Confirm that high values of vet_ben are pensions from generals #73
Comments
|
@Amy-Xu, See taxdata issue 281 for more discussion of very high values of |
|
@martinholmer Sorry about the delay. Somehow this issue slipped through my hands. I traced back to the code to see which part of CPS data supplies the Look at the raw CPS documentation, it does distinguish the source of pensions in terms of whether it is from Social Security, a military, a state or local government or private company etc. It will definitely be worthwhile to dig into issue more. |
|
I don't see where in C-TAM this discussion of veterans benefits is being
recorded.
What issue is this?
…On Sun, Sep 30, 2018 at 5:45 PM Amy Xu ***@***.***> wrote:
A quick update on the experiment I mentioned above - switching the assumed
base of veteran's benefits. Here is a chart plotting the veteran's benefit
from the raw CPS and the imputed.
[image: screen shot 2018-09-30 at 4 28 38 pm]
<https://user-images.githubusercontent.com/10549597/46262957-76c92180-c4ce-11e8-8b20-c8837bb59efb.png>
The average benefit per percentile for the imputed is a lot lower than the
raw CPS. The raw 2016 CPS has about 3.9 million participants and $58
billion benefits. In this experiment, I imputed the total participants to
21 million and the total benefits to $153 billion that includes
expenditures on Comp & Pension, Education & Vocational Rehab, and Medical
care. Thus, looking at the raw CPS numbers and the imputed average, I'm not
too surprised the average becomes a lot lower.
That being said, I don't think this particular issue is a major source of
the extremely high benefits, because we only impute average benefit values
to the imputed individuals. The maximum of benefits in the raw CPS is
capped at $99,999, and the maximum of imputed benefit is $86,537. So I
think the 140k benefit very likely is a result of tax unit creation.
@MaxGhenis <https://github.com/MaxGhenis> @martinholmer
<https://github.com/martinholmer>
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#73 (comment)>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/ALm1-VbPAtVnTLSkoSTVMcA6gPyYR2Qgks5ugTuWgaJpZM4V3lUZ>
.
|
|
@martinholmer Thank you for the attention. I removed my previous post because I realized there were some details I needed to check with our then contractor, so I don't misspeak about anything. I will re-post once I confirm with her. |
|
I'm still looking into the nuts and bolts on this matter. A quick update: the compensation and pension column in the VA administrative data (as Max linked in the initial post) seems to only include the payments for disability, according to the spreadsheet's note No.2.
In addition, according to this summary (page 4), it seems the retirement pay portion of pension might not be currently considered as a part of Veteran's Benefit, at least not the benefits administered by the Veteran Benefit Administration (VBA). While the CPS data uses a more general term 'veteran payment'. So it could be the case that the retirement pay is included in the CPS data although it is not included in the administrative data of veteran benefit (programs administered by the VBA). I need to dig into this section more. Last note on the taxability of veteran's pension: the portion for disability is not taxable, but the portion for old-age is. |
|
@Amy-Xu, thanks for your update on "veterans benefits" in this comment. The information you've supplied in that comment only reinforces my belief that we should do two things:
My reason for holding this view is that the cash veterans pensions (disability and retirement) are already included in the PUF and CPS data, so including them in the So, I continue to look for an answer to my question: Why was |
|
Thanks @martinholmer.
Could you remind me which variable includes the cash veterans benefit for disability? In 2014 and 2015, this portion of benefit is around $75 billion. In terms of the military retirement pay, if the variable we currently use for the CPS tax unit dataset is Along the same line, the way that CPS constructs their retirement pay variable seems to confirm that the military retirement pay is not a part of VA benefits.
|
|
@Amy-Xu said in C-TAM issue #73:
From the user documentation linked in the Tax-Calculator
So, as you can see, military disability pensions are in |
|
@Amy-Xu said in C-TAM issue #73:
OK. But that brings up again the question that started this issue: Why are some The working hypothesis for months has been that the largest observed values are retirement pensions for multiple-star generals (see the title of this issue). Now after months of speculation along these lines, you're telling us: NO, the large values of |
|
I went to the IPUMS website and downloaded a dataset with
year
incvet
qiincvet
I think incvet is the right variable. Here is the description:
INCVET indicates how much pre-tax income (if any) the respondent
received from
payments from the Veterans' Administration (VA) during the previous
calendar year.
Such payments could include service-related disability compensation,
survivor benefits,
pension, educational allowance, or other veteran payments.
I asked for 2008 through 2018. I can see that for unallocated data the
maximum value of incvet over the 11 year period is $94,000. I think
that there is some mistake in extracting the value of veterans
benefits from the CPS file.
…On 10/16/18, Martin Holmer ***@***.***> wrote:
@Amy-Xu said in C-TAM issue #73:
> the way that CPS constructs their retirement pay variable seems to confirm
> that the military retirement pay is not a part of VA benefits.
OK. But that brings up again the question that started this issue: Why are
some `vet_ben` values so large?
The working hypothesis for months has been that the largest observed values
are retirement pensions for multiple-star generals (see the title of this
issue). Now after months of speculation along these lines, you're telling
us: NO, the large values of `vet_ben` is not caused by large pensions
received by generals. So, then why are some of the `vet_ben` values so
large? See the empirical results in the still-open [taxdata issue
281](PSLmodels/taxdata#281) for the
magnitude of some of the `vet_ben` amounts.
--
You are receiving this because you were mentioned.
Reply to this email directly or view it on GitHub:
#73 (comment)
|
|
@feenberg Thanks for checking the data from the IPUMS website. Correspondingly I went into the 2015 CPS ASEC and checked the veteran benefits there ( During the imputation, these original benefit recipients went through two steps of changes. First, at state level, a universal ratio has been applied to adjust the state level total benefits. Second, a insurance value of medical benefits was added to their benefits. Because this state level variance, these high benefit recipients could end up with different values in the end. @MaxGhenis Max mentioned at the beginning of this issue that there is one recipient who receive about $140k benefit. I tracked this recipient ( Procedure wise, all the steps mentioned are a part of our standard routine (1-impute recipients, 2-assign imputed recipient an average benefit, 3-adjust total benefit). In occasional cases, some already high benefit recipient could get their benefit bumped up to even higher level, but the overall chance of getting benefits at this high is relatively low, because all imputed individuals usually have benefit level much lower since they are assigned the average of their state. That being said, if their benefits deem to be extremely unrealistic, I can add a line to kick them out. |
|
@martinholmer quoted the readme file of the Tax-Calculator
I'm not an expert on this but looking at the 1040 instruction, I'm not sure whether the disability part of pension is included in
I cannot find anything more specific in the PUF documentation either. By any chance @feenberg might have more knowledge on this? |
|
@Amy-Xu, What PUF variable is Form 1040, line 7? |
|
On Wed, 17 Oct 2018, Amy Xu wrote:
@feenberg Thanks for checking the data from the IPUMS website. Correspondingly I went into the 2015 CPS ASEC
and checked the veteran benefits there (vet_val). This particular variable is capped at $99,999 and in 2015
there are 10 individuals who reported benefit at the cap value $99,999. Among these 10 individuals, 5 of
Isn't 99999 the missing value indicator? Still, there are values above
$80,000, which is almost as great. I suppose that could be a disability
pension.
dan
|
|
@martinholmer asked:
1040 line 7 is wage, salaries, tips, etc. I think that's probably somewhat related to e00200. |
|
@Amy-Xu said:
Exactly, so how do you square that information with your earlier assertion that military disability pensions are not taxable. If they're in |
|
@martinholmer Thanks for your response. But if you read my comment again, I have already stated I'm not sure -- the relevant bullet point about disability pension is very vague. It talks about 'your employer', which could possibly not related to the VA. I tried to see whether the PUF documentation has more detailed description but didn't find any. So in the same comment, I tried to seek more explanation from Dan, if he happens to know more about this. In this separate comment, I attached a link to a IRS webpage to support why I think disability compensation is not taxable. In that page, you can find a particular section for the VA disability benefits, which provides a further link to Publication 525. Here's the relevant section in Publication 525:
My impression after reading the section above is simply that the disability portion of VA benefit is not included in the PUF. |
|
@Amy-Xu said two weeks ago in C-TAM issue #73 (with respect to high veterans benefits in the CPS data):
Actually, I now think that all veterans benefits (as opposed to military retirement benefits) are non-taxable under the federal income tax. This view is based on the Wikipedia entry for Veterans Benefits. Here is the first section of that entry:
So, this section makes clear that military pensions (administered by DOD as deferred military compensation) are completely different from veterans benefits, and that veterans benefits are for those who do not qualify for military pensions. It is also clear that veterans benefits can go to the disabled and to the non-disabled elderly:
And the Wikipedia entry says this about the magnitude of veterans benefits:
So, it is clear that annual veterans benefits for an individual can be nowhere close to $100,000 or more. Now we get to the thing that seems to me is wrong with veterans benefits in C-TAM. This issue is entitled "Confirm that high values of vet_ben are pensions from generals." But a full-career general (like the ones @feenberg is imagining) will qualify for a large military pension, but not get any veterans benefits. So, the notion that the high values of The more likely explanation for the high values of
And @feenberg responded with this question:
It would appear that the answer to his question is YES. So, it would seem as if you have misread the CPS code book. Here is what the CPS code book actually says:
So, it would seem that the C-TAM Do you think I've misunderstood something here and that my conclusion that the benefit data in the CPS file used by Tax-Calculator are buggy is incorrect? If not, it would seem that you should assign a very high priority to fixing these bugs in the C-TAM benefit data. |
|
@martinholmer Sorry for the delayed reply. Thank you very much for all the information here. So up to this point it seems we have reached an agreement that the VA benefits are not taxable. Can I interpret your quote from the Wikipedia as that the VA benefit is not a part of the contract with the DoD, thus theoretically repealable? The IPUMS looks like a very neatly put website that provides access to the CPS datasets. However, I'm afraid this is not the source we currently use, since the variable I use for the VA Benefits is called As you can see, missing values are coded separately from all the numerical values. When I was working with the 2015 CPS, the missing values are used to be explicit as 'None or Not in universe'. But somehow in the 2016 CPS, this string doesn't show up anymore. I speculate that is due to a different buffer processing script which turns it to zero, rather than a radical change on markers. @andersonfrailey Could you specify a bit on the source of CPS dataset and the associate processing scripts if there's any? Thanks a lot. With regard to the range of this VA benefit, I have not been able to dig into the documents to estimate the theoretical range. But Martin, Dan mention in this comment that the
I came across the number on monthly benefits before, but wasn't convinced that is the only source of benefit. I need to do more search on this part, and will post again once I find more. |
|
At this point all I can suggest is to contact the CPS people and ask for
an explanation. Or perhaps the VA.
dan
…On Tue, 6 Nov 2018, Amy Xu wrote:
@martinholmer Sorry for the delayed reply. Thank you very much for all the information
here. So up to this point it seems we have reached an agreement that the VA benefits
are not taxable. Can I interpret your quote from the Wikipedia as that the VA benefit
is not a part of the contract with the DoD, thus theoretically repealable?
The IPUMS looks like a very neatly put website that provides access to the CPS
datasets. However, I'm afraid this is not the source we currently use, since the
variable I use for the VA Benefits is called vet_val rather than INCVET. Thus there
could be a discrepancy on the notations for missing values. We have been using the CPS
dataset compiled by the NBER. You can find the data files and documentation at this
NBER page. In terms of the meaning of '99,999' in the dataset, I would argue that the
chance of it being a marker of missing value is low. This is a snippet I found in the
codebook:
screen shot 2018-11-06 at 4 21 57 pm
As you can see, missing values are coded separately from all the numerical values.
When I was working with the 2015 CPS, the missing values are used to be explicit as
'None or Not in universe'. But somehow in the 2016 CPS, this string doesn't show up
anymore. I speculate that is due to a different buffer processing script which turns
it to zero, rather than a radical change on markers. @andersonfrailey Could you
specify a bit on the source of CPS dataset and the associate processing scripts if
there's any? Thanks a lot.
With regard to the range of this VA benefit, I have not been able to dig into the
documents to estimate the theoretical range. But Martin, Dan mention in this comment
that the INCVET variable from IPUMS indicates the largest payment is $94,000. I tried
to plot a histogram with vet_val from the 2016 CPS and here's what I got.
screen shot 2018-11-06 at 4 34 51 pm
I came across the number on monthly benefits before, but wasn't convinced that is the
only source of benefit. I need to do more search on this part, and will post again
once I find more.
cc @feenberg @MaxGhenis
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub, or mute the
thread.[AHvQVQfZoyJyN91vYB1Yl6ZSHEEXm6z3ks5ushYfgaJpZM4V3lUZ.gif]
|
|
@Amy-Xu asked:
Yes. |
|
@Amy-Xu said:
But this IPUMS table seems to suggest they are just two different names for the same variable: |
|
@Amy-Xu said:
Which contradicts your recent assertion that:
His statement and your statement cannot both be true. What you forgot to mention is that Dan @feenberg concluded in that comment the following:
So, he thinks (as I do) that there is a mistake in the (high) |
|
@martinholmer said:
I totally agree. I'm trying to figure out where is the mistake, if there is one. So there're two things I want to verify: 1) Are the IPUMS variables carrying exactly the same values as the original CPS ASEC dataset? 2) Where is the CPS dataset we use coming from? With regard to the first issue, the IPUMS elaborates what they did for the missing values, N.I.U.s, and topcodes at this page (https://cps.ipums.org/cps/inctaxcodes.shtml). It seems they indeed have made minor adjustments in what they called as harmonization process. In particular, they mentioned at the end of the N.I.U. cases section:
Thus I don't think @andersonfrailey Anderson, could you describe the source of the CPS file under the name |
|
@Amy-Xu the CPS file came from NBER and the processing scripts can be found here. In the .DAT file we download from there all of the values are numerical so |
|
@Amy-Xu said in C-TAM issue #73:
But it sounds like IPUMS has corrected the original CPS file, right? I'm not sure why C-TAM is using the NBER version of the CPS when the IPUMS version sounds to be very high quality. Take a look at this What-is-IPUMS page for a description of their international reputation. |
|
@martinholmer said:
I would stick with 'adjusted' rather than 'corrected.'
IPUMS in my perspective is very useful if one wants to do time series analysis across multiple data sources. However, I don't see how NBER's compilation is low quality by any means. In addition, with regard to your quote of Wikipedia, I argue it is possible for a veteran to receive more than $100,000 a year from the VBA. The compensation amount is dependent on disability degree - veterans with higher degree of disability will get more benefits. A veteran with amputation may quality for Special Monthly Compensation (SMC) on top their regular Disability Compensation. Suppose this veteran has a spouse, children and two parents, he or she could potentially receive $3,527.22 from the regular compensation and an additional $5198.96 from SMC per month, which add up to $104,714.16 for a year. This case of course is a special one, but it suggests that getting over $100,000 is possible. Thus, I don't see why the CPS is apparently erroneous. |
|
@Amy-Xu said in C-TAM issue #73:
@Amy-Xu, I know this is complicated, but I'm not sure why you think the SMC benefits are on top of regular VA disability benefits. Your assumption that this is true allows you to argue that people get non-taxable cash VA benefits of more than $100,000 per year. But when I start looking around the Internet, it doesn't seem that the large SMC benefits you mention are in addition to regular VA disability benefits; rather they replace the regular VA disability benefit. First consider this introduction to SMC benefits:
Here's what the VA website says abount the size of the only add-on amount (SMC K):
Here is another example of the relationship between regular VA disability benefits and SMC benefits:
So, unless this website, which is devoted to helping veterans understand the VA benefit rules, is completely wrong, it would seem that your understanding of VA benefits is wrong, and that even a severely disabled veteran will be getting nowhere close to $100,000 per year in non-taxable cash disability benefits. So, your recent comment is not convincing to me. What do others think? |
|
I'm happy to hear more from others. But Martin, if you go into the link for SMC K in the same page you cited, it says the following:
If I understand the 'in addition to' part right, the SMC K can be stacked on top of the regular disability compensation. Of course there's a chance of me confusing myself, cannot comprehend English properly after a whole day reading Foucault. |
|
@Amy-Xu said:
Yes, I agree. That's exactly what I said in this comment a few days ago. But SMC K, which is very small (about $100 per month), is the ONLY SMC benefit that is "stacked on top of the regular disability". So, I'm still highly skeptical of your claim that people can get tax-free cash VA disability benefits of $100,000 per year or more. |
|
I just called the Census Bureau Customer Service asking about the high values ($99,999) in |
|
@Amy-Xu said:
OK, thanks for checking. If the high veterans benefits that are imputed here are correct, then UBI reforms that repeal veterans benefits are going to generate a non-trivial number of people who are substantially worse off. |
|
You two are closer to this issue so I'll let you close it if it's resolved. Thanks for so much research on this! |
|
@Amy-Xu, Do you want to close C-TAM issue #73? |
|
@MaxGhenis @martinholmer Thank you both for all the time spent on this issue. Closing. |
In #68 I showed that
vet_benmaxes out at $140k:@Amy-Xu said that this is likely compensation and pension, per this VA site.
@feenberg then said:
This issue is to pursue this investigation.
The text was updated successfully, but these errors were encountered: