We first explain the construction of the 32,124 observations for the individuals' application decision presented in Table 4.1. We assume that each individual makes a decision whether or not to apply once in each wave. If an individual did not apply they are assumed to make their decision at the interview date. If on the other hand an individual did apply, then the decision date is assumed to be the first application date. An individual was assumed to make a decision only if they were not already receiving SSDI or SSI benefits and did not have a pending application. As a result, each individual has a maximum of three, and a minimum of zero application decisions. At each decision date we assigned the appropriate set of income, health and demographic variables to the individual. In this assignment we matched each decision with the variables' values obtained from closest interview information. The only exception is for people that applied right after an interview but reported not being disabled at that interview. In this case we assigned the individual with the variables' values of the subsequent interview, even if it was long after the application.
The 694 observations used in the appeal estimation were constructed in a similar fashion. It is important to note that in this estimation we combined all observations for first, second, and even third, appeals. We did that because the number of individuals for whom we observed second and third appeals is rather small, making it impossible to estimate models for individuals' decision for these appeals.
Constructed variables:
An important issue for the construction of the income and wealth variables is the fact that questions on these variables were only answered by the primary respondent of the household, usually the financially knowledgeable person of the family. Therefore we had to merge this information in order to obtain the relevant values of these variables for the spouses, and to compute total family income for each respondent.
The definitions of the income and wealth variables are as follows:
The definition of the employment history variables:
Imputations:
It is worthwhile to briefly summarize some of the imputations used in constructing the data extract, which was carried out in an attempt to minimize the number of observations which were eliminated from the estimations. Imputations were carried out only for dates of different events connected to the application and appeal process. It was common to find missing months of application, appeal, onset of disability, or the starting month of disability benefits reception. In some cases even the year of the event was missing. In some other instances the dates were not consistent with other information provided in the survey. Our imputations were aimed at avoiding any systematic biases. When other available dates could not provide bounds for our imputations, we simply assigned the number ``6'' for the missing month. When the year was missing we dropped that observation, unless we could unambiguously restore it given the other available information. Although 52% of the observations had some imputations, a number of internal consistency checks using independent information from the employment, disability and income sections of the HRS survey have shown that reported date of disability onset, exit from the labor supply, and receipt of DI benefits match up in predictable fashion. These results (omitted due to space limitations) have been highly encouraging and reinforce our confidence in the quality of the HRS data, the quality of our imputations and the truthfulness of the respondents.