...appeal.

Parsons (1991a) employed aggregate data on the mean delays and success rates at various appeal stages in his critique of Bound's (1989) analysis of the health and earnings of rejected DI applicants. The debate between Parsons and Bound about how many of rejected DI applicants would actually return to work suggests the need for a deeper analysis of the incentive effects of the DI appeals process.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...paper.

The biases resulting from measurement error and endogeneity probably have offsetting effects with measurement error leading to downward bias and endogeneity leading to an upward bias in the coefficients of self-reported health variables. The literature on this subject has conflicting findings. For example, recent papers by Dwyer and Mitchell (1998) and Bound, Schoenbaum and Waidmann (1995) do not find significant evidence of bias in self-reported health measures in the HRS, whereas Kreider (1997) finds evidence of significant endogeneity bias.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...disability.

Workers over age 31 are disability insured if they had 20 quarters of coverage (during the last 40 quarters and are fully insured. That is, they have at least one quarter of coverage for each year elapsed after 1950 (or age 21, if later) and before the year in which he or she attains age 62 or becomes disabled).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...Figure 2.1.

This procedure is also described in the Social Security disability redesign web site.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...satisfied.

The waiting period start date is the later of: (a) the date of disability onset; and (b) the date the applicant first attained disability insured status (see footnote 3). This waiting period is exempted if the applicant had a previous period of disability that ended within 5 years of the latest date of disability onset.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...month.

Once accepted, a SSDI beneficiary receives a benefit based on their computed Primary Insurance Amount, a concave piecewise linear function of their Average Indexed Monthly Earnings (AIME). There is no actuarial reduction in the primary insurance amount: SSDI benefits are the same as the Social Security old age benefits of someone who retires at age 65 with comparable AIME.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...respectively.

The income and asset thresholds for couples are $794 per month and $3,000, respectively. The asset threshold excludes home, auto, household items, burial plots, and life insurance policies of face values under $1,500.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...three.

Additional households and individuals who were not contacted during the wave one interview were added in waves two and three. To simplify our data processing tasks we decided not to include these additional 1,000 individuals in our extract.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...process.

Details of the methods used to construct these nonparametric density estimates, including data dependent bandwidth selection via cross validation, are described in Appendix B.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...stage.

When a respondent reported being denied benefits after initial application and having appealed, the respondent was asked for the date of the last appeal. This date could correspond to a reconsideration or a further stage in the appeal process.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...higher.

The GAO report, discussed in the introduction, suggests that the reason for the high reversal rate at the ALJ stage is due to inadequate documentation of reasons for denials by the DDS.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...probability.

The marginal effect for continuous variables is computed as the average derivative of the estimated choice probability with respect to the variable in question. For binary variables the marginal effect is the average of the difference between the probability with the binary variable set to 1 and the probability with the binary variable set to 0.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...applying.

Peter Diamond has noted an interesting puzzle: why don't more people simultaneously apply for early retirement and DI benefits? Apparently the hassle or ``stigma cost'' of submitting an application for DI benefits exceeds the expected utility of the extra 20% benefit margin and the ability to qualify for Medicare up to one year earlier than otherwise.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...significant.

The only anomalous finding is the estimate of variable 14, the proportion of months worked in the last six, which has a significant positive coefficient. All other things equal, we would have expected individuals who have recently stopped working to be more likely to apply for DI benefits. However, if we drop HLIMPW altogether then this variable is estimated to have negative and significant effect. Hence we conclude that the net overall effect of increased labor supply prior to application is negative.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...model.

See Appendix C for detailed description of the marginal probability model.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...by

Each of the stage-specific probabilities 21#21 were specified as binary logits. See Appendix C for more details.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...al. (1995).

We also estimated the marginal probability model with constants entering at each of the four steps. However, this model yield almost identical results to the results presented in Table 4.3, with the constants in the last three steps being statistically and economically insignificant.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...decisions.

We are currently working on extending Rust and Phelan's (1997) DP model of retirement behavior to include disability as an exit route from the labor force.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...decisions.

New solution algorithms such as Rust's (1997) ``random multigrid algorithm'' have been proven to break the curse of dimensionality in the sense that the computational complexity of these alternative randomized methods increases polynomially rather than exponentially fast in the number of state variables. But the polynomial growth rates are still sufficiently fast that there is a substantial premium on finding parsimonious specifications of the state and control variables entering the DP model.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...variables:

A more detailed explanation of the calculation algorithms is available from the authors upon request.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.