Unit Value Imputation Methods Using Household Scanner Data: A Case Study of Milk Purchases

Abstract We compared three common unit value imputation methods using household purchase data from 2018 to 2020 concerning five milk categories. Regression-based imputation outperformed household mean and retailer mean imputations, based on root mean squared error, mean absolute error, and mean absolute percent error. In a censored QUAIDS model, retailer mean imputation yielded statistically different estimates from the other two methods concerning compensated own-price and cross-price elasticities. We demonstrated that different price imputation methods used in household demand estimation generate different results in predicted prices and estimated price elasticities, and these differences may not necessarily be trivial.