Neuroeconomics for development: Eye-tracking to understand migrant remittances

1. NGO personnel recruited potential participants who had to meet the following criteria: (A) be older than 18; (B) have sent remittances at least four times in the past year to El Salvador, Guatemala, or Honduras; (C) be able to read and use a computer; and (D) not wear bifocal glasses (for purposes of eye-tracking).

2. Participants gave informed consent, were assigned a study ID, and completed a pre-survey (see Docs & Materials).

3. Each participant reviewed three webpages containing attributes of 10-11 money-transfer operators (MTOs) such as Western Union and MoneyGram (more under Experimental Design). These webpages were offline variants of a World Bank-certified comparison platform. Participants had five minutes to review each page and choose the MTO with which they wanted to send $100 or $300 (randomly determined) to El Salvador, Guatemala, or Honduras (also see step 5). While reviewing webpages and making decisions, participants' eyes were tracked. As part of the experiment instructions, participants were asked not to move back and forth once the eye-tracker had been calibrated, in order to safeguard accuracy of the data.

4. A subset of the participants completed a short post-survey (see Docs & Materials).

5. Participants were paid $50 in cash for completing the study. In addition, one in eight participants won a lottery in which $100 or $300 were sent by means of one of the MTOs chosen in step 3. This increased the chance that decisions made in the study were given careful consideration, as tends to be the case in the day-to-day environment. Lottery winners were notified by phone after the study and asked to confirm the contact information for their preferred recipient in the country of origin. After the money was sent, they were provided with a confirmation number so the recipient could claim the funds.

2017-05-01

2017-10-31

The primary outcome is a participant's choice of MTO on a given results page.

We are also interested in (i) the extent to which a participant chooses the same MTO across all three results pages; (ii) the extent to which a participant's typical choice of MTO (elicited in the pre-survey) is the same as what is chosen in (i); (iii) different measures of visual attendance; and (iv) their interactions.

(i) and (ii) are derivatives of the primary outcome, e.g., they will be measured as a dummy for whether or not choices meet such conditions. (iii) will be constructed based on eye-tracking data. (iv) will be based on (i)-(iii). See the pre-analysis plan for additional details on these two outcome measures.

The experimental design comprised two main treatments: Group A and Group B. Both groups saw three webpages (as explained under Intervention).

The first page was the same across groups. It contained the following attributes for each MTO: the name and logo; the exchange rate, transaction fee, and thus, amount that would arrive; and whether the funds would arrive as a deposit or in cash for pick-up or home delivery.

The second page is where Groups A and B differed. If the participant was in Group A, page 2 added how long the money would take to arrive (i.e., the delivery speed) to page 1. If the participant was in Group B, page 2 added customer reviews instead.

Finally, the third page added customer reviews to page 2 if the participant was in Group A and delivery speed if the participant was in Group B. So, while Groups A and B were the same again by page 3, the order in which additional information had been presented varied.

Assignment to Group A or B was random, since odd study IDs were assigned to A and even IDs were assigned to B.

Participants were also randomly assigned to (i) stakes of $100 or $300 (as referenced previously) and (ii) three possible versions of webpages, which varied in terms of the order of the MTOs and the MTO that had a 50 percent price discount.

The randomization was done in office by a computer.

Individual.

No

394 individuals. Also see next item.

In the NSF proposal, we proposed 400 individuals. We actually ended up with 394 individuals.

There are four treatments: A-$100, A-$300, B-$100, and B-$300.

The 394 individuals should be distributed as follows across these treatments: 115, 77, 115, and 77. This is because there was a 50-50 chance of being assigned to A versus B and a 60-40 chance of being assigned to $100 versus $300. The exact sample size per treatment will be verified once the data are analyzed.

As mentioned previously, in the NSF proposal, we committed to a sample size of 400 migrants. This number was not based on ex ante power calculations for the following reasons. First, the NSF gave a maximum award budget. Considering subject payments, implementation, and other budget categories, we thus committed to a sample of 400 participants across 6 potential between-subject treatments. Second, given the plan to collect the outcomes of interest (in particular, visual attendance) in the field from a sample of migrants with relatively little education and income, there were no reliable priors for assessing possible effect sizes. Third, compared to several studies that use eye-tracking data, N=400 was a relatively large sample. When analyzing the data, we will conduct ex post power analysis along the lines of Maniadis et al (AER, 2014).

Neuroeconomics for development: Eye-tracking to understand migrant remittances