Measurement error in cassava variety identification: results from comparison of DNA fingerprinting, farmer-reported data, and a phenotypic protocol for cassava varietal identification in Malawi

This paper quantifies the magnitude and implications of the measurement error associated with alternative approaches for collecting data on the adoption of improved crop varieties. Using data from 1,260 cassava-producing households across five major cassava-growing districts in Malawi, we compare farmer self-reported varietal information and a photo-based phenotypic identification protocol against DNA fingerprinting using the DArTseq platform, which serves as the benchmark. The results indicate substantial misidentification across all subjective methods. Only 30% of farmers correctly identified the cassava variety they cultivated, while identification based on photo-assisted phenotypic characteristics was accurate in just 5% of cases. Farmer self-reported data substantially overestimate adoption of improved cassava varieties; DNA evidence suggests that adoption of formally released improved varieties is nearly zero (0.2%) in the study sample. Asking farmers to state the name of the cultivated variety yields more accurate information than asking whether the variety is “improved” or “local”, likely reflecting ambiguity surrounding the classification of “local selected” varieties. We conclude that subjective methods of varietal identification introduce significant measurement error. Given the relatively modest marginal cost of DNA fingerprinting – approximately USD 25 per household – the gains in accuracy strongly justify its inclusion in empirical studies where varietal status is a key variable.