Abstract

Household-based surveys play an essential role in collecting data on income, employment, health, and demographics. However, a common challenge in survey implementation is nonresponse, which can affect the quality and representativeness of survey results. Nonresponse is typically addressed by randomly selecting additional samples as replacements within the sampling frame. While this approach is operationally practical, this may lead to bias if replacement households differ significantly from the nonresponding ones.
This study examines whether a score-based matching approach, which selects replacements based on similarity in household characteristics, offers any advantage over randomly selecting replacements. Using data from the 2021 Family Income and Expenditure Survey–Labor Force Survey (FIES-LFS) linked with the 2020 Census of Population and Housing (CPH), the study compares the similarity between nonrespondents and their replacements under both methods. Predicted household income is used as a summary score to assess comparability, as it reflects various characteristics of the household and its members.
Results show that score-based matching yields replacements with smaller differences in predicted income compared to random sampling, indicating closer alignment with nonrespondents. However, despite these numerical differences, statistical tests reveal no significant differences between nonrespondents and their
replacements under either method at both national and regional levels. This suggests that both random sampling and score-based matching are acceptable approaches for replacing nonresponding households.