Higher socioeconomic status does not predict decreased prosocial behavior in a field experiment

Does higher socioeconomic status predict decreased prosocial behavior? Methodological issues such as the reliance of survey studies on self-reported measures of prosociality, the insufficient control of relative incentives in experiments, and the use of non-random samples, have prevented researchers from ruling out that there is a negative association between socioeconomic status (SES) and prosociality. Here, we present results from a field experiment on the willingness of unaware individuals of different SES to undertake an effortful prosocial task—returning a misdelivered letter. Specifically, using the rental or sale value of homes as indicators of SES, we randomly selected households of high and low SES and misdelivered envelopes to them. Despite controlling for numerous covariates and performing a series of ancillary tests, we fail to find any evidence that higher SES predicts decreased prosocial behavior. Instead, we find that misdelivered letters are substantially more likely to be returned from high rather than low SES households.


Supplementary Note 1: Details on the selection of high and low SES households
All households were located in the same medium-sized city in the Netherlands. Households were randomly selected through a two-step process. In the first step, to ensure that we selected either high or low SES individuals to participate in our experiment, we compiled lists of high and low SES households following the procedure detailed below. In the second step, we randomly selected households from these lists for participation.
Our first concern was to compile a list of high and low SES households. CBS Netherlands was willing to provide us with such data for each household after the experiment was conducted. This way, they could encrypt the data, guaranteeing the privacy and anonymity of all participants. For safety and privacy concerns, no institution was willing to share with us a list that would be linking individual addresses with household wealth, prior to data collection.
For this reason, we had to compile our own list of households to select from using a measure that would be highly correlated with wealth. We used the property value of one's house for high SES individuals, and the rental price for the low SES individuals.

Selecting low SES households
The city has several social-housing corporations. For one of these, the explicit purpose is to rent out apartments to the lowest SES people in the city. The social-housing corporation was kind enough to share with us a list with the addresses of their cheapest apartments, from which we selected 152 households. As this number did not suffice for our study, we selected additional apartments by using the website of the same corporation in which apartments are advertised for rent. In particular, we selected apartments in listed buildings, but not the apartments that were for rent.
After compiling the list of low SES households, we randomly selected 225 of them to participate in our experiment (180 used in our main experiment and 45 used in the "Low SES Joost" treatment), provided that two conditions were met. First, most apartment buildings have four stories and share entrances with eight apartments. Although an official from the social-housing corporation informed us that residents have little contact with each other, we decided to limit misdeliveries to three mailboxes per entrance to minimize communication between households. We randomly selected 159 households sharing an entrance with two other apartments in the same building, and 66 households that do not share an entrance. We find no economically or statistically significant differences across these two groups. Households sharing entrances returned 38.4% of misdelivered envelopes while household not sharing entrances in our sample returned 47.0% (N =225, p = 0.24, two-tailed, Fisher-exact).
The second condition was that the family living in a selected household is native Dutch.
The reason is that the ethnicity of the sender and receiver can influence pro-social behavior [1,2,3,4]. This could be a problem as non-natives in the Netherlands tend to have a greater likelihood of having a lower SES. To avoid this confound, we took pictures of all family-name signs on doorbells and ensured that all apartments in our sample were likely to be native Dutch. Using data from CBS Netherlands, we can control for the effect of any remaining non-Dutch households.

Selecting high SES households
To compile a list of high SES households, we started by consulting www.Funda.nl, a website in the Netherlands that advertises houses for sale. Our goal was to identify neighborhoods and streets where wealthy individuals live. To do this, we found all houses with a sale price of at least e750,000. All houses on the same street as the identified houses were included on our list of high SES households, excluding the houses that were up for sale. Once this procedure was done, we consulted www.Postcode.nl. This website contains information on all houses in the city in which our experiment was conducted such as the postal code, house number, and surface. We used this website to include on our list all houses that were not for sale, but had a comparable surface as those that were for sale.
Once the list was compiled, we randomly selected high SES households to deliver the misdelivered letter. To minimize effects of communication between subjects, we excluded houses that were close to each other. A total of 100 houses in our sample have no neighboring house that is selected for any of the core treatments. The other 80 houses do have at least one neighbor that is a subject in one of the core treatments. (Almost all of these houses are villas with a large surface and driveways that are far removed from the driveway of the neighbor.) Differences in return rates between houses with a neighbor (76.3%), or without neighbour (85.0%) are insignificant (Fisher exact test, p = 0.18).

Supplementary Note 2: Household-level data and randomization check
We collected information from CBS Netherlands on the households in our sample for three purposes: (i) to ensure that we correctly identified high and low SES households, (ii) to ensure that our randomization of households into treatments was successful, and (iii) to check the robustness of our experimental findings when adding additional controls. An overview of this can be found in Supplementary

Definitions of socio-economic variables
CBS Netherlands provides administrative data on all citizens of the Netherlands. Data are available from various sources, and they can be merged on an individual or household level.
To guarantee anonymity, CBS Netherlands encrypts each person and household by means of a code called RIN-address and RIN-person. CBS Netherlands has encrypted the addresses in the database that we provided them, by adding the variable RIN-address. Then, we merged different databases of CBS Netherlands by using the RIN-address variable. Supplementary Table 2 gives an overview of the different databases that we used. 1 The number of observations per treatment was 90 in all cases. CBS Netherlands was unable to provide us with information on certain variables in a small number of cases, as noted.

Supplementary Note 3: On the envelopes and their content
The letters misdelivered in the experiment contained either a banknote of e5 or e20, or a bank-transfer card (BTC) of the same value. The analysis presented at the paper pooled the data from the different amounts. Here, we present the disaggregated analysis. Let BTCx and Ex denote the treatment with x ∈ {5, 20}. Return rates by treatment can be seen in Supplementary Fig. 1. For high SES households, the return rate decreases insignificantly from 87% in BTC5 to 84% in BTC20 (N = 90, p = 1.00, two-tailed, Fisher-exact). For low SES households, the return rate decreases from 51% in BTC5 to 44% in BTC20, but the difference is again insignificant (N = 90, p = 0.67, two-tailed, Fisher-exact). The difference in return rates between E5 and E20 is insignificant both for high SES individuals (82% vs. 71%; N = 90, p = 0.32, two-tailed, Fisher-exact) and low SES individuals (33% vs. 22%; N = 90, p = 0.35, two-tailed, Fisher-exact). Importantly, we find that the high SES group is more likely to possibility that someone could do X, there is no reason to be suspicious. This is why we decided to have the sender be a grandfather; a person that may recall more innocent times and be unaware of the dangers of sending money via post.
Anecdotal evidence from our experiment does not hint to participants becoming suspicious.
Some individuals (N = 16) returned our envelopes after placing them in non-transparent A4sized envelopes. All of them were accompanied by letters explaining that the mailman made a mistake (and usually added that Joost should advice his grandfather not to send money via post). None of these letters suggested that those returning the envelopes were suspicious.
Still, in light of the above, even if some people became suspicious, we would expect them to be a small fraction of our sample. Therefore, we have no reason to expect that they would have a noticeable effect on our estimates, even if we had reasons to expect low SES individuals and high SES individuals to be differentially affected -which we do not.

Supplementary Note 4: Robustness checks
In this section, we present additional robustness checks by extending our regression analysis (see Table 1) to control for a number of variables obtained from CBS Netherlands. Supplementary Table 3 shows linear probability models. These controls are: (i) "Density" which measures the ratio of houses in a street participating in our experiment over the total houses in a street; (ii) "HH Foreign" is a dummy variable taking the value one if all adult members of a household are non-Dutch; (iii) "Benefits" is a dummy variable taking value one if at least one adult member of the household receives unemployment benefits, social welfare, social security, or disability insurance; (iv) "Pension" is a dummy with value one if at least one adult member of the household receives pension; (v) "HH Size" measures the average number of people in a household, including children; (vi) "HH Age" measures the average age of all adults in a household. As can be seen, controlling for these additional variables does not affect our estimates appreciably or our conclusions.

SES individuals
In order to understand whether the lower return rates of low SES households might reflect differences in the use of postal services, we approached males and females on a street in the center of the city. Possible candidates were asked whether they were interested to participate in a short survey. If they responded positively, they were asked if they rented their home from a social housing corporation. If they responded negatively to this second question, then they were not interviewed. We set a target of 45 respondents, i.e., the same number as that of low SES households in each of our treatments.
Overall, 89 citizens responded that they were not interested to partake in our survey.
128 citizens were interested, but did not qualify. All 45 interviewed subjects answered all questions. The interviewer read the questions to respondents sequentially. The responses are given in parentheses in the questionnaire.

The questionnaire
We are conducting a survey regarding the use of postal services in different countries. We would be grateful if you could spent a couple of minutes answering a few questions. Week -0.08 * * -0.08 * * * -0.08 * * -0.08 * * -0.08 * * -0.08 * * * -0.08 * * * -0.08 * * (0. Distance Mailbox -0.14 * * -0.14 * * -0.14 * * -0.14 * * -0.14 * * -0.14 * * -0.14 * * -0.14 * * (0. Estimates from a linear probability model. The dependent variable is a dummy taking the value of 1 when an envelope is returned and 0 otherwise. "High SES" and "Cash" are dummy variables indicating whether an observation is associated with a high SES household or the Cash treatment, respectively. "Week" measures the number of weeks since the last payday. Distance Recipient's house (Distance mailbox) presents the distance from a subject's household to the recipient's house (the nearest mailbox). Standard errors are shown in parentheses (clustered at the street level). The results are virtually identical when we use a Probit specification. Reported results are from two-tailed tests. * * * , * * , * indicate significance at the .01, .05, and .10 levels, respectively.

4.
Has the postal service ever misdelivered a letter in your letterbox that was meant to be delivered to a different address? YES/NO (100% yes) 5. If such a misdelivered letter were put in your letterbox, can you tell me some ways you can use so that the letter is properly delivered to the intended recipient? (Check all that they If YES, about how long would it take you to walk there? minutes (mean=3.2 minutes)

Supplementary Note 6: Survey on trusting institutions
To explore whether low SES individuals trust the public mail company less than high SES individuals, we chose to conduct a survey using Dutch students at Erasmus University Rotterdam. In total, we surveyed 140 students, but omit 7 observations from the analysis as they are from non-Dutch students. Students were given ample of space and time to fill out the survey. During the survey, students were forbidden to communicate with each other. Below, we present the survey instrument, as well as summary statistics in parentheses.

The questionnaire
Dear student, We would like you to spend a couple of minutes to fill out this survey. After everyone is done, we will collect all forms. It is important to realize that there are no 'right' or 'wrong' answers, and that you are fully anonymous. In the instruction letter, mention was made of the principal investigator, along with his contact information and a website that contained more details on the study (see below). The Private Incentive treatment was conducted on one single day in October 2014. To test for noise that could be introduced in our data due to the mail company, we posted ourselves seventeen envelopes from this treatment directly to the university's mail address. These envelopes were sent from the neighborhoods that the subjects live in. As expected, all of these envelopes arrived the day after sending them out. We conclude that our data are free of any noise due to the mail company.

Instructions (translated from Dutch)
To Will I get paid for my participation? Yes you will! In addition, to the enclosed e5 which is yours to keep, if you complete the study, we will send you an additional e20 in cash.
What do I have to do? It is really easy to complete this study. All you need to do is put the card in the enclosed envelope, and then mail the envelope. If you complete the study by October 3, we will send you the additional e20, which you should receive by October 17. If you choose not to mail the envelope, then you will not complete the study, and you will not receive the e20 completion payment. You may naturally keep the enclosed e5 whether you complete the study or not.
Will my personal information be revealed? No, it will not. You will notice that the enclosed card does not have your name on it, but only a participant number. This participant number links the card to your address, so that we can send your e20 for completing the study to your address. We do not record your name. Your participation therefore will be anonymous.
Thank you very much for participating in this research.
The Erasmus University Rotterdam Postal Study (EURPS) is examining how people in different parts of the Netherlands use the postal service and send letters.

Funding for this Research:
This study is supported by grants from the Erasmus School of Economics and the Netherlands organisation for Scientific Research.

Privacy Statement:
The privacy of participants in this research is assured. All addresses used in this study are randomly selected. As is required by research ethics, participants' names are never known to the EURPS researchers, and after the data collection EURPS will destroy any record of the addresses used. There will be no information in our data what will allow us to identify any of the participants. In addition, our report will not mention any specific parts of Netherlands.
The data will not be shared with anyone outside of the research group, and will be used only for scientific purposes. We are studying how people in different parts of the Netherlands use the postal service.
We have randomly selected households to participate in the study. For their effort, e5 was included. The task required subjects to send a card that was given to them. If the card was received, then e20 was returned in cash. The results are interesting to us, because they shed light on the level of compensation needed for people to take the effort to mail a letter.