Non-parametric preprocessing for the estimation of equivalence scales

Empirically analyzing household behavior usually relies on informal data preprocessing. That is, before an econometric model is estimated, observations are selected in such a way that the resulting subset of data can be assumed to be sufficiently homogeneous with respect to the specific research question pursued. For example, households with members above retirement age may be excluded where it seems important that they differ from other households with respect to time use and home production. We propose the use of matching techniques and balance checking at this initial stage. This can be interpreted as a non-parametric approach to preprocessing data and as a way to formalize informal procedures. To illustrate this, we use German micro-data on household expenditure to estimate equivalence scales as a specific example. Our results show that matching leads to results which are more stable with respect to model specification and that this type of formal preprocessing is especially useful if one is mainly interested in results for specific subgroups, such as low-income households.