A model of unemployment duration is estimated with weekly micro
data on a sample of Canadian men during the 1975 through 1980 period.
Entitlement provisions in the unemployment insurance program and demand
conditions are found to have a significant impact on the probability of
leaving unemployment. The probability of a worker leaving unemployment
declines with duration of unemployment, holding unemployment insurance
entitlement constant. When entitlement is allowed to vary, the probability of leaving first falls and then generally rises with unemployment
duration as the declining entitlement induces a greater willingness to
accept offers or search more intensively. These results are robust to
alternative specifications of duration dependence and to allowing for
person—specific unobserved heterogeneity.

The labor force participation behavior of married women, particularly their responses to husbands’ labor
market outcomes and the effects of fertility variables, is modeled using longitudinal data to control for
a rich dynamic structure. Simulation methods provide a feasible approach to overcome the computational
difficulties inherent in classical maximum likelihood estimation of models with non-trivial error structures.
The models are estimated using the method of maximum simulated likelihood (MSL) estimation. The
empirical results imply that women’s participation outcomes are characterised by significant structural state
dependence, unobserved heterogeneity, and serially correlated transitory latent component of error. The
results show that the effect of husbands’ permanent earnings on the participation decision is significantly
stronger than that of current earnings; however, the implied income elasticities of participation are small,
on the order of -0.10. The results also provide strong evidence that fertility variables are not exogenous
to women’s participation decisions. Although MSL estimation is biased for a finite number of simulations,
I provide Monte Carlo evidence that suggests the simulation bias in the estimators is generally not large
relative to the sampling errors, except when there is positive serial correlation and either significant
heterogeneity or state dependence, or when the form of the unobserved heterogeneity is misspecified. In
these cases, the estimated serial correlation and state dependence effects have substantial negative and
positive bias, respectively.

In this paper we present several alternative estimators of the
effect of CETA participation on movements in and out of employment.
Using Social Security earnings records for 1970 to 1979, we construct
employment histories for adult males in the 1976 CETA cohort and control group drawn from the Current Population Study. Our results suggest
that CETA participation had a small to moderately large impact on the
probability of employment. The estimated effects of classroom training
programs are uniformly larger than the effects of non—classroom programs
For both programs, the largest estimates are obtained from a random-
effects specification which expresses the probability of employment as a
function of year effects, previous employment experience and training
effects. We find that a relatively simple first~order Markov model
together with a four grid-point distribution of individual effects is
remarkably consistent with the employment data for both trainees and
controls. The smallest program estimates are obtained from an exact
match procedure which compares post-training employment outcomes of
trainees and controls with identical pre-training histories. The
matched comparisons also highlight some of the difficulties in using
nonexperimental data to evaluate the effectiveness of training.