FYI: Comparison to Stanford's California Poverty Measure

[MaxGhenis]

The Stanford Center on Poverty & Inequality produces a California Poverty Measure, which does some tasks also done in C-TAM. Here are relevant sections of their methodology, for comparison to C-TAM and in case there's value in reaching out to them. Note that ACS is their base dataset.

SNAP & TANF

We assign eligibility for SNAP at the simulated program unit-level based on income less than 175% of FPL for SNAP and 125% of FPL for TANF. We take self-reported participation as given and randomly assign participation to other income-eligible units within county-race-household size cells to match administrative totals [...] we predict the probability of participation for each eligible non-reporting unit in the ACS. We then order the data based on the predicted probability within the county-race-composition cell, and assign SNAP or TANF participation to enough units until we match administrative totals.

Max note: In a previous version they assigned eligible units randomly within cells to meet targets.

Housing subsidies

We thus impute housing subsidy receipt by first developing a linear regression model predicting subsidy receipt in 3-year California CPS data, Technical Appendices 12 then applying the regression coefficients to the pool of renter household heads in our ACS data. We assign housing subsidy receipt to household heads identified as having the highest probability of subsidy receipt until we match the percentage of renters reporting subsidies in the CPS data. We disallow receipt for households where all individuals are identified as likely unauthorized immigrants. We then estimate the value of the imputed subsidy as equal to the county-specific Fair Market Rent for the housing unit size, less the tenant payment, estimated at 30 percent of household income. The housing subsidy amount plus the tenant payment is capped at the value of the shelter portion of the poverty threshold, following Census SPM procedures.

WIC

We compute eligibility using age of child (age 0-5 in the data, which covers the 12 months prior to the survey month). Women ages 16-44 who meet other criteria are deemed eligible, as are women who have infants (age 0-1 in the data). Income eligibility is defined as family income less than 1.33 times the eligibility ceiling (185 percent of FPL). All those who report Medicaid, or who are foster children, or who are imputed to get SNAP or TANF benefits are also deemed income eligible. We then randomly assign receipt to match administrative totals for women, infants, and children served by county. Months on WIC are also assigned randomly, assuming that a constant share of infants and children will age into and out of eligibility throughout the year, and that a constant proportion of women will become pregnant throughout the year. Monthly benefit amounts are based on Vericker and Zhen (2013).

Misc

They also impute school lunch programs and medical out of pocket expenses, and form tax units from ACS data based partly on strategically maximizing EITC.

@Amy-Xu

[MaxGhenis]

@MattHJensen indicated interest in this on today's call.

[Amy-Xu]

Does this mean

1) we could compare the California portion of data to see how the distributions different under two methodologies?

2) if the differences are significant and sensible, we should expand the model for other states?

What are the plans you have in mind?

@MaxGhenis @MattHJensen

[MaxGhenis]

Columbia's Center on Poverty and Social Policy also does a lot of antipoverty work involving recreating the Supplemental Poverty Measure, and part of this is imputing benefits. In particular, Jane Waldfogel and Christopher Wimer have authored many SPM-related papers, such as an anchored SPM (2013) and an evaluation of 2020 candidates' antipoverty bills for Vox (Stanford's Sara Kimberlin, who also works on CPM, also contributed to this). The anchored SPM paper goes through their imputation procedure.

In terms of your questions, I think it could be worth reaching out to Stanford and/or Columbia to let them know about this project, and see if they have suggestions on what to prioritize, since they seem to be experts in assistance programs. Who knows, they might be interested in sharing their code as a new PSL project, or even working toward a unified poverty model with C-TAM/taxdata in the long run.

I suggested bringing in SPM to taxdata/taxcalc last year (https://github.com/PSLmodels/Tax-Calculator/issues/1896) and it was decided not to be a fit at that time. But I think at some point this would still be a great feature, and the fact that other poverty researchers do it--albeit imperfectly compared to official Census data--suggests there's opportunity to unify around open-source models as the project has done for tax analysis.