Experimental Design Details
The primary outcome of interest is (1) a user’s decision to undergo the assessment after the prompt. Additionally, we will consider (2) the attrition rates measured as a difference between the number of people who started the assessment and those who completed it, (3) time needed to complete the assessment, (4) if users open the report after completing the assessment.
Due to the field setting of the experiment, we do not have comprehensive data on each user’s demographic or other potentially relevant characteristics and have only limited ability to elaborate on mechanisms. The user data is anonymized.
As users were randomized at the registration, main outcomes will be analysed using t-tests. In addition, the compliance decision and attrition will be analysed using probit regressions controlling for a proxy of a geographical location as supplied by the data provider, age and gender as estimated from the chat data in the app. Additionally, we will use machine learning techniques to analyze the chat data (where available and feasible). In the human condition, we will control for the gender of a human mediator as revealed by the displayed name and estimated via gender.api.