Experimental Design Details
Incentivized tasks:
1) Risk aversion is elicited using the static version of the Bomb Risk Elicitation Task (BRET, Crosetto and Filippin, 2013). The BRET is sufficiently intuitive and does not require to describe explicitly all the lotteries in terms of outcomes and the corresponding probability. This feature renders the task suitable for an online study where the experimenters do not have much control over the instruction phase.
2) Trust and trustworthiness are measured by means of a mini-trust game involving binary options. The first player decides whether to keep one dollar or to pass it to the second player. If passed the second player receives 4$ and decides whether to keep the whole amount or instead to return 2.2$ back. As noted by Ermisch and Gambetta (2006), this structure of the game without equal payoffs prevents choices to be driven by reasons of fairness.
A discrete number of choices is necessary to play the game in strategy method. The binary choice has been chosen in order to reduce the noise due to confusion and errors.
The game is played one-shot with role uncertainty, i.e. choices are made both as the first and as the second player with resolution of uncertainty at the end of the study. Role uncertainty has the advantage of increasing the number of observations collected, but it can affect the choices. In fact, subjects decide behind a veil of ignorance, thereby being more likely to identify themselves with the opponent when making a decision. This caveat does not constitute a major problem in our setting, because the design entails repeated decisions and we are interested in the change.
3) Solidarity is measured using a simplified version of the classic Solidarity game (Selten and Ockenfels, 1998) played one shot. In this three player game each subject has a 2/3 chance of getting 3$ and must decide before uncertainty is resolved (at the end of the study) how much to pass to the ex post loser in case (s)he wins. The player can pass any amount (from 0 to 3$ in 50 cent steps) to the loser with the possibility of giving different amounts to the two opponents, something that is relevant towards the Ingroup-Outgroup analysis (see below).
Every player is assigned four winning numbers between 1 and 6 and the outcome is decided rolling a dice. The main difference as compared to the classic Solidarity game is that in our version a single dice rolling is made (by the experimenters) for all the three players, rather than resolving uncertainty independently for every player. Assigning 1,2,3,4 to player A, 1,2,5,6 to player B, and 3,4,5,6 to Player C the common resolution of uncertainty ensures that one and only one player loses in the game.
This simplified version retains the incentives and the main features of the original game, while at the same time reducing the range of possible outcomes by fixing at two the number of winners.
Treatments
The design entails a mixed between-within subject design that allows us to use a Difference-in-Difference (DID) framework to get rid of time-constant observable and unobservable heterogeneity.
Main treatment: stay at home policy
In wave 1 (31th March) we administered the incentivized tasks to subjects in a city (New York) with a stay-at-home policy already in place since March 20 as well as and in two cities in the State of Arizona (Phoenix and Tucson) on the first day in which a stay-at-home order took effect.
We launched the first wave when it was morning in Arizona, while the order took effect at 5 p.m. The effects of the stay-at-home policy cannot be regarded as instantaneous, and therefore we can safely assign the respondents in Arizona to the “No” condition. In any case, we have the possibility to test this assumption. Given that more than half of the participants in Arizona responded before the order took effect, we can check whether their choices significantly differ from those of the other subjects, who participated while the order was already effective. The orders are valid until April 30 in Arizona, and until further notice in New York.
Hence, we identify the following conditions according to the natural experiment implied by the different timing in the introduction of the stay at home policy:
Wave 1 Wave 2
Arizona (AZ) NO YES
New York (NY) YES YES
Additional treatment: Ingroup-Outgroup
Social distancing does not necessarily affect subjects’ attitudes towards others in a unique direction, as the effect may depend on the closeness of the relationships. For instance, social distancing can strengthen the ties within a community increasing solidarity, trust, and trustworthiness, while at the same time triggering the opposite reaction towards people outside the community. For this reason the design of the experiment also entails a between subject manipulation in order to separate Ingroup Vs. Outgroup effects of social distancing on trust, trustworthiness, and solidarity.
Prolific allows the researcher to filter the participant according to the State of residence. In the screening phase of the study we have asked the participant to report their city of residence. Using this information we measure whether the behavior of the participants differs towards other people living in the same city Vs. in a different State. In case of the Phoenix metropolitan area we can distinguish between the different cities and suburbs (Phoenix, Mesa, Chandler, Scottsdale, Glendale, Gilbert, Tempe).
In the trust game we implement an Ingroup-Outgroup manipulation in a between subject design. In the Ingroup condition each participant is told that (s)he is matched with an anonymous participant from the same city. In the outgroup condition a participant from Arizona (New York City) knows that the opponent is from the State of New York (Arizona), ceteris paribus:
-Treatment Trust Ingroup (90 AZ, 90 NY): two players from the same city are matched (N.B. city not State)
-Treatment Trust Outgroup (90 AZ, 90 NY): one player from Arizona and one from New York are matched.
The solidarity game involves 3 players. In every game there are two players from the same city and one from the other State. The sample of invited participants is therefore split in two groups:
-Treatment Solidarity NY-NY-AZ (60 AZ, 120 NY)
-Treatment Solidarity AZ-AZ-NY (120 AZ, 60 NY). In this case combinations the two players from Arizona comes from the same city (Tucson, or the same city within the Phoenix metropolitan area)
The Ingroup-Outgroup favoritism can be measured as the difference between the amount given to opponents that only differ for their origin. For instance, suppose that Player A and B are from New York City and Player C from Tucson. Player A has two anonymous opponents, and she knows that they only differ because Player B is from New York, while Player C lives in a different State (Arizona). The difference in the amount passed to Player B and Player C can be used to proxy the Ingroup favoritism of Player A. The same argument applies to Player B. In contrast, Player C faces two identical (Outgroup) opponents and therefore we do not expect any systematic difference in the amount passed to them. The (average) choice of Player C can be regarded as a pure Outgroup decision, while where the two opponents have a different origin we measure an Ingroup favoritism as the sum of Ingroup-Outgroup effect. Hence, whether the amount passed to Outgroup opponents does not systematically differ according to the number of Outgroup opponents, i.e. whether the decision is taken in a NY-NY-AZ or AZ-AZ-NY combination, constitutes a test of a pure Outgroup effect.
Independent Observations
We adopt a “Perfect Stranger Matching” approach: participants are explicitly told that they are never matched with the same player across all tasks. Given that there is no feedback until all choices are made, the decisions at the individual level can therefore be regarded as independent observations.
Protocol
Wave 1:
1) BRET
2) Mini-trust game with role uncertainty and Ingroup-Outgroup treatment
3) Reduced version of the Solidarity game with Ingroup-Outgroup distinction
4) Questions, also including the amount of interpersonal relationships (expected as well as retrospective)
Wave 2:
1) BRET (asking whether the subject wants to change the choice in wave 1)
2) -3) Same as wave 1
4) Questions, also including the amount of interpersonal relationships (expected as well as retrospective) as well as on the impact of the pandemic.
Payment Protocol (conditioned on the successful completion of both waves)
Fixed payment of 1.5$ for each wave, plus incentivized tasks:
Bomb task. The expected payoff is 2.5$ assuming risk neutrality (min 0, max $10), paid once. Fairly likely the realized payoff will be lower (say in the order of 2$) because typically subjects are risk averse on average. In the literature about the elicitation of risk preference it is well-known that choices repeated over time tend to be inconsistent. In order to decrease the amount of noise, in wave 2 we plan to remind the subjects about the choice in wave 1, telling them that they are free to change it. Having the choice appearing as the change of the first, the Bomb task will be paid only once. At the same time, systematic changes of the choices will signal a shift in the risk aversion of (some groups of) subjects.
In the trust game the individual earning is between 0 and 4$. The average payoff depends on the fraction of subjects passing the dollar as first mover in the game. Based on previous experiments it is reasonable to assume that about a half of the players will do so, implying an average payoff of 1.25$ per capita.
In the solidarity game there are always two players out of three earning 3$, while the other gets nothing. The average earning is therefore 2$ per capita by construction, regardless of the amount of money passed among subjects before the resolution of uncertainty.
In each wave we pay one of these two games at random.