20(27)
Table 4. Estimation results, from discrete-time proportional hazards
model (exponentiated coefficients), for the risk of
overcrowding, incorporating both normal and gamma frailty.
Clustered sandwich estimator of standard errors within
parentheses (no. of households 58,168, failures 9693).
Exp. coeff.
(Std. errors)
Exp. coeff.
(Std. errors)
Frailty
Normal
Gamma
Reform variable
T*[d97] (t)
1.060 ***
(0.010)
1.061 ***
(0.010)
Decrease in housing allowance
0.944 ***
(0.006)
0.943 ***
(0.006)
Year dummy, 1997–1999 [d97] (t)
1.054
(0.035)
1.053
(0.035)
Household characteristics
Age (t)
0.956 ***
(0.002)
0.957 ***
(0.002)
Two children (t)
2.251 ***
(0.072)
2.244 ***
(0.067)
More than two children (t)
4.484 ***
(0.281)
4.459 ***
(0.261)
Housing expenses/income, 1996
1.293 ***
(0.029)
1.293 ***
(0.028)
Housing market characteristics
Tobin’s
q
(t)
0.586 ***
(0.026)
0.588 ***
(0.025)
Duration dependence
Time (t)
1.752 ***
(0.110)
1.738 ***
(0.101)
Time^2 (t)
0.926 ***
(0.007)
0.927 ***
(0.007)
Log likelihood
–36,700.2
–36,698.9
Notes
: *** Significant at the 1% level. (t) indicates time-varying variables. The null hypothesis
of zero variance of the unobserved heterogeneity was tested using a likelihood ratio test.
Unobserved heterogeneity, incorporated as normal and gamma frailty, was significant at the 5%
level.
We have a left-censoring problem in our data, i.e., we have no information
on when those recipients who received a housing allowance in January
1994 started to receive a housing allowance, i.e., whether they received it
before or in 1994. To see whether the inclusion of left-censored
observations affects the results, we also made the estimates excluding all
recipients who received a housing allowance in January 1994
(approximately 30%); this did not affect the results.
