This paper develops a model of job search where some information of value
is revealed only after a job starts and where job prospects need not be
identical. Using results from the theory of multi—armed bandits, it is shown
that the optimal sampling strategy consists of an ordering of the job
prospects to be searched. Those jobs with greater 'residual' uncertainty
remaining when the job begins, will, ceteris paribus, be placed higher in the
sampling order and be associated with a lower reservation wage. Thus, if jobs
differed only with respect to this match uncertainty, reservation wages would
increase over an unemployment spell. In general, when search costs and wage
distributions also differ across jobs, the behavior of reservation wages over
an unemployment spell need not be monotonic.

This paper develops a model of occupational matching where, within an
occupation, information at one job may be useful for predicting the match at
other jobs. Recent developments in the theory of superprocesses are used to
derive the optimal sampling policy which predicts that those currently working
their second job within an occupation are less likely to separate from this
job then those working their first job. Also, this difference should increase
with tenure in the previous job since, for those with long tenures, it is more
likely that occupational sorting has taken place. These predictions are tested
using weekly tenure data from the National Longitudinal Survey: Youth Cohort.
Controlling for unobserved heterogeniety and employing semi-parametric
estimation techniques, it is found that one’s previous job tenure
significantly lowers the likelihood of leaving the current job only if both
jobs are of the same occupation. However, overall, occupational switchers are
more likely to leave the current job only if the tenure in the previous job is
greater than one year. Similar results are found for job quitters when the
data is analyzed using a competing risks framework.