« Back to Results

Data Quality and Measurement of Earnings, Poverty, and Inequality

Paper Session

Sunday, Jan. 3, 2021 12:15 PM - 2:15 PM (EST)

Hosted By: American Economic Association
  • Chair: Darren Lubotsky, University of Illinois-Chicago

The Poverty Reduction and Targeting of the United States Safety Net

Bruce D. Meyer
,
University of Chicago
Derek Wu
,
University of Chicago
Grace Finley
,
Harvard University
Carla Medalia
,
U.S. Census Bureau

Abstract

This paper analyzes the poverty reduction and targeting effects of taxes and transfers in the United States, using a groundbreaking set of linked survey and administrative data called the Comprehensive Income Dataset (CID). The administrative data consist of IRS tax records and program participation data from various federal and state agencies. These data cover earnings, asset, and retirement income, tax liabilities and credits (including the EITC and Child Tax Credit), and transfer income for a myriad of safety net programs including Social Security, SSI, SNAP, Unemployment Insurance, Veterans’ Benefits, Public Assistance, housing assistance, Medicare, Medicaid, WIC, and energy assistance. We link these data to the Current Population Survey Annual Social and Economic Supplement (CPS), the source of official poverty and inequality statistics and the Survey of Income and Program Participation (SIPP), the most comprehensive survey of income sources in the U.S. Linking the administrative data to the surveys is vital given that a large and rising share of benefits and other income sources is not recorded in the surveys.

To analyze the targeting of government programs, we start by calculating the share of programs dollars paid that go to families below various income cutoffs, focusing on how estimates using the CID compare against estimates using survey data alone. We then assess program targeting on other dimensions by examining the material well-being of families receiving a given program. We examine family characteristics including permanent income and mortality in both the CPS and the SIPP. A wide group of additional measures of material well-being are available in the SIPP including material hardships, appliances owned, and home quality problems. By comparing targeting on a wide range of dimensions of material well-being, we obtain a more comprehensive picture of the families to whom transfer dollars are paid out.

The Good, The Bad and the Ugly: Measurement Error, Nonresponse and Administrative Mismatch in the CPS

Chris Bollinger
,
University of Kentucky
Barry T. Hirsch
,
Georgia State University
Charles M. Hokayem
,
U.S. Census Bureau
James P. Ziliak
,
University of Kentucky

Abstract

Using the March Current Population Survey Annual Social and Economic Survey matched to the Social Security Administration Detailed Earnings Records, we link observations across year to investigate a relationship between item non-response and measurement error for the CPS earnings questions. Prior research has found that (1) non-response is linked to earnings: individuals in the tails of the earnings distribution are less likely to respond to the earnings question. Other research has suggested that (2) individuals with income above the average are more likely to under-report their earnings, while individuals with earnings below average are more likely to over-report their earnings. We examine whether these two phenomenon are related. The overlapping samples in the CPS data allow us to observe individuals who switch from response to non-response. This allows us to investigate whether those who fail to respond in both years have different response patterns than those who provide earnings data in both years.

Addressing SNAP Under-Reporting to Evaluate Poverty

Liana Fox
,
U.S. Census Bureau
Jonathan L. Rothbaum
,
U.S. Census Bureau
Kathryn Shantz
,
Urban Institute

Abstract

The Supplemental Nutrition Assistance Program (SNAP) is a major piece of the social safety net. At peak participation during the Great Recession in 2013, an average of 47.6 million individuals received SNAP benefits each month with an annual cost of $76.1 billion. However, research has found substantial and systematic under-reporting of the receipt of SNAP benefits. This makes it challenging to assess the impact of SNAP receipt on outcomes of policy interest. For example, estimates of the relationship of SNAP receipt and labor force participation or earnings as well as SNAP's impact on material well-being could be biased by SNAP under-reporting. Unfortunately, because SNAP is administered at the state level, comprehensive nationwide administrative program data is not available. We address this issue by using administrative SNAP data from 8 states to impute "true" SNAP participation in the other 42 states and DC. We validate our approach by implementing a "leave-one-out" imputation, where we leave each state with administrative data out of the imputation model separately and compare our imputed SNAP benefits in those states to the actual observed administrative data. We then show how estimates of the Supplemental Poverty Measure (SPM), a measure of poverty that includes resources from SNAP and some other in-kind benefits, are affected by correcting for SNAP under-reporting. To evaluate how the imputed data might affect the kinds of analyses of interest to researchers, we also compare regression results using the imputed values to regressions using the administrative values. For example, we show that survey estimates of the relationship between having earnings and SNAP receipt are biased in the survey data, but we do not find evidence of bias in our imputations.

Partisanship and Survey Refusal

Mark Borgschulte
,
University of Illinois-Urbana-Champaign
Heepyung Cho
,
University of Illinois-Urbana-Champaign
Darren Lubotsky
,
University of Illinois-Chicago

Abstract

Survey refusal in the Current Population Survey (CPS) has tripled over the last decade. This
rise coincides with the emergence of rhetoric, largely from the political right, questioning the
accuracy and integrity of government statistics. We examine how support for the Tea Party and
the Republican party have affected CPS refusal rates and whether households are more likely
to participate in the survey when their preferred political party holds the White House. Using
state and metro vote shares or an individual-level model based on the longitudinal structure
of the CPS, we find no evidence that Republican or Tea Party supporters drive the long-term
upward trend in refusals. We do find evidence of a political cycle in response rates. Refusal
rates since 2015 exhibit polarization, with the fastest growth in refusals among those least likely
to support Trump and the Tea Party. Evidence from an analysis which generates exogenous
variation in Tea Party support using rain on the day of the first Tea Party rally indicates that
exposure to anti-survey rhetoric decreases refusal rates, consistent with the findings from our
other analyses.
Discussant(s)
Martha Stinson
,
U.S. Census Bureau
Dan Black
,
University of Chicago
Moises Yi
,
U.S. Census Bureau
Mel Stephens
,
University of Michigan
JEL Classifications
  • J0 - General
  • I3 - Welfare, Well-Being, and Poverty