Irenaeus Wolff

2020 (with Sebastian Fehrler and Baiba Renerte), download

• We use a standard ingredient of many models and experiments to obtain a setting in which both social projection and information neglect could happen: two urns (states of the world), a light one (2 light balls, 1 dark ball) and a dark one (2 dark balls, 1 light ball). Participants get a signal (a single draw) from the actually-drawn urn and have to guess on the colour of the next-to-be-drawn ball + to indicate the probabilities of the urn being light or dark and of the next ball being light or dark. Up to 70% report a 50-50 belief on the urn question, indicating that they cannot learn anything from the signal. We go to great lengths to induce learning in three consecutive experiments, but to little avail.

2018, download 

Current title: Predicting Voluntary Contributions by 'Revealed-Preference Nash Equilibrium'

• we don't have good predictions for one-shot public-good situations (few models are calibrated to even allow for such predictions). This paper looks at whether Nash-equilibrium predicts aggregate behaviour out-of-sample when based on empirical conditional-contributions measurements (prediction from Wolff, 2017). Answer: yes, pretty well!
• even better when we provide conditions close to the mutual-knowledge-of-preferences assumption (by disclosing interaction partners' conditional-contribution vectors [in a deception-free way]). It even predicts behaviour of those who don't believe in the displayed preferences, and of those who show stochastic preferences.
• in case of multiple equilibria, Fehr & Schmidt's (1999) invoking of Pareto-dominance gets empirical support, even though about 1/5 of the population (each) also select the average- and the lowest-contribution equilibria.

with D. Bauer, 2018, download

Current title: Why is Belief-Action Consistency so Low? The Role of Belief Uncertainty

• We experimentally induce different levels of belief uncertainty and show that more belief uncertainty leads to more inconsistent behaviour.
• We point out that more information may lead to more uncertainty – if the information goes against one‘s prior
• We provide an individual-level experimental test of the `belief-sampling‘ model that has been shown to predict behaviour in multi-armed bandit settings (→ OR literature) as well as aggregate behaviour in games, markets, and surveys (→ Felix Mauersberger).

with Urs Fischbacher and Moath Hussien

• We look at when it is ok or even expected to punish, analysing data from several countries

with Regina Anselm, Fabian Dvorak, and Urs Fischbacher

• Two projects in which we do what the title suggests. :-)

with Adrian Chadi, Urs Fischbacher, Moath Hussien, and Katrin Schmelz

• Strategic behaviour of candidates may lead to certain types of candidates being selected/promoted
• This may yield unequal outcomes when focusing on certain (job-unrelated!) characteristics of the (non-)selected candidates (e.g., gender). How are such inequalities evaluated?

• Can we trigger (or block?) reciprocity norms by inducing sharing (for `survival’)? How does the environment relate to the evolving norms?

with Malte Baader and Urs Fischbacher

• We study the consequences of patent protection depending on whether reverse-engineering is easy or difficult in the controlled world of a lab experiment

with Urs Fischbacher, Moath Hussien, and Baiba Renerte

• A huge new research area that we are only starting to discover

with Dominik Folli, Journal of Economic Psychology, 88: 102458, 2022, download

• Does it matter whether you ask: „what do you think did your opponent do?“ or „what do you think did everybody else do?“ We show it does.
• In general, people show a clear consensus effect. But when it‘s a bad thing to do what my opponent does and you ask me for my guess about the opponent‘s action, I‘ll re-adjust and ex-post rationalise my action. Therefore: don‘t ask me about my opponent‘s action in such cases (my reports are going to be distorted twice rather than only once)!
• We find no evidence of a hindsight bias or wishful thinking in any of the treatments.

Journal of Economic Behavior & Organization, 184: 16-29, 2021, download

• Why do we see a bunching of choices in a pure discoordination game played on the options "A", "B", "A", and "A"? And why do we see the same bunching in other games with the same unique mixed-strategy equilibrium prescribing uniform mixing (as I show)? And – why does the bunching correspond to the pattern we observe when people simply have to bet on one of the options (as I show)?
• My answer: many people reason strategically in one way or the other, but at some point they realise that this doesn‘t help in the situation at hand (it doesn‘t eliminate even a single option). At this point, they "switch off" their "strategic-thinking engine" and simply bet on any of the options (which ones is predicted by findings from psychology). This behaviour could be exploited by others who best-respond to it. But it isn‘t. Models that include layers of best-responding to betting-like behaviour predict clearly worse out-of-sample than a simple mix of betting-like behaviour and uniform randomisation. This answer accounts for two well-established findings, one from the literature and the other one from the real world: the seeker advantage in hide-and-seek games, and the bunching in national-lottery bets.

Journal of Economic Psychology, 74: 102197, 2019, download

• Paying participants as soon as they finish the questionnaire ruins the data quality.
• If data quality is key, wait until everybody has finished before starting to pay. This often is costly, of course. If data quality is not absolutely key, consider paying after most participants have finished the questionnaire.
• Paying too little overall is bad too. Framing the show-up fee as being „for the completion of the questionnaire“ doesn‘t help, though.
• Neither providing progress feedback nor having people put in their names to prepare the receipts seems to affect the data quality. In our case, asking people to enter their names does not lead to more socially-desirable answers even on the very next screen, while providing progress feedback after two thirds of the pages at least improves the motivation to fill in open-answer questions.

with Bettina Rockenbach, Review of Behavioral Economics, 6: 19–37, 2019, download

• Studies a setting in which interactions work as in the usual VCM experiment, but what people have at the end of a round is what they have at the beginning of the next one (think of a bank account that is endowed with a number of tokens only once, at the beginning). This also means that punishment destroys future contribution capabilities.
• In a treatment without peer-punishment, initial contribution levels predict final wealth levels. In a treatment with punishment, this is not the case. Here, what is important is that cooperators do not start punishing non-cooperators straight away, as doing so triggers retaliation cycles. In contrast, `patient‘ administration of punishment fosters wealth.

with Joachim Plesch, games, 9(2), 24, 2018, download

• People don‘t seem to give away their data just because the true size of loyalty-programme benefits often remains hidden. Instead, it seems like the programmes successfully hide that customers are paying with their data – or that the ambiguity of which data is eventually used helps the companies. Finally, we find that political support for the `liberal democrats‘ (the most business-friendly party) is correlated with weaker privacy concerns.

Economics Letters, 150: 83–85, 2017, download

• Shows that empirically (judging by people‘s revealed preferences), there are positive-contributions equilibria far more often than we would expect, e.g., by the parameterised model of Fehr & Schmidt (1999). That re-opens the question of why contribution levels don‘t converge to these more socially-beneficial equilibria in repeated settings.

with Bettina Rockenbach, German Economic Review, 17(3): 316–336, 2016, download

• We let participants come up with their own rules for a standard public-good setting on four consecutive occasions. Many start out introducing (centralised!) punishment but abandon it later on. Also, many use openly-fictional framing, and surprisingly successfully so. A number of rule designers choose to conceil individual contributions, and no designer introduced a leader or ostracism. Finally, rule designers typically combined several rule components.

Economics Letters, 141: 134–137, 2016, download

• Can we explain behaviour in a Hide&Seek game on „A“, „B“, „A“, and „A“ using a level-k model that uses empirical measures of salience as level-0? Short answer from 9 different measurements of salience: No.

with J.Philipp Reiss, Experimental Economics, 117(4): 586–614, 2014, download

• Do people respond to incentives of credit contracts in a limited-liability setting as predicted by theory? The answer is yes, but only after sufficient learning opportunities.
• After learning, non-monotonic contracts perform much better than standard debt contracts. However, under non-monotonic contracts, „entrepreneurs“ misreport their earnings in order to repay less, albeit much less so than they could (mirroring the lying literature).

with Arne R. Weiss, Economics Bulletin, 33(1): 794–796, 2013, download

• In elections that are based on self-descriptions of the candidates, voters on average expect elected candidates to line their pockets more than randomly-selected candidates. However, this is not the case. The difference in voter-expectations comes exclusively from those who voted for the losing candidate. They are all too pessimistic with respect to the winner. A comforting finding for real elections, if your candidate didn‘t make it this time...

Journal of Theoretical Biology, 315: 128–138, 2012, download

• Asks whether the mix of behavioural types typically observed in public-good experiments could `survive‘ in an evolutionary-game-theory model (even though punishment and counter-punishment carry costs). Answer: yes it can, so we shouldn‘t be surprised to see what we see.

with Andreas Nicklisch, Journal of Economic Behavior & Organization, 84(3): 892–905, 2012, download

• Shows that the popular reciprocity models systematically miss an important aspect: Most of the time, people don‘t kill others just because these others were slightly unkind even if doing so was costless (i.e., marginal punishment costs were negligible; the point is important because it means fitting any of the models and predicting out-of-situation is doomed to fail whenever marginal punishment costs differ).
• Increasing the fixed costs of punishment makes punishment more extreme, but only among the `reciprocity types‘: the fractions of selfish and inequality-averse players are small and stable.

with Andreas Nicklisch, Journal of Public Economic Theory, 13(5): 791–827, 2011, download

• Examines the contribution norms in a public-good setting by looking at punishment patterns.
• Full contribution is the norm, even though people tend to hold others to the standard of their own contribution: the more tokens I keep to myself, the more likely I will be punished; however, this effect is much stronger once I contribute less than my potential punisher.
• In the first (of many potential) punishment stage(s), we find no evidence of anti-social punishment (in the sense of players punishing more cooperative others). This strongly suggests that anti-social punishment in typical public-good experiments is mostly `pre-emptive retaliation‘.
• When others are allowed to voice their opinion on whether a player should punish a certain other player before the potential punisher decides about how strongly to punish (if at all), the punishment choices rely on the observers‘ judgements rather than directly on the norm. However, the judgements follow the same principles as the norm.
• Precedes a similar and often-cited GEB paper with similar conclusions by two years.

with Hong Geng and Arne R. Weiss , Journal of Public Economic Theory, 13(5): 695–719, 2011, download

• Why do citizens fare better when there are (fair) elections? Why do elections keep the elected from lining their pockets too much? Reciprocity towards the electorate? No: it is their promises:
  • In elections based on self-descriptions without promises, the elected appropriate (as) much of the cake (as randomly selected people).
  • In elections based on promises, the elected on average feel partially bound to their promises (they knowingly let their voters down, but not completely). Put in terms of a simple guilt-aversion model: candidates‘ second-order beliefs match voters‘ first-order beliefs, albeit at a more pro-social level than candidates‘ actual choices.
• Precedes a very similar and often-cited AJPS paper by three years.