There is an increasing need for the creation of comprehensive measures of household consumption for the U.S., especially measures that consider how families allocate their time into household production.
While there is access to high-quality data on expenditures (Consumer Expenditure Survey – CE) and time data (American Time Use Survey - ATUS), the availability of joint time and expenditure data is limited.
This motivates the need for the creation of synthetic datasets that combine information from the CE and ATUS. This synthetic dataset could be used to construct comprehensive consumption measures, and to analyze the interaction between expenditure and time spent on household production.
In this paper, we use data from the PSID to evaluate the potential benefits and limitations of three imputation approaches for the creation of this type of synthetic dataset: a) Statistical matching, where individuals are linked to statistically similar observations, b) conditional mean imputation, where imputations are obtained based on the predictions from parametric modeling, and c) stochastic imputation, which aims to account for the error component in the imputation process.
|