We use a randomization design which includes cluster-level assignment of treatment saturation and individual-level assignment of treatment. Out of 1.2 million urban households in 3,802 clusters, 100,000 households are treated in a direct-marketing intervention. In the relative saturation design used in this study, each cluster is randomly allocated to one of the five bins (0%, 25%, 50%, 75% and 100%) of saturation, that means in a cluster allocated to 25% bin, only 25% randomly selected households are solicited for charitable donation via direct-marketing.
Within a cluster, households are randomly assigned to treatment groups. The content of the letter is determined using a cross-randomized design where specific messages are provided individually or jointly. On the top of the “basic appeal” content, we add four type of messages- Social distance, Information on the monetary value of subsidy, Signaling, Moral suasion. We further create two variations each under “Information on the monetary value of subsidy” and “Signaling”. This provides us total 36 groups, where a treated household is randomly assigned to. Finally, within the “Social distance” message, we randomly vary the examples on lines of caste-religion similarity.