Add Phase 1 data examination results

[martinholmer]

This PR adds the examination/results.md document and provides a link to it in the high-level README.md document.

[donboyd5]

results.md:

• Would it be possible to add the output-file variable names for the two current-law runs:
  • CY2023 Payroll Tax Liability ($ billion) (federal employee plus employer share) -- this looks like payroll tax
  • CY2023 Individual Income Tax Liability ($ billion) (federal individual income tax) -- it looks to me like this is iitax
• Those are some big differences vs. CBO in current law liability (especially IIT), for both taxdata and pe! We will need to dig into that.
• As a result of this, I have added the following variables to the ad hoc analysis: c05800, taxbc, othertaxes, iitax, payrolltax. (See this, at bottom of table: https://boyd-psl-adhoc.netlify.app/analysis.html#comparison-of-weighted-sums-for-selected-variables)
• You can see that my ad hoc run shows, at 2023 levels, for baseline 2023 law (I think), the following.
• For iitax, I get $2,154.3 for taxdata, which about matches your $2,154.4 for taxdata. However, you get $2,012.9 for the pe phase 1 dataset but I get $1,540.4.
• For payrolltax, I match your taxdata number and am somewhat closer for pe than I am with iitax: you have 1696.7 but I have 1630.1 --far too different to be rounding differences.
• This suggests for PE that we're doing something different or using different data or comparing different results. I am using the Feb 20 version of the PE file (see top line here: https://boyd-psl-adhoc.netlify.app/prelims.html).
• Maybe we can discuss when we talk tomorrow.

[donboyd5]

@martinholmer @nikhilwoodruff @MaxGhenis
Perhaps this is the cause of the difference for PE?

• I stack 3 files: (1) pe, (2) taxdata as grown by you to 2023 and then run through tax-calculator, i believe, with all variables, and (3) taxdata as... with only the variables that are in pe
• this means that the variables in pe and in td same-variables-as-pe file that are not in the grown-td all-variables file are missing
• I set missing values to zero. I did this because I think/thought tax-calculator could not handle missing values
• and then run through tax-calculator

I am guessing that as a result I have some important input variables in my stacked file that I set to zero, and they are affecting tax-calculator results, giving results that are not what I intended.

It seems like the fix is for me to do tax calculations in two steps:

• create a stacked file of pe and td same-variables-as-pe and run this through tax calculator
• run the td all-variables file through tax calculator

Then, get the two resulting files and stack them. It will have missing values for the variables that are in td but not in pe, for the pe records and for the td-same-variables-as-pe records. Leave them missing. Calculate comparisons on this file.

I think this should fix it. I had not thought through the implications of setting missing to zero in the stacked file. I'll try to do this now and will report back.

[donboyd5]

That fixes it. The 2023 baseline-law results in the ad hoc analysis now match the results from @martinholmer. The updated ad hoc analysis is here. The revised R code is here.

The results generally are much closer now between pe and taxdata. @nikhilwoodruff and @MaxGhenis I'm sorry for any grief or head-scratching this caused you. Still plenty of questions to investigate, but not the massive differences my erroneous earlier ad hoc analysis gave.

[martinholmer]

@donboyd5 asked in the discussion of PR #17:

Would it be possible to add the output-file variable names for the two current-law runs:

• CY2023 Payroll Tax Liability ($ billion) (federal employee plus employer share) -- this looks like payroll tax
• CY2023 Individual Income Tax Liability ($ billion) (federal individual income tax) -- it looks to me like this is iitax

The "output-file variable names" are at the top of the td23.res-expect and pe23.res-expect files on the second line. I have added a mention of these two files in the data examination methods document.