Peer Effects In Street Donations: A Field Experiment

Treatment 1 -- Peers' visible donations with two levels: seeing someone donate and not seeing someone donate. In this treatment, we use natural variation of donations (whose car is near you in the light is random) as well as experimentally manipulated donations. More precisely, we record the donations made by regular drivers. However, since these donation events are rather infrequent, we also hire two drivers who position themselves in the front rows in the red light (so they can be seen) and make donations every other light
Treatment 2 -- Gender of actors with two levels: male vs. female
Treatment 3 -- Income background appearance of actors with two levels: low vs. high

2020-04-15

2020-10-15

Probability of donation and donation magnitude made by passing-by vehicles

Survey question 1: "Do you believe that giving to a street performer is similar to paying for a service or making a donation?"
Survey question 2: "If you see other people giving, would you decide to give or not to give, and why?"

Question 1 aims to determine whether people perceive giving to street performers as charitable giving, and question 2 helps us disentangle the mechanism behind peer effects on people's donation behavior.

The unit of analysis in our experiment is the vehicle. We study the behavior of passing-by drivers and passengers in terms of how often they donate to a street performer and how much.

We deploy a between-subject design in an experiment with two components. In the first component, called "peers' visible donations", we use a hybrid of experimental and observational approaches to assign units of analysis (vehicles) to treatments. The two treatments in this component are: seeing someone donate and not seeing someone donate.
We use natural variation of donations (whose car is near you in the light is random) as well as experimentally manipulated donations. More precisely, we record the donations made by regular drivers. However, since these donation events are rather infrequent, we also hire two drivers who position themselves in the first row in the red light (so they can be seen) and make donations every other light.

The second component of the design is purely experimental and it manipulates the income-background appearance and gender of the performers. We utilize a 2 by 2 full factorial design to test their individual and interaction effects on people's willingness to donate. More precisely, we will have four conditions: (1) male, low-income appearance; (2) male, high-income appearance; (3) female, low-income appearance; (4) female, high-income appearance. All actors perform the same act repeated times in each red light.

Design details:

Our experiment is conducted at specific traffic light locations on the streets of Lima, Peru. Each team of experimenters will consist of one juggler/actor, two drivers, and three observers. The selected streets may range from two to four lanes.

When the traffic light turns red at the designated spots, the actor will start his/her presentation. While the actor makes her presentation, the observers will position themselves on both sides of the arriving vehicles and take notes of the characteristics of the vehicles as well as the people inside. Upon completion of the presentation, the actor will pass through vehicles to receive money. Once the actor has passed through all the vehicles, he/she will approach one of the observers to report the amount of money collected from each vehicle. In our experiment, the measurement unit will be vehicle-based, not individual-based as the number of individuals may vary in each vehicle.

In order to accurately collect the donation data of our experiment, we designed a special data filling booklet to facilitate the process of information collection and minimize the error rates. On each page of the data filling booklet, we have boxes representing the place of each vehicle relative to the traffic light. After the performer collects donations, he/she will approach the observers with the booklet and the reported donation amount will be recorded in the box corresponding to the specific vehicles.

Days after the experiment, we implement surveys in the city to collect descriptive statistics to disentangle the possible mechanisms of peer effects.

Visible donation treatment: randomization done by every red light, i.e. our drivers make visible donations every other red light
Performer race and social background treatments: randomization done in office by a computer

Experimental sessions (every red light is one session)

Yes

225 sessions (red lights)

2000 vehicles

500 vehicles for each of the gender*income cross treatments
donation visibility treatment has natural variations in each round, expected number of vehicles that see others donate will be 600

The minimum detectable effect size for the main outcome will have a cohen's f of 0.07. With 2000 observations, we will be able to obtain a power that's at least 0.8