Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

How CBO handles imputing benefits #71

Closed
martinholmer opened this issue May 4, 2018 · 16 comments
Closed

How CBO handles imputing benefits #71

martinholmer opened this issue May 4, 2018 · 16 comments

Comments

@martinholmer
Copy link

martinholmer commented May 4, 2018

There has been recent discussion of how best to impute to filing units a Medicare benefit amount and a Medicaid benefit amount. Most of this discussion has taken place in C-TAM issue #68, C-TAM pull request #70, and taxdata pull request 185. One issue in those discussions has been whether it makes sense to impute different actuarial values (not "insurance values") for Medicare (Medicaid) depending on the Medicare (Medicaid) enrollee's income.

Thinking about how to do the imputations is essentially a benefit/cost analysis of two approaches. There is a simple approach that assigns each enrollee the same actuarial value. Others want to assign different values that vary by income (see the use of 16 income subgroups in taxdata pull request 185). But the focus on income subgroups seems too narrow. For example, the government's cost of providing Medicare seems likely to vary more by age subgroup than by income subgroup. So, a sensible more complicated approach would include more subgroups than just income. Of course, that complicates the imputation process and thus increases the cost of the complicated approach.

A potential cost of the simple approach could be the fact that CBO used something like the complicated approach. I've said in the earlier discussion that as far as I understood, CBO follows the simple approach to assigning a benefit amount to each enrollee. But I didn't supply any documentation to support that view. A couple of minutes of Google searching produced this September 2017 CBO slide presentation by Kevin Perese and Bilal Habib entitled Methodological Improvements for CBO’s Analysis of the Distribution of Household Income.

In that CBO presentation, the authors describe how CPS under-reporting of the receipt of benefits is corrected using a regression approach (see the Medicaid participation discussion on slides 43-48). But then slide 55 says each Medicaid participant is assigned the same benefit amount:

For Medicaid, CBO derives the average cost to the government per recipient from administrative data (by eligibility category). Those averages are then assigned to all recipients (CPS “reported” values are overwritten).

The phrase "by eligibility category" likely refers to subprogram distinctions made in both the CPS and in aggregate administrative data (for example, the actuarial cost of providing CHIP benefits is likely to differ from the actuarial cost of providing Medicaid benefits to those who qualify for Medicare by virtual of being an SSI beneficiary). But it is clear that within each Medicaid subprogram all the enrollees are assigned the same actuarial value as their benefit amount.

And then on slide 66 there is this description of assigning Medicare benefit amounts (emphasis added):

To impute Medicare benefits, CBO makes no change to [CPS] reported recipients. CBO assigns the average cost to the government per participant to all recipients. Benefits from the Low-Income Subsidy for Medicare Prescription Drug Coverage are allocated separately.

Notice again the Medicare subprograms can have different actuarial values, but there is no income or age variation in the imputed Medicare benefit.

In conclusion, it seems to me that the simple approach has a much higher net benefit than does the complicated approach. This is mainly because the complicated approach takes a lot of extra work that CBO is not doing. What do the rest of you think?

@Amy-Xu @feenberg @andersonfrailey @MattHJensen @MaxGhenis

@MattHJensen
Copy link
Contributor

This assessment makes sense to me, but I could be missing something -- I'm interested to know what @feenberg and @Amy-Xu think. And Amy is probably in the best position to make the final call on the cost/benefit tradeoff, given her perspective on the maintenance burdens.

@feenberg
Copy link

feenberg commented May 4, 2018 via email

@Amy-Xu
Copy link
Member

Amy-Xu commented May 7, 2018

@martinholmer Thanks very much for this detailed outline. I agree that we should assign the same actuarial value to everyone for the moment.

@andersonfrailey
Copy link
Collaborator

An issue that @Amy-Xu pointed out with using just a simple average benefit amount, for Medicare at least, is that Medicare provides a lot of benefits for the institutionalized population that we don't have in our datasets. Should we worry at all about maybe trying to back out spending on that subset of the population? @Amy-Xu do you think it's a large enough issue to warrant the extra work?

cc @martinholmer @feenberg

@Amy-Xu
Copy link
Member

Amy-Xu commented May 11, 2018

I imagine institutionalized population might have a different actuarial value, and this part of population has not been included in the CPS tax unit dataset yet. It would be the best using one universal value for everyone if this is a negligible point. But I want to hear more thoughts before we proceed with the universal value.

@martinholmer @feenberg @MattHJensen

@martinholmer
Copy link
Author

I'm confused by this discussion because I was under the impression that the CPS excludes the institutionalized population. Isn't this what this says?

screen shot 2018-05-11 at 11 58 29 am

@Amy-Xu
Copy link
Member

Amy-Xu commented May 11, 2018

@martinholmer You're absolutely right. But is it a problem when CPS doesn't while Medicare, and potentially Medicaid, do?

@martinholmer
Copy link
Author

@Amy-Xu said:

You're absolutely right [that CPS does not include the institutionalized population].
But is it a problem when CPS doesn't while Medicare, and potentially Medicaid, do?

I thought what you said yesterday was that the "scaling factors" applied to the MEPS Medicare and Medicaid amounts were constructed to exclude program costs attributable to the institutionalized population.

@andersonfrailey

@Amy-Xu
Copy link
Member

Amy-Xu commented May 11, 2018

@martinholmer

I interpreted your citation of CBO report

CBO assigns the average cost to the government per participant to all recipients.

as we need to calculate the actuarial value by dividing the total government cost by the total number of enrollee. Is that what you interpreted from the CBO report?

It seems in the current context that "total non-institutional benefit expenditure over total non-instituional enrollee' would be more accurate. What do you think?

Lastly, because the total benefit target is coming from the Medicare Trustee's Report, I assume this assignment of actuarial value has nothing to do with the MEPS imputation (for individual level benefit) anymore. Thus my understanding is that this issue is a separate one from C-TAM PR #70. Do you agree?

@feenberg
Copy link

feenberg commented May 11, 2018 via email

@feenberg
Copy link

feenberg commented May 11, 2018 via email

@Amy-Xu
Copy link
Member

Amy-Xu commented May 11, 2018

@feenberg

Suppose 1% of the population is missed, and 50% of the expenditure is on that 1%. Then
dividing the total expenditure by number of enrollees give the correct
insurance value to each enrollee.

I see your argument, but somehow I feel your argument supports my conclusion rather than yours. Before I explain my logic, I think we need a clarification on the terms we use in this issue.

  • Insurance value: It seems Martin, in his initial post of this issue, interprets as how much it worths to an enrollee. cc @martinholmer (Am I right?)

  • Actuarial value: how much is paid.

I assume you're speaking about actuarial value. (Correct me if I'm wrong)

Then my logic is that non-institutional population shouldn't get assigned the actual whole population average because a huge amount of the expenditure is not on them.

@feenberg
Copy link

feenberg commented May 12, 2018 via email

@Amy-Xu
Copy link
Member

Amy-Xu commented May 13, 2018

@feenberg said

But that is always true of insurance. Most of the ex post benefit goes to
a few. The ex ante benefit is spread evenly. We have chosen to model the
ex ante benefit for good reason.

I see. That makes sense to me. Then it seems we can just proceed with the average calculated as the total benefit expenditure divided by the total number of enrollee.

Is that sensible to you? @martinholmer

cc @andersonfrailey @MaxGhenis

@andersonfrailey
Copy link
Collaborator

Sorry for letting this slip through the cracks. Has this discussion been settled? And if so are we just going with the changes made in TaxData PR #185?

@Amy-Xu @martinholmer @feenberg @MaxGhenis

@Amy-Xu
Copy link
Member

Amy-Xu commented Sep 17, 2018

Since the corresponding TaxData Issue has been merged, I'm closing this issue. Thanks to everyone.

@Amy-Xu Amy-Xu closed this as completed Sep 17, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants