishment imposed on the A-player, be it by the C-player or (eventually) the D-player. If the D-player decides to punish the A-player (and, thus, confirms the decision taken by C-player), it corresponds to the level of sanction chosen by the C-player. If the D-player’s punishment deviates from that by the C-player, the D-player’s decision is relevant, superseding (in all directions) C-player’s choice. Hence, the ultimate punishment for the C-player corresponds to the level of sanction decided by the D-player. It may be equal to 0 (if the final decision is not to punish the A-player for the theft), 100, or 150 tokens.

SD = 20 is the “sanction” the C-player receives (the disutility that the C-player suffers) if the D-player overturns her/his decision.

The reason to choose this pay-off structure, where we consider the costs of punishment without the “potential” benefits, is twofold. First, we do not want to present an overly complicated design. Secondly, in Italy, where we run our experiment, additional monetary benefits with respect to the initial endowment for third parties may sound strange for the experiment participants. In real life, judges receive a wage.

4 Experimental Hypotheses

In this part, we present our hypotheses regarding the final level of punishment in the different treatments. Hypotheses 1 and 2 consider both the homo economicus and behavioral perspectives. The third hypothesis is based on two different ways of reasoning. In fact, we divide the hypothesis in two possible results: the first is the opposite with respect to the second.

Hypothesis 1: The level of punishment is higher in the VCT than in the BT. ^[30]

Both the baseline treatment and the second-instance treatment involve cases of possible altruistic third-party punishment (be it, in the BT, that the C-player has the punishment option or, as in the VCT, that the D-player has the last word on punishment). In a narrowly conceived homo-economicus world, costly punishment will not be meted out, because it implies cost without (instrumental) benefit. Anticipating thus zero costly punishment by the C-player, the rational choice for the A-player is to take the B-player’s entire endowment. In the VCT, nothing would change because, in backward induction, there are again no instrumental grounds for the D-player to reverse, in a manner costly for herself/himself, the C-player’s punishment decision. Anticipating the lack of incentives for reversal, the C-players may disregard the presence of a second instance and simply pursue their rational choices as before, i.e. to avoid, on their part, costly punishment. The A-player would again “steal with impunity”.

Only if we introduce, in line with previous experimental results, a behavioral perspective, may we expect a certain degree of punishment in the baseline treatment. Third-party punishment is considered to be a social order enforcement device. A sort of golden keystone of social stability (see Chaudhuri 2011; Gachter and Herrmann 2009; Lewisch 2020). Studies (e.g., Fehr & Gächter 2002) show that people are willing to sacrifice resources to punish norm violations, even when they gain nothing. Even when not personally involved, people care about how others treat each other. Experimental evidence confirms this point (see Fehr and Fischbacher 2004, Ottone et al. 2015, among others).

What changes may we expect in a VCT-world? Given that reversals are costly and confirmations are not, there is, first of all, a certain status quo bias (Kahneman et al. 1991) in favor of any decision taken by the C-player. One may, moreover, further distinguish between punishment and non-punishment decisions by the C-player. In principle, punishment by the C-player can be construed as an altruistic (“social”) act, namely as a positive interference with (or as a “correction” of) a seemingly unjust act committed by A-player (therefore, as a type of unilateral – costly – dispute resolution). Will we expect the D-players to correct a punishment decision by C-player? In light of the foregoing, it seems rather unlikely that the D-players would sacrifice their money to reverse what could be framed prima facie as the “social act” of costly punishment. Would the D-players, in turn, reverse a non-punishment decision by the C-player? This is a less straightforward question. Whereas any reversal is costly, the D-players may wish to punish the A-players (in order to answer their “theft” vis-à-vis the B-player), but they may also be driven by the wish to “correct” (and insofar to “punish”) the C-players, who selfishly did not carry out altruistic punishment to protect their own resources. In such a situation, even self-interested C-players may carry out punishment to avoid the potential reversal of their sentence by the D-players. Altogether, these factors are likely to contribute, in a behavioral perspective, to a slight overall trend towards an increase in punishment (i.e. more punishment to be expected in the VCT than in the baseline treatment). These factors lead us to predict an increase in the level of punishment in the VCT with respect to the BT. This would confirm the result found in Lewisch et al. (2015).

Hypothesis 2: The level of punishment in the GRT is higher than or equal to the level of punishment in the BT.

Would requiring participants to provide reasons for their punishment decisions increase the incidence of punishment? Obviously, this requirement would be without any effect for the hard-nosed homo economicus, because she/he would not carry out costly punishment in any situation. In a more behaviourally oriented world, the obligation to motivate one’s decision is likely to have an effect. On the one hand, “giving reasons” forces the potential punisher to “think twice” about his decision and reflect on both the underlying deed and her/his reasons to answer or not answer this unsocial act, with a slight tendency to not leave such an act unchallenged. On the other hand, this requirement may also lead to reflexions by the potential punisher regarding the question whether her/his spontaneous impulse for punishment is worth the cost associated therewith.^[31]

In the realm of this hypothesis, we assume that a (costless) no-punishment decision is quick and easy, whereas the requirement to motivate one’s-decision triggers reflections that instead increase the likelihood of punishment. Whereas the required reasons could, in cases of non-punishment decisions, also evoke the absence of meaningful deterrence in one-shot interactions, they would often also implicate a confrontation with shame/guilt when putting on paper a selfish reason for not punishing. Depending on the composition of the population on the experiment, we would cautiously predict an ultimate punishment level that is equal to or higher than the level of punishment in the BT.

In the GRT treatment, the reasons by the C-player are not communicated to other players. Consequently, we can rule out the possibility that a higher level of punishment may be due to the fact that a non-punishment choice is difficult to justify to the B-player (the victims).

Hypothesis 3A: The level of punishment in the VGRT is higher than the level of punishment in the VCT and GRT.

This hypothesis is based on the assumption that both mechanisms, viewed separately, are likely to generate, in a behavioral perspective, an increased incidence of punishment, so that this effect shall only be reinforced when both institutional changes apply jointly. Or put differently, if setting GRT as well as VCT is likely to foster punishment, we should expect an even stronger positive effect on punishment if we bring the two settings together.

Hypothesis 3B: The level of punishment in the VGRT is lower than the level of punishment in the VCT. The effect with respect to the BT and the MT is uncertain.

This hypothesis is based on the assumption that, whereas both mechanisms have a positive effect on punishment, the combination of vertical control and “giving reasons” need not necessarily generate super-positive results. Provided for communication between the punishers, opportunistic C-players could attempt to abuse the opportunity to motivate their decision for purposes of influencing the D-players and not carry out costly punishment against them. The underlying intuition would be the following: In a scenario with vertical control, where choices cannot be justified, C-player’s decision to not punish may be risky in terms of triggering punishment by the D-players. In turn, in a motivated decision, the C-players are able to explain their motives for non-punishment (be it the absence of deterrence purposes, the lack of an explicit prohibition of stealing in the formal instructions, or a homo economicus reasoning) and also attempt to convince the D-players that it would be in their best interest to leave the punishment decision by the C-players unaltered (in which case, both the C-player and D-player could save their resources). To be sure, any such attempt to influence the D-players would also be possible if the C-player actually punishes the A-player (explaining to the D-players that it would be unwise to alter such a “social” decision in light of the costs associated therewith). The “communication tool”, however, seems more attractive for the non-punishment decision, where the C-players can also explain their “good” reasons for the seemingly unsocial act of non-punishment and the advantage for the D-players to leave such a decision unaltered.

5 Results – Data Analysis

We ran three sessions for each treatment. Overall, 319 students at a University in northern Italy^[32] participated in the experiment.^[33] With the final questionnaire, we collected data concerning standard demographic characteristics (gender, age, occupation, religious beliefs^[34]), as well as information revealing participants’ attitudes towards intervening to correct unfairness and solve real or potential conflicts (volunteering activities, chosen option in the classic trolley game^[35]). Finally, we checked whether participants were naive experimental subjects by asking them whether it was the first time they participated in an experiment.

About half of participants (53.3 %) are male. Less than 24 % are currently employed. More than half of participants are religious (61.8 %). 21.3 % are volunteers. 19 % were participating in an experiment for the first time. Table 2 reports how these variables are distributed across the treatments. Statistical analyses show some differences, especially between MT and VCMT. In any case, the econometric analysis allows us to examine whether potential differences between the treatments are affected by these characteristics.

This means that, during the experimental campaign, we collected observations from 173 cases. Due to the composition of the groups and the fact that each C-player participates in a single game, whereas each D-player participates in three games, we collected 173 independent observations from 173 C-players (42 in BT, 45 in VCT, 42 in MT, 44 in VCMT) as well as 90 choices from 30 D-players (15 in the VCT and 15 in VCMT).

We conducted our analysis in several steps, focusing on the relative changes in terms of punishment (5.1.) and of the underlying “criminal” behavior (5.2.).

6 Discussion and Conclusion

The aim of this paper is to shed some light on the basic question of whether an implementation of institutional variants like “vertical control” and “giving reasons” influences the carrying out of costly punishment by potential punishers. We focus on these specific institutional instruments since both “vertical control” and “giving reasons” are generally presumed to be devices that improve the attractiveness of a procedural system. Moreover, both instruments aim to make judges’ decisions more deliberative and less intuitive.^[45] The former because it drives first-instance judges to think more carefully about what is a correct and fair sentence in order to not be reversed by second-instance judges. The latter because it makes judges focus more on the rational and normative aspects of a conflict rather than on instinctive and emotional factors.^[46]

The paper analyzes four different experimental settings, each of which stands for a separate institutional world (“naked” third-party punishment, vertical control, giving reasons, motivated punishment under vertical control). We believe that the paper is novel, namely that the impact of neither an (overriding) second instance,^[47] a giving-reason requirement, nor the combination of the two have been studied previously.

Our main result is that both instruments (“vertical control” and “motivated punishment”) increase the level of costly punishment by non-involved observers (= potential punishers) in relation to the institution-less baseline scenario. Whereas the two instruments generate similar results, the changes are even more pronounced in a “giving reasons scenario” (which amounts to a remarkable 10%-points increase in the incidence of punishment vis-à-vis the baseline scenario). In contrast, a combination of the two (in themselves powerful) institutional devices, in a somewhat counter-intuitive manner, does not additionally increase punishment rates. This last-mentioned finding highlights a general point in procedural rule making, namely the possibility of countervailing effects in the overall procedural system. That is, procedural rules (aimed as such at a well-defined goal) allow the players in the system, even only as a side effect, to pursue their self-interest in an opportunistic manner. In addition, there is always the danger that the accomplishment of overall policy goals is weakened.^[48]

Overall, our results show that the institutional framework of the respective setting has a considerable impact on the level of third-party punishment. Both the introduction of a second instance and the requirement of motivated decisions trigger a substantial effect on the incidence of punishment. A comparison between these two institutional tools, however, shows that these similar results come at a different cost, given that a second instance requires both more resources and more time than a mere giving-reasons-requirement. Whereas in our experiments the combination of the two tools generated the somewhat surprising “negative result” of a decrease – and not a further increase – in the incidence of punishment, this is not to suggest that such a combination is generally unadvisable on policy grounds. Our findings just alert us to the fact that players in procedural systems behave opportunistically and that they may misuse certain procedural instruments if they have the option to do so. Procedural rules shall, therefore, be designed such that they minimize the potential for such opportunistic behaviour. Under the appropriate incentives, a combination of a second instance and a requirement to provide reasoned decisions may work very well indeed.