This paper presents estimates of the union wage effect while controlling for
person specific unmeasured variables which do not change over time (‘fixed
effects‘) as well as specification tests for the person effect models commonly in
use. a sample of men from the PSID, we find that controlling for a person effect
POSUH in estimates of the union wage effect on the order of 5-81, as opposed to 201
in the cross section. An omnibus test based on an unrestricted reduced form and an
instrumental variables test based on a differencing idea provide no evidence against
the conventional model. A third test, based on a comparison of those who enter and
leave union coverage, does provide evidence against the usual model.