Experimental Design Details
Experimental Design
We will closely collaborate with two organizations in Armenia: 1000+ foundation and Nikita mobile (www.nikita.am). Since Nikita mobile has a rich database of individuals with their contact details, the database of potential donors will be borrowed from this company. The database includes the mobile phone number of the potential donors and their gender. Unfortunately, the database does not include any other information, such as the donation history of individuals. Thus, we do not know whether these individuals donated to 1000+ foundation before or not.
The intervention will be implemented via targeted text messages and will last for 6 months. Each text message in the experiment will contain:
• A donation request. Depending on the treatment, we will have either an altruistic, or a reciprocal, or an ordinary request.
• A short link to the 1000+ webpage to make the donation. Each individual will receive a trackable individual link through which we can check whether she clicked on the link or not, made a donation or not, and how much money she donated.
• Instructions how to unsubscribe from the messaging campaign. To unsubscribe the individual will need to send STOP to the short number indicated in the text message. To make sure that this message is not case sensitive, the system is instructed to recognize the following words “STOP”, “Stop”, “stop” in English, Russian, and Armenian language (i.e., ՍՏՈՊ, Ստոպ, ստոպ, СТОП, Стоп, стоп, STOP, Stop, stop).
25,600 individuals will be randomly sampled from the database of Nikita mobile and randomized into 6 arms:
- Donation appeal 1 - Low Frequency Ordinary Message [N=4,266]: We send a quarterly neutral solicitation message. This is the control group.
- Donation appeal 2 - High Frequency Ordinary Message [N=4266]: We send a monthly neutral solicitation message (e.g., Please make a donation to 1000+).
- Donation appeal 3 - Low Frequency Reciprocity Message [N=4,268]: We send a quarterly message involving a strong reciprocity component asking to donate.
- Donation appeal 4 - High Frequency Reciprocity Message [N=4,266]: We send a monthly message involving a strong reciprocity component, asking to donate.
- Donation appeal 5 - Low Frequency Altruism Message [N=4,266]: We send a quarterly message involving a strong altruism component, asking to donate.
- Donation appeal 6 - High Frequency Altruism Message [N=4,268]: We send a monthly message involving a strong altruism component, asking to donate.
All the messages will be sent on the same day in batches to make sure that the unusually high traffic does not overload the web-page of 1000+ foundation in the highly unlikely case if everyone who receives the message taps on the link (the probability of a webpage crash is tiny even in this case, however we opt for a safe option). There will be a gap of 10-20 minutes between the dispatch of each batch. The phone numbers of each treatment are randomly allocated to batches and in each batch there are equal number of observations from each treatment.
Frequent monthly messages are sent in an interval of roughly 30 days. We do not send the messages on a given date in each month to exclude anticipation by the recipients. This design choice is motivated by a recent theoretical literature about anticipated and unanticipated reminders (Ericson, 2017), since surprise reminders/messages can be more effective than anticipated ones. In our view, sending the frequent messages on exactly the same day each month can undermine their effectiveness and somewhat reduce the “surprise effect” of these text messages. In other words, the main aim of this design choice is to make the frequent text messages less anticipated.
We investigate the following hypotheses:
H1: The presence and content of the message (reciprocity/ altruism) affect the unsubscription rate, the donation rate and the amount donated.
H2: The frequency of communication affects the unsubscription rate, the donation rate and the amount donated.
We will also explore how the frequency of communication interacts with the presence and content of the message.