Experimental Design Details
The population of potential participants is composed by the owners of small businesses that completed an open online survey carried out in Qualtrics. They were recruited through social media ads targeted to owners of small businesses. The survey has yielded around 34,000 responses coming from the 8 countries of our study. The eligibility criteria used to select the experiment sample from this population consisted of the following points: (i) provided consent to receive additional information from the research team, and (ii) provided contact information (email and phone number). A sample of 15,367 firms fulfill these requirements.
In Argentina, Bolivia, Brazil, Chile, Colombia, the Dominican Republic, Mexico, and Peru, firm owners were randomly assigned to one of the following arms:
Treatment arm 1: Email
Treatment arm 2: Email + Phone session + Chatbot
Resource constraints prevented us from implementing the chatbot in Argentina, Bolivia, and the Dominican Republic. Firms from those three countries, however, did receive the one-on-one phone session. Therefore, we have (in practice) an “Email + Phone session” arm:
Treatment arm 3: Email + Phone session
Since Mexico was the country with the most respondents, there is an additional treatment arm which consists only of receiving the email and chatbot. This arm was part of the original design: business owners in Mexico were randomly assigned to Treatment arm 1 or Treatment arm 2, previously mentioned, or to:
Treatment arm 4: Email + Chatbot
In sum, there are three comparisons we will be evaluating:
Email (T1) vs. Email + Full information (T2), leveraging our samples from Brazil, Chile, Colombia, Peru, and a subset of the firms from Mexico (N: 10,895)
Email (T1) vs. Email + Phone session (T3), which involves business owners from Argentina, Bolivia, and the Dominican Republic (N:2,804), and
Email (T1) vs. Email + Full information (T2) vs. Email + Chatbot (T4), which involves all Mexican firms (N:3,334).
Notes:
In the baseline survey we asked business owners if they wanted to be contacted by a NGO to receive information about policies that had been announced in their countries. This offer was intended to increase the likelihood that business owners would share their contact information with the research team. Since we generated these expectations, we decided that everyone who showed willingness to be contacted would receive at least the summary of policies through email. For this reason, we do not have a pure control group.
Randomization is stratified by country and date of response.
The baseline survey was initially also targeted at business owners from the United States and more countries from Latin America (Costa Rica, Ecuador, El Salvador, Guatemala, Honduras, Nicaragua, Panama, Paraguay, Uruguay, Venezuela). The firms from the United States were not included in the experiment because 1) first round of aid funds (PPP) ran out when we were planning our intervention, and 2) we encountered difficulties for timely hiring of temporary workers that we would have needed to set up a call center. The firms from the other Latin American countries were not included because the initial take up of the baseline survey was low so we targeted our resources to deploy more ads in the 8 biggest countries.
We also plan to include more Brazilian firms that will be recruited using another strategy: they are sampled from a public registry of firms in the country, that includes contact information. We are still planning the logistics of the intervention with this sample because it is larger than the one we manage so far. Our Pre-Analysis Plan will be updated accordingly when the implementation details are ready.