Can anyone help me interpret this report from the Social Security Bulletin on variation in Social Security disability participation? I know I'm in trouble when I'm trying to interpret something that starts out with this sort of thing:
The variance relationship is given in equation (7):If I had been interested in this sort of thing maybe I would have gone to grad school in my undergraduate major, political science (which, at least at the graduate level, is a lot sciencier than you might think). There are reasons people go to law school and some of them have to do with figuring out what you don't want to do.
Var[ln(part)]=Var[ln(disprev)]+Var[ln(partdis)]+2Cov[ln(disprev),ln(partdis)]
I think this report shows that persons of Hispanic origin and those who live in households where English isn't spoken at home are less likely to be on Social Security disability (take that, Donald Trump!) but there's a lot more in this report that I'm struggling to interpret.
4 comments:
They were interested in the fact that the proportion of the population receiving SSDI ("part") varies across the country (they also examined SSI). For each area of the country, in addition to the SSDI participation rate, there is other data which could help in understanding this variation. The main ones are the prevalence of disability ("disprev") and the fraction of the disabled that are on SSDI ("partdis"). Equation (7) follows directly from the fact that part is the product of disprev and partdis, and each of the terms on the right can be calculated from the data [the last term depends on the correlation between ln(disprev) and ln(partdis), but this was not very large]. The first term was 68% for SSDI, and the second only 21%, so most of the geographic variation in SSDI rates is likely caused by differences in disability rates (this is not true for SSI).
The other part of the study tries to determine how much other factors might explain variation in the partdis component, with most factors having only weak effects. The "Hispanic/non-English" factor (this is a weighted average of the original variables, the biggest contributors being those in the "correlated variables" columns of Table 3) was largest, about 10%. The statement you make is likely correct, although all the data can say in that "Areas with higher proportions of people who are Hispanic, speak a language other than English in the home, were not born in the United States, or are noncitizens have lower DI participation" - it doesn't say anything about individuals. This conclusion is not from the results shown in the document, which just indicates how big an effect this factor likely has. The direction of the effect was determined by separate regression of "part" on the Hispanics factor (and so it doesn't account for geographic variation in disability rates, but just looks at the overall correlation, which was negative, so more Hispanics gives lower "part").
I would skip the statistics and read the discussion and other explanations of the results. One thing to keep in mind is that they are trying to determine the causes of variation in certain variables, but this analysis cannot show causation - only relationships among the variables, which is why I qualify statements above about causation.
What 8:55 said
I am acquainted with two of the authors and some of their previous work using SSA data. Their findings support most of our intuitive assessments based on many years observing disability workloads across the country. The work of this team runs the data through algorithms that emulate those those observations statistically. Nothing new, just another way of observing trailing indicators. What’s important about the outcomes is that they support the notion that differences are demographic and not administrative. Duh!
Warning: SSA steers the variables used for reports depending on the story line or policy push. The agency alters the outcomes depending on which variables are omitted, edited, merged, readjusted, and so forth. Tell me the story line and someone will make it so. Yipes!
Post a Comment