Nonclassical measurement error in farmland markets with implications for Ricardian analysis

Nonclassical measurement error threatens the validity of empirical economic models. We examine the extent to which land value measures that are commonly used in studies of the US farmland market are subject to nonclassical measurement error. We consider differences in county‐level land values from two popular data sources: (1) self‐reported land value assessments from the 2017 US Census of Agriculture and (2) observed sales prices from agricultural land transactions (2013–2017). We find that the difference between these data sources, which we interpret as measurement error, is correlated with multiple determinants of land value. Farmland market characteristics, including market size, are negatively related to this difference, implying that more information in a market allows respondents to better update their priors about their land's value. Survey respondents are also slow to update their assessments in areas of rapid recent price changes. Surprisingly, we find measurement error in farmland value to be correlated with climate. To illustrate the empirical implications, we conduct a Ricardian analysis of agricultural climate change impacts and find that temperature, in particular, is differentially capitalized into self‐reported land value estimates and sales prices. Under certain climate specifications, using self‐reported land values generates climate damage estimates that are more pessimistic but ultimately not significantly different from those using sales prices. However, regional heterogeneity highlights where this difference is significant, namely the northern half of the eastern United States. More generally, our findings highlight important considerations for researchers relying on farmland values from alternative data sources.