The Distribution of Income and Consumption Using Linked Data

• In Event: New Approaches to Measuring Poverty, Well-Being, and Program Enrollment

Abstract

There is a vast literature on the distribution of income, with substantial attention given to low-income families and the role of various government programs in alleviating poverty.  However, high and rising rates of underreporting of income in surveys, particularly among those at the bottom of the reported income distribution, raise substantive concerns about bias in studies relying on such measures. Two approaches have emerged to address income underreporting. One approach involves directly linking survey and administrative data to improve income measurement. A second approach relies on expenditure data, rather than income data, to create measures of consumption. Numerous studies have documented striking differences between consumption- and income-based measures of inequality and poverty (e.g., Slesnick 1993, Meyer and Sullivan 2012, 2023, Aguiar and Bils 2015, Attanasio and Pistaferri 2016). A consistent finding is that consumption poverty is both lower and has declined more rapidly than income poverty. While a likely explanation for this pattern is the large and growing degree of income underreporting in survey data, no study to date has used individual-level linkages to directly quantify the extent to which this underreporting can explain differences in income and consumption poverty.

 

In this paper, we examine the joint distribution of income- and consumption-based measures of well-being in the United States, applying both of the aforementioned approaches to the same individuals within the same dataset. Specifically, we leverage a unique dataset that links data from the Consumer Expenditure (CE) Survey for 2014-2016 to administrative tax records and program participation data from the Social Security Administration (SSA), Department of Housing and Urban Development (HUD), Department of Veterans Affairs (VA), and state SNAP agencies. This is the first study to undertake such an analysis, with the linked data correcting a substantial portion of income misreporting and having important impacts on the analyses.  Indeed, our “blended” income measures, which combine survey and administrative data, compare much more favorably than survey-based aggregates to benchmark totals from national accounts and other sources.

 

We begin by examining the full income and expenditure distributions and find that blended income measures align more closely with expenditure distributions than survey-reported income. We also see less evidence of underreporting with the blended income measures, with 21% fewer individuals having expenditures that exceed their income. The changes are especially pronounced at the very bottom of the distribution, where measurement challenges are greatest and government program benefits play a critical role. Notably, the linked data resolve a longstanding puzzle in prior research: individuals reporting the lowest incomes in surveys seemingly report consumption levels many times higher (Meyer and Sullivan 2011, Brewer, Etheridge, and O’Dea 2017). While families in the bottom 5 percent of the survey income distribution report expenditures exceeding average survey income by a factor of 7, this gap narrows to 2.5 when conditioning on the bottom 5 percent of the blended income distribution. Moreover, blended income deep poverty rates closely mirror consumption-based deep poverty rates, closing almost all (97%) of the existing gap measured using survey-reported income and consumption.

Author

• Derek Wu, University of Virginia
  Presenting Author