Paternalism with hindsight: do protégés react consequentialistically to paternalism?

We investigate experimentally whether the protégés’ reaction to paternalism depends on the consequences of the paternalistic action to their well-being. Thus, our research is concerned with the perception of paternalism by those who are directly affected by it, rather than with the justifiability of paternalism from an ethical perspective. We find that the protégés punish a paternalist restricting their freedom of choice. Yet, this negative reaction is not based on principled grounds because, with the wisdom of hindsight, the protégés punish the paternalist only if the restriction makes them worse off. Conversely, if the restriction makes them better off, the protégés on average do not punish and, sometimes, they even reward the paternalist. This suggests that the protégés take predominantly a consequentialist stand on paternalism. In addition, a regression analysis reveals that our main finding is not altered when we control for the intentions (malevolent vs. benevolent) that the protégés attribute to the paternalist.

1. Johansson-Stenman (2012) is a notable exception. Based on survey evidence, he studies whether people see an action as ethically bad because of its overall consequences or on some other grounds. His results suggest that most people are consequentialists in their ethical perceptions. While Johansson-Stenman focuses on people’s normative views, we look at people’s actual behavior.

2. The laboratory provides an ideal environment for studying unambiguous outcomes, whereas real-life outcomes can be more or less ambiguous depending on the context.

3. The choice of a lottery seems to be the appropriate starting point for an experimental analysis because paternalistic policies often deal with choices whose outcomes are risky/uncertain.

4. Apart from the first sentence, we formulated the instructions as neutrally as possible, avoiding suggestive terms like patron and protégé. A translated version of the instructions can be found in the supplement.

5. Although in our experiment placing a positive bet does not hurt the protégés in expected value terms, the patron may still be motivated by benevolence and restrict gambling choices in order to protect the protégés against a negative outcome.

6. By this means, the protégés could, without cost, express disapproval or approval of the paternalistic action. An alternative would have been to make the choice of \(\varDelta \) costly. Asking the protégés to pay for altering the patron’s endowment would reveal if their feelings are strong enough to let them incur the (chosen) cost. Yet, if the cost is perceived to be too high, protégés may be reluctant to pay it, thus biasing our results against the hypothesis that consequences matter. To avoid this possibility, we opted for a design where the choice of \(\varDelta \) is free.

7. The strategy method has considerable advantages compared to the direct response method, e.g. it facilitates collection of richer data sets and makes people think harder about all possible outcomes. Its main disadvantage, however, is that it may induce experimenter demand effects (Zizzo 2010). For example, in our design, the strategy method could cause subjects to make different \(\varDelta \)-choices for different levels of \(R\). While we acknowledge this potential problem, we note that we are not interested in how \(\varDelta \)-choices vary with \(R\), but in differences between treatment groups.

8. Nor can differences in \(\varDelta \)-choices conditional on \(R=0\) between treatment groups be explained by efficiency concerns (i.e., interest in increasing the sum of payoffs) and competitive preferences (i.e., desire of lowering the others’ payoffs). Since random assignment to treatments should minimize differences between the groups, concerns for efficiency and competitiveness could question the interpretation of our data only if all protégés, regardless of the treatment group, were choosing \(\varDelta =50\) and \(\varDelta =-50\) respectively.

9. While we acknowledge that preferences for inequality may interact with risk attitudes (for experimental research on this topic, see Amiel et al. 2001; Kroll and Davidovitz 2003; Carlsson et al. 2005; Bolton and Ockenfels 2010), our aim here is to provide a simple scenario that clearly illustrates why we state and test our hypotheses conditional on \(R=0\).

10. It is worthy of note that out of 367 protégés, 55 refrain from betting (\(x= 0\)), 20 bet their entire endowment (\(x=100\)), and the remaining 292 bet a positive amount lower than the endowment (\(0<x<100\)). The decision of how much to bet in the lottery resembles the risk elicitation method used by Gneezy and Potters (1997) who are interested in differences in risk taking between treatments and, differently from us, consider a lottery with a positive expected return. In our setting, a risk averse participant would always prefer keeping the endowment rather than placing a positive bet and possibly receiving nothing. Thus, a protégé that bets a positive amount is not risk averse, but either risk neutral (if \(x \le 100\)) or risk seeking (if \(x=100\)). This means that, in our experiment, close to 15 % of all protégés behave in a risk averse manner. This is in sharp contrast with previous results showing that about 80 % of participants are risk averse (see, e.g., Dohmen et al. 2011) and may be due to the fact that some people do not like making extreme choices. However, it is not the purpose of this paper to assess risk preferences.

11. The median values equal zero in all except two cases, both of which refer to protégés in the positive outcome-hindsight treatment group (namely median(\(\varDelta \))=0.50 and 1.50 when \(R=80\) and \(100\), respectively).

12. We will provide an interpretation of \(\varDelta \)-choices in the foresight treatment in Sect. 4.3.

13. Note that we find no difference when comparing \(\varDelta \)-choices made in the foresight treatment with average \(\varDelta \)-choices over the two hindsight treatment groups (\(p\) value 0.527; two-sided Mann–Whitney test). This suggests that protégés are consistent in the sense that their reaction after observing the realized outcome does not, on average, contradict their reaction in foresight.

14. Benevolent intentions were identified by asking the protégés (in the post-experimental questionnaire) “Why do you think that the participant in role B set a limit to your choice?”. If a protégé thought that the most correct option was either “She does not want to see that I lose money” or “She is benevolent”, the variable “benevolence” takes value 1; otherwise, it is coded 0. The complete list of options available to the protégés is given in the supplement.

15. To ease interpretation of the coefficients, the emotion variables are centered. This means that the average self-reported emotion is subtracted from each observation of these variables (the interaction terms are also computed using the centered variables). The effect of “outcome” on the dependent variable is therefore given for the “average” person on the emotion scale.

16. See Verme (2009) for a more comprehensive list of possible views of how people value freedom.

17. Recall that any positive restriction would result in different actual bets, so that inequality aversion might explain \(\varDelta \)-choices.

18. Unawareness of the lottery outcome appears to be the best scenario for looking at this issue, since the analysis in Sect. 4.2 has shown that knowledge of the outcome affects average \(\varDelta \)-values.

19. Note that subjects are not randomly assigned to the two groups that are compared. Rather, they are assigned to the groups depending on whether their intended bet was greater than or equal to zero. Consequently, the \(\varDelta \)-choices of the protégés in the two groups may be different for reasons other than whether or not the restriction is binding. For example, protégés who wanted to bet zero are likely to be risk averse while the others are not. Our conclusions then hold only under the assumption that people with different risk attitudes do not value freedom of choice differently. Although we are not aware of any evidence to the contrary, we acknowledge that the results are suggestive rather than conclusive. Not distinguishing between the two concepts of freedom provides a way to make controlled inference. Based on a sign test, we can reject the null hypothesis that average \(\varDelta \)-choices are equal to zero, in favor of the alternative that they are negative (\(p\) value 0.004). That people value freedom of choice is thereby confirmed.

We thank the managing editor Marc Fleurbaey, the editor in charge, and two anonymous referees for their constructive comments. We also thank Dominique Cappelletti, Alon Harel, Elina Lampi, Matteo Ploner, and Birendra Rai for helpful suggestions. The authors bear the responsibility for any errors and omissions.

Authors and Affiliations

1. Max Planck Institute of Economics, Kahlaische Strasse 10, 07745 , Jena, Germany

   Mitesh Kataria & M. Vittoria Levati

2. Department of Economics, University of Verona, Via dell’Artigliere 19, 37129 , Verona, Italy

   M. Vittoria Levati

3. Technical University of Munich, Lothstrasse 17, 80335 , Munich, Germany

   Matthias Uhl

Corresponding author

Correspondence to M. Vittoria Levati.

Reprints and permissions