WEATHER FORECASTS AS A CREDENCE GOOD: VERIFIED ACCURACY, FARMER BELIEF AND THE DEMAND FOR AGROMETEOROLOGICAL INFORMATION IN PAKISTAN

Abstract

A farmer who acts on a weather advisory and still harvests a poor crop cannot tell a bad forecast from bad luck. Forecast quality is consumed jointly with soil, seed, pests and the weather draw, so consumption does not reveal it. Almost every study of weather-information uptake observes what farmers believe about forecast quality but not what that quality is. This paper measures both sides at the same place and the same time. Verified day-ahead accuracy is constructed for 89 Pakistan Meteorological Department stations from 34,610 scored station-days and matched to the perceived accuracy reported by 689 farmers. The level comparison depends on which verification score is used. Against percentage correct, mean belief of 0.692 sits 0.19 below a verified 0.883, but percentage correct is beaten on this panel by a no-skill climatological rule, and the sign of the gap reverses against the hit rate on rain days. What does not depend on the score is the decoupling. Across seven verification metrics the correlation between belief and verified quality never exceeds 0.112 in magnitude at station level or 0.056 at farmer level. Demand follows belief alone. Conditional on the perception, verified accuracy has no detectable effect on willingness to pay or reported outcomes. A learning model with misattribution, at calibrated coding thresholds, reproduces about two thirds of the level gap measured by percentage correct. Its policy ranking places de-biasing extension and localised advisories far above subsidised access.