Rational Inattention in Online Dating

2018-05-07

2018-08-31

1 Whether online users acquire information about a potential partner based on their race, and
2 Whether attention and endogenous information acquisition affect the number of replies and their length received, within different races and genders

1 Attention is measured by the number of clicks on the user's profile, and
2 Whether a user replies or not is a binary variable (yes/no). The length of the reply is measured in characters, the exact content of the reply is not recorded nor analyzed.

We often exhibit strong same-race preferences when deciding whom to date. Recent theoretical work (Bartos et al, 2017) suggests that an underlying cause of this phenomenon is that we have limited attention to pay to potential partners. If this theory is correct, it would imply that intraracial preferences either become milder or disappear when agents pay attention to all potential partners. I intend to test this theory using a field experiment. The details of it are provided in the following hidden box, which will become public after the experiment is completed. Making the details of the experimental design public may compromise the research project.

In the field experiment, I will contact online dating users from three virtually created accounts. The accounts represent virtual persons of different race, but identical in every other aspect: meaning that they describe in their profile to have the same level of education, the same height, the same profession, etcetera (the website requires users to fill these characteristics). They will contact once each person with a short polite message. We will monitor when people pay attention to the sender (i.e. read their profile) and when do they reply to the original message. Privacy will be protected and it will be impossible to determine which users were contacted: the only way to link a particular observation to a particular user is via the username, which will be deleted and substituted by a number ID. The key to match users with their number IDs will be deleted. Names and related sensitive information will not be recorded.

All users contacted will be over 25 years old, and they will be contacted at most once. Those who reply will be debriefed.

The sample size will be
1110 observations for heterosexual female users
1110 observations for heterosexual male users
432 observations for homosexual male users

Each group will be equally split between control and treatment groups.

Randomization done in office by a computer.

Individual

Yes

3 clusters depending on the characteristics of the receivers. Two of them have 1,110 receivers (total 2,220) and one has 432 receivers, due to the small number of users who fit the third characteristic.

2*1110=2,220 + 1*432=432

Each of the three clusters will be equally divided between control and treatment. Therefore, the treatment group is 555 for the first two clusters, and 216 for the last cluster.

2652 total participants give us 0.8 power to detect an effect size of 0.25 of a standard deviation between the treatment and the control group in the main study at a .05 significance level.