Fix AGI concept used in EITC phase-out logic

[martinholmer]

This pull request fixes a bug in EITC phase-out logic. This issue was raised by @MaxGhenis in #1904. Thanks, Max, for pointing this out. Prior to this pull request the EITC amount was phased out using the maximum of earnings and a concept called eitc_agi, which was defined as the sum of AGI (c00100) and non-taxable interest income (e00400). As Max pointed out e00400 is included in the investment income EITC eligibility test, but he couldn't find an mention of it being added to AGI and used in the phase-out formula. After an extensive Google search, I found an IRS document that said in a discussion of the EITC that the AGI concept used was from line 38 of Form 1040, which is standard AGI (c00100). My search also found an online EITC calculator that implied the AGI concept used in EITC calculation was standard (that is, does not include non-taxable interest income).

The results of fixing this bug are quite small. In some years, total refundable credits (in which the EITC is included) changes by $0.1 billion and in other years it is unchanged. When it does change, the aggregate individual income tax total and the combined tax total also changes by $0.1 billion.

It appears that this bug went undetected in our validation efforts because Internet TAXSIM does not have non-taxable interest income as one of its input variables, and because our efforts to validate Tax-Calculator against other tax calculators were never completed.

@MaxGhenis, thanks again for raising and pressing this issue.

A diligent reader of the changes in this pull request might wonder why the aggregate EITC payments changed slightly when using the CPS input data but did not change enough to cause different aggregate EITC payments (rounded to the nearest $0.1 billion) when using the PUF input data. The likely answer to that question is that the distribution of non-taxable interest income in the PUF data is realistic while the distribution in the CPS data is known to be unrealistic. Even before the EITC logic bug was reported, work had begun on trying to impute a more realistic CPS distribution as part of taxdata pull request 165. A more realistic distribution (with most of the non-taxable interest being received by high-income filing units) would probably have produced no aggregate EITC changes as big as $0.1 billion in the CPS data (which is what happened in the PUF data).

[codecov-io]

Codecov Report

Merging #1907 into master will not change coverage.
The diff coverage is n/a.

@@          Coverage Diff           @@
##           master   #1907   +/-   ##
======================================
  Coverage     100%    100%           
======================================
  Files          37      37           
  Lines        3316    3316           
======================================
  Hits         3316    3316

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 13a9a33...26936fc. Read the comment docs.

[MaxGhenis]

Would it be clearer to use c00100 here instead of introducing a new variable name of agi, which might be interpreted as a different concept? functions.py currently only uses the name c00100 to refer to AGI in code.