Extended Data Table 5.

Overlapping sources of exclusion from rural Novissi

Progressive sources of attrition from the rural Novissi program, where each row shows exclusion conditional on exclusions from preceding rows. The final three columns show characteristics of the population “succeeding” at each step. Panel A: Results estimated using the 2020 phone survey (N = 8,915). Panel B: Results estimated for just the poorest 29% from the 2020 survey (N = 3,209). There is no attrition based on mobile phone ownership or past phone use in this sample (in contrast to Table 2) since only active phone users were sampled for the phone survey. Values reweighted using sample weights. (In some cases, sample weights create large differences in the weighted and raw percentages. For instance, 5,145 out of 8,898 voters (57.8%) attempt to register (Panel A), but the weighted percentage is 45.5%. The importance of sample weights is consistent with the wide distribution of sample weights shown in Supplementary Fig. 10).