Imputing Consumption From the Consumer Expenditure Survey to the Panel Study of Income Dynamics

Andreas Lehnert

Description

This research note presents in some detail the data and estimation procedure used by Blundell, Pistaferri, and Preston (BPP) to impute a value of total non-durable expenditure to households in the PSID conditional on observable characteristics and reported expenditures on food at home. Their technique depends on estimating the partial food-demand function, f(c), where f is spending on food and c is total consumption spending. To successfully impute a value of c given a value for f the function f(c) must be monotone.

I test the monotonicity of f(c) by using non-parametric methods; that is, I do not impose any kind of a priori functional form. Use the higher-order differencing techniques of Yatchew to estimate a partially-linear regression (PLR) model of f(c). I find that at high levels of total spending food consumption is non-monotone. However, the affected region is small, likely leading to little bias.

Download:

• 200kb pdf file
• 328kb postscript file
• 32kb LaTeX source file