How accurate are app-based farmer self-reports? Evidence from coffee production records in Burundi
Year: 2026
Type: preprint
Abstract: Governments and donors increasingly collect agricultural data through digital tools, but evidence remains limited on how production estimates from farmer-mediated tools compare with those from enumerator surveys. We address this with coffee production data from about 300 farmers in Burundi. Farmers logged sales in a mobile application and answered an end-of-season in-person survey (CAPI); we benchmark both against records from six Coffee Washing Stations where cherry is weighed. App reports fall 21 per cent below the benchmark; the in-person survey stays within 4 per cent. The app's ranking accuracy varies with use: among farmers with above-median participation, the app ranks farmers about as well as the in-person survey (Spearman 0.82 against 0.87, not significant); below the median it does worse (0.60 against 0.83). The shortfall comes from unlogged transactions rather than understated amounts, and it grows with recorded production, so a constant correction cannot remove it. Our results suggest combining the two modes rather than replacing one with the other: app data can rank farmers where use is sustained, but aggregate production figures still require enumerator surveys or administrative records.
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
SDG: Zero hunger
Open Access status: green
OpenAlex ID: W7223465943