288 - 
—------------ 
Table 2 Here 
——----------- 
While there indeed are a number of short spells of 
unemployment, the median is more than 12 weeks. There are a 
fairly substantial number of long spells -- 208 are 
estimated to last more than a year. 
From this information event history models for the rate of 
reemployment can be estimated with variables listed in Table 
1 as co-variates. A fully parameterized model such as the 
Weibull or the Gompertz would seem desirable. These models 
would allow the estimation of a time dependency parameter 
from which the direction of the duration dependence could be 
inferred. However, available software did not permit the 
estimation of the interaction between co-variates and the 
time dependency 
Because the statistical significance of 
such interactions is of interest here, the discrete time 
approximation to the continuous time models proposed by 
Allison (1982) was chosen. This approach also easily allows 
for the incorporation of time varying co-variates. 
Allison shows that a discrete time approximation allows for 
a very flexible formulation of most event history models. 
Using a slight generalization of the framework proposed by 
Allison, general models of the sort: 
p;(t) - f(A; (t)-BX;) 
(1) 
can be estimated with the proper specification of f. Here 
p.(t) is the probability that an event takes place in period 
t for individual i, A, (t) is a vector of constants denoting 
the time dependency for individual i, and BX, is a