same direction, but larger. Emerson and Napasintuwong’s result similarly suggested a longer
duration for authorized rather than unauthorized workers. Their result referred to the number of
years working in U.S. agriculture, rather than individual jobs as the above three analyses do.
Their model did not directly address the sample selection issue, as the other three analyses do.
Conclusion
We have proposed and estimated a Heckman-type two stage model with legal status
ordered probit model in the first stage and a duration model in the second stage. This methodology
aims at overcoming two sources of inconsistency of farm work duration study: selection bias and
the censoring problem. Our first methodology deals with the former problem adequately, but it
takes only a rudimentary measure on the second problem: we have used only completed spells.
Our current estimation result is based on this method.
The current estimation has significant coefficients on the selection bias correction term
for all legal status equations except for that of citizen workers. That is, using ordinary least
squares would lead to inconsistent estimates in all equations except for citizen workers. The most
important finding from our estimation is that unauthorized workers do not necessarily have
shorter farm work duration than legal workers. This is supported by two statistics. First, average
predicted farm work duration for unauthorized workers is second longest. Second, the simulation
analysis shows that the duration of the “typical” unauthorized worker will be longer when
working under an authorized or permanent resident status.
15