Expand the database to include 2016 CPS

[Amy-Xu]

The benefit database created from the 2016 CPS should be a better fit for the 2016 CPS tax-unit dataset. Currently all the administrative targets are for 2015 CPS, roughly the expenditure in 2014. In this issue I will be updating the progress of finding 2015 administrative targets. Meanwhile, I'll think about how to organize the scripts and targets in a more efficient way. Appreciate any suggestions.

@martinholmer @andersonfrailey @MaxGhenis

[Amy-Xu]

Program	Total Recipients (million)	Total Benefit (billion)	Source
TANF	4.1	26.0 (Admin: 3.2)	Recipients; Benefits
WIC	1.9 women; 1.9 infants; 4.1 children	4.0 food + 1.9 Nutrition Service & Admin	FY 2015 & National level Summary
Veteran's Benefit	21.6	75.8 Comp & Pension + 17.8 Insurance & Construction & Rehab + 65.5 Medical Care + 8.1 Operating Expenses	Expenditure Tables: 2015
SSI	9.2	55.0	Tables 1 -3
SNAP	45.8	69.7	Participation and Costs 1969 - 2017

[Amy-Xu]

Program	Total Recipients (million)	Total Benefit (billion)	Source
Medicaid	70	527.8 + 26.0 admin cost	2016 Actuary Report
Medicare	55.3	638.7 + 8.9 admin cost	2016 Trustee's Report

[Amy-Xu]

Housing is a bit tricky - there doesn't seem existing any benefit totals on the HUD website, so we calculate the targets by summing up the state level statistics from the HUD database. Following the ReadMe, I downloaded the state-level data for 2015, and then used a modified version of this script for the calculations.

Eventually I got 4.5 million families reporting $37.4 billion benefits. Although the script confuses me a little bit, it seems we should use 'Average HUD expenditure per month' rather than 'Average Family Expenditure per month', since the former represents the actual benefit and the latter is simply family spending according to the HUD's description of their data.

Mostly likely CPS has over-reported beneficiaries, just as the case in 2015 CPS. So, we should probably expect to kick out some reported beneficiaries.

[Amy-Xu]

A quick update on the housing imputation for 2016 CPS: the variable we used as the value of housing subsidy fhoussub is not available anymore. So all the imputation will be instead based on the estimate from Special Poverty Measurement (SPM), which is also produced by the Census.

[Amy-Xu]

With regard to the organization of files, I plan to refactor the python imputation script to accommodate different years. Specificifically, I want to have the scripts called and run by a command line that requires input arguments, including year, input file name, output file name. This is a work in progress. I'll open a PR soon to demonstrate in detail.