TaxData Dataset Documentation#

TaxData is an open source library used to prepare micro-data for Tax-Calculator, a microsimulation tax model. This document details the process we create the datasets. It contains the following sections:

  1. Summary

  2. PUF

  3. CPS

Summary#

TaxData offers two datasets created to run in Tax-Calculator: PUF.csv and CPS.csv. Both files are representative micro-datasets containing tax unit level data. The key differences are the sources used to create each file and the aggregate results and distributions.

The PUF is created using a statistical match between the IRS-SOI Public Use File and the Census Current Population Survey. The use of IRS-SOI data is restricted and therefore we cannot provide access to them to the general public.

The CPS is created using the Census Current Population Survey, Annual Social and Economic Supplement data for March 2013, 2014, and 2015. Household and family information from the surveys are used to create tax units based on income and relationship data. Because the CPS is not designed for tax reporting purposes, results from constructed tax units are expected to be different from PUF in term of both aggregates and distributions. In addition, the CPS includes very limited data on high income households and has under-coverage issues in lower income households. Therefore, the wage distribution of CPS is different from that of PUF. Additional data on welfare and transfer program participation and benefits is added to the CPS using the open source CPS Transfer Augmentation Model (C-TAM). C-TAM is used to impute benefit values for SNAP, SSI, Social Security, Medicare, Medicaid, veterans benefits, housing assistance, TANF, and WIC.

Variable Availability#

Because benefit data is imputed onto the CPS, it is not available in the PUF. Similarly, the CPS is missing some income and deduction variables only available in the PUF.

A full list of variables available in each file can be found here. The availability of each variable is specified in the availability section. PUF availability is indicated by taxdata_puf, CPS availability by taxdata_cps.

Due to missing variables, certain policy parameters will be ineffective depending on which file you’re using. These will be greyed out in TaxBrain. The policy parameters available in Tax-Calculator can be found in here. The dataset(s) compatible with each parameter are indicated by true/false values in the Has An Effect When Using section.

Aggregate Totals#

The weighted totals of each variable will differ between the two datasets. Totals by year for the CPS can be found here. For the PUF, here.

PUF#

puf.csv (PUF) is created by combining the 2011 IRS-SOI Public Use File and the 2016 Current Population Annual Social and Economic Supplement using a statistical matching technique. The code used to generate puf.csv and its associated files can be found in the various puf_* directories. Because use of the IRS-SOI file is restricted, we cannot share it with the general public.

Input files#

Data Limitations and Clarifications#

As with all datasets, the PUF has some limitations that we are aware of. As we more are discovered we will add them to this section.

Mortgage Interest Expense and Investment Interest#

These two deductions have been folded into a general “interest paid deduction” variable. Unfortunately there is not currently any way to separate them.

Claiming Dependent Filers#

There is no way to link dependent filers in the PUF with the tax unit that is claiming them as a dependent. This should be accounted for when performing tasks such as calculating average income.

State, Local, and Real Estate Taxes#

The PUF includes two variables that encompass some aspect of state and local taxes: e18400 includes either state and local income taxes or state and local general sales taxes. e18500 includes all state, local, and foreign taxes on real estate. There is no way to separate the exact sources of those real estate taxes.

CPS#

cps.csv (CPS) is created by forming tax units out of the 2013, 2014, and 2015 Current Population Annual Social and Economic Supplement. Household and family information from the surveys are used to create tax units based on income and relationship data. Because the CPS is not designed for tax reporting purposes, results from constructed tax units are expected to be different from those in the PUF in terms of both aggregates and distribution. Additional data on imputed benefits receipts are merged onto all three CPS files before the tax units are formed. These imputations are performed by the open source CPS Transfer Augmentation Model (C-TAM).

Current Input Files#

Data Limitations#

As previously stated, the CPS was not intended for tax policy analysis. Thus, we do see less reliable data on income, particularly at the extreme ends of the spectrum.