I would like to generate an estimate for 2022 of the average number of people per policy (policy holder plus dependents) enrolled in an employer-sponsored health insurance plan, by tier (self, self plus one, family) with standard errors.
It looks like this might be possible with the IPUMS CPS analysis tool.
I tried the following :
|Row|hiunpers|HIU number of persons|
|Column|grptypnw|Type of current employment-based plan|
|Weight|sdawt|Supplement Weight|
|Filter|year(2022)|Survey year|
|Filter|grpcovnw(2)|Currently covered by employment-based insurance(=Yes)|
|Filter|grpownnw(2)|Policyholder for current employment-based insurance(=Yes)|
The results show that around 35% of policyholders with a self only plan have more than 1 person in the HIU, 25% of policyholders in self plus one plans have a HIU of 1 or greater than 2, and 4% of policyholder with family plans have 1 in the HIU.
Does this approach work using these variables and the SDAWT weight ?
Is this likely the closest I can get or is there a better approach using IPUMS CPS or other IPUMS analysis tool?
The variable HIUNPERS is the number of people within each health insurance unit (HIU) as indicated by HIUID. A HIU is a group of individuals is a group of individuals likely to be considered a “family unit” for the purposes of determining eligibility for private or public health insurance coverage. These HIUs are defined by researchers at SHADAC, a research center at the University of Minnesota. SHADAC uses information from the CPS, such as family interrelationships and the employment status of adults, to construct the most likely HIUs within each family. HIU variables that are created by SHADAC, rather than self-reported by respondents, are flagged as such in the variables’ documentation. The procedure used to define HIUs is described in the documentation of HIURULE and this SHADAC report, “Defining Family for Studies of Health Insurance Coverage.”
The variable GRPTYPNW, however, is self-reported (see the 2022 ASEC codebook, specifically item 966 on page 6C-42). Therefore, there will inevitably be discrepancies between attributes of HIUs as reported by respondents in the CPS and as imputed by SHADAC researchers. Either of these sources of information can be sources of errors—individuals can misreport information about their insurance coverage for a number of reasons, and the algorithms used by SHADAC to define HIUs inherently include some amount of error. There can also be discrepancies because individuals included in an HIU may not be part of the household of the policyholder, and therefore would not be captured in the HIU defined by the SHADAC-constructed variables.
An alternative way to estimate the number of people covered by employment-based health insurance that is based on respondents’ answers to CPS questions is to add up the number of “yes” responses to the variable GRPCOVNW for each policy (you can identify individual policy-holders, and therefore distinguish between multiple employment-based policies within a family, using the variable GRPWHO1). I would expect to see less discrepancy between a set of self-reported variables than a combination of imputed and self-reported variables. However, note that there will still be some discrepancy, since there may still be non-household members covered by a given insurance plan. Such cases are reported in the variable GRPOUTNW. Also note that this approach requires that you download a data extract as the online analysis tool doesn’t support this type of data manipulation.
To respond directly to your question about your proposed SDA query, this seems reasonable to me overall. The approach you have outlined will estimate the percentage of policy holders in each insurance tier (self, self plus one, and self plus family) who have HIUs of various estimated sizes, according to the algorithms used by SHADAC. You may want to compare the results you get using HIUNPERS to results you get using GRPCOVNW and GRPWHO1. I will note that your first filter, GRPCOVNW(2) is redundant if you also use the filter GRPOWNNW(2), since all policy holders should be covered by the plans they hold. It is important to keep this second filter (i.e., GRPOWNNW), however, to ensure you are counting just one person from each health insurance policy.