Testing for sample selection bias in pseudo panels
Abstract: Sample selection bias is common in economic models based on micro data. Despite the continuous generalization of panel data surveys most countries still collect microeconomic information on economic agentsÂ´ behaviour by means of repeated independent and representative cross-sections. In this paper we develop a simple testing procedure for sample selection bias in pseudo panels. In the context of conditional mean independence panel data models we specify a pseudo panel model in which under convenient expansion of the original specification with a selectivity bias correction term the method allows us using a Wald test of H0: ρ=0 as a test of the null hypothesis of absence of sample selection bias. We show that the proposed selection bias correction term is proportional to the inverse Mills ratio with argument equal to the normit" of a consistent estimation of the observed proportion of individuals in each cohort. This finding can be considered a cohort counterpart of HeckmanÂ´s selectivity bias correction for the individual case and generalizes to some extent previous existing results in the empirical labour literature. Monte Carlo analysis shows the test does not reject the null for fixed T at a 5% significance level in finite samples and increases its power when utilizing cohort size corrections as suggested by Deaton (1985). As a "side effect" our method let us a consistent estimation of the pseudo panel parameters under rejection of the null."
|Date of creation:||01 Mar 2007|
|Date of revision:|
|Contact details of provider:|| |
When requesting a correction, please mention this item's handle: RePEc:col:000130:003558. See general information about how to correct material in RePEc.
For technical questions regarding this item, or to correct its authors, title, abstract, bibliographic or download information, contact: (Ximena Duenas)The email address of this maintainer does not seem to be valid anymore. Please ask Ximena Duenas to update the entry or send us the correct address
If references are entirely missing, you can add them using this form.