Publication type
Thesis/Degree/Other Honours
Author
Publication date
June 1, 2001
Abstract:
The impact of response measurement error in duration data is investigated using small parameter asymptotic approximations. The change in the form of duration dependence in the relevant statistical functions is characterized and shown to be closely related with the elasticity properties of the density and survival functions of the error-free duration. The general effects of measurement error in the form of duration dependence are described and illustrated for the generalized Gamma distribution family. The approximations lead to a specification test to detect measurement error which is shown to be related to the class of Information Matrix tests. It is shown that in a commonly used class of models the test statistic is exactly pivotal. The second order asymptotic properties of the alternative forms of the test statistic are derived and the quality of the approximations and the performance of the test are investigated via Monte Carlo experiments. The approximations allow to characterize the inconsistency of M-type estimators suggesting a GMM-type estimator that corrects the bias in the moment conditions. The score for the variance of measurement error under the null hypothesis used to construct the measurement error specification score test is used as an additional moment condition that defines an extended DGP under the null and allows identification of the extended parameter vector under the proposed GMM estimator. This leads to the result that score tests can be 'constructive'. The performance of the estimator is investigated via Monte Carlo experimentation. These methods are applied to the problem of estimation of the parameters of the conditional distribution of time to leaving unemployment from data collected retrospectively from the BHPS survey.
Subject
Notes
not held in Res Lib - bibliographic reference only
#508509