Value of Sample Separation Information in a Sequential Probit Model: Another Look at SSA's Disability Determination Process

We have estimated a 4-step sequential probit model with and without sample separation information to characterize SSA's disability determination process. Under the program provisions, different criteria dictate the outcomes at different steps of the process. We used data on health, activity limitations, demographic traits, and work from 1990 SIPP exact matched to SSA administrative records on disability determinations. Using GHK Monte Carlo simulation technique, our estimation results suggest that the correlations in errors across equations that may arise due to unobserved individual heterogeneity are not statistically significant. In addition, we examined the value of administrative data on the basis for allow/deny determinations at each stage of the process. Following the marginal likelihood approach adopted by Benitez-Silva, Buchinsky, Chan, Rust, and Sheidvasser (1999), we also estimated the above sequential probit model without the sample separation information for the purpose of direct comparison. We found that without the detailed administrative information on outcomes at each stage of the screening process, we could not properly evaluate the importance of a large number of program-relevant survey-based explanatory variables. In terms of both in-sample and jackknife-type out-of-sample predictive analysis, the value of modeling the sequential structure of the determination process in generating correct eligibility probabilities is confirmed.