The Estimation of Measurement Error in Panel Data *

Abstract

This chapter considers models for the estimation of measurement error on the same population over time. It is useful to distinguish between a causal model, describing the structure of causal relations between “true scores,” and a measurement model, which describes the relationship between measured scores and true scores. Models for empirical phenomena may be evaluated with respect to their formal mathematical properties or with respect to their correspondence with the phenomenon. As an illustration we apply the lag-1 model, assuming homogeneous-error variance, to repeated observations on reported earnings. An increase in the time interval between measurements should lead to more dramatic differences between the models. The distinction between path analysis and path-regression analysis is crucial where one is attempting to compare the parameters of several populations. The distinction is equally important for models characterizing stable processes or where stability is assumed in order to achieve identifiability.