At intake we record how often each rider currently rides. A few months later we follow up to ask again — in person or by SMS. The increase is attributed to our program.

Extrapolated across 6,150 unique participants (name + contact + setting deduplication):

6,150 unique receivers × global survey mean (8.72 extra rides/month) × 12 months ≈ 635,000 extra rides/year

Individual settings on the map use a more precise estimate. Rather than a hard cutoff — a setting's own average once it has 5+ survey responses, otherwise the global average — each setting's figure is now blended between its own average and the global average using empirical-Bayes partial pooling, weighted by how much survey data that setting has. Well-surveyed settings rely mostly on their own data; lightly-surveyed settings lean more on the global average. This avoids both overfitting to small samples and flattening genuine differences between settings.