same. We also find that both age and US farm experience have a significant nonlinear effect on
legal status. US farm experience has positive effect on legal status up to thirty-five years. Age has
positive effect on legal status up to eighty years. Education has a significantly positive linear
effect on legal status. We find that the greatest positive marginal effect is from the female dummy
followed by English speaking ability and before 1993 dummy. The greatest negative marginal
effect is from the Hispanic dummy followed by the after 2001 dummy. Note that, holding all other
characteristics the same, the workers interviewed before 1993 are eleven percent more likely and
those interviewed after 2001 are fourteen percent less likely to be legal compared to those
interviewed between these periods.
Finally, Table 2 shows actual-predicted legal status table. A worker is predicted to be
status 0 (unauthorized) if xi 'ɑ < μ0, and is predicted to be status 1 (authorized) worker
if μ0 < xi α < μ1 and so on. Table 2 shows that 80 percent of unauthorized workers are correctly
predicted to be unauthorized. In the same way, 21 percent of authorized workers, 70 percent of
permanent resident and 26 percent of citizens are correctly predicted in their legal status. Our
ordered probit model does a very good job in distinguishing type 0 workers from legal workers,
but many of type 1 workers and type 3 (citizen) workers are mistakenly predicted to be type 2
(permanent resident) workers.
Duration Model with Selection Bias Correction
Here we estimate the duration model with selection bias correction using the results from
the ordered probit legal status model in the first stage. Table 3 shows estimates for parameters and
asymptotic standard errors (given in the parentheses) for farm workers with each legal status.
Status 0 (unauthorized) workers have 33,865 observations, status 1 (authorized) workers have
11