Back to History

Fields Changed


Field Before After
Last Published June 08, 2019 02:12 PM June 10, 2019 02:58 AM
Was the treatment clustered? No Yes
Planned Number of Clusters 750 users drawn from 750 distinct households 150 grids of 100 x 100 meters
Sample size (or number of clusters) by treatment arms 250 users in control 250 users for Treatment 1 250 users for Treatment 2 50 grids with 5 users and total of 250 users each in control, treatment 1, and treatment 2
Power calculation: Minimum Detectable Effect Size for Main Outcomes We carried out power calculations for our primary outcome, using the mean and standard deviation of the probability of correctly identifying a news as true or false, observed in the pilot data. Our calculations show that with a cluster randomized design, and user level outcomes measured at baseline and endline, assuming a significance level of 5\%, 80\% power, cluster size of 5, and intra-class correlation of 0.05, and 50 grids per arm, it will be possible to detect a standardized effect of 0.30. This amounts to a 23\% increase in the mean of the primary outcome from 0.63 to 0.77 (standard deviation 0.48). We expect to gain statistical power as we will measure and control for user's baseline score in correctly identifying news as well as user's baseline characteristics.
Back to top