On intra-annual consumption-poverty in the U.S: Response to SNAP and the importance of within-year variation

This research was supported by the Southern Rural Development Center (SRDC) via USDA-ERS Research Innovation and Development Grants in Economics (RIDGE). We thank George Norton, Christopher Parmeter and seminar participants at the Economic Research Service, SRDC and Department of Applied Economics, University of Minnesota, Twin cities for valuable comments. We thank Kristine West for excellent research assistance. All errors and omissions are our own.

Abstract

We use U.S. quarterly consumption data and decomposable poverty indexes to study consumption-based intra-annual poverty and its relationship to participation in the Supplemental Nutrition Assistance Program (SNAP). Intra-annual spells of consumption-poverty account for half of the incidence and one-third of the severity of all consumption-poverty among U.S. households. Households experiencing consumption-poverty for at least one quarter, but not for the whole year, are more likely to self-select into SNAP than the general population but less likely to do so than those who are poor for the year. SNAP participation, in turn, reduces annual and intra-annual poverty.

Keywords:

• Intra-annual poverty
• Transient poverty
• Consumption-poverty
• Food assistance

Notes

This research was supported by the Southern Rural Development Center (SRDC) via USDA-ERS Research Innovation and Development Grants in Economics (RIDGE). We thank George Norton, Christopher Parmeter and seminar participants at the Economic Research Service, SRDC and Department of Applied Economics, University of Minnesota, Twin cities for valuable comments. We thank Kristine West for excellent research assistance. All errors and omissions are our own.

1 A series of U.S. Census Bureau reports on short-term poverty dynamics based on the various SIPP panels can be found at: http://www.census.gov/hhes/www/poverty/data/sipp/.

2 Some of the alternative poverty lines and definitions published by the Census Bureau, and estimates that account for the recommendations of a report by a special National Academy of Sciences panel (Citro & Michael, 1995) may deal with some of the issues underlying the differences between income and consumption measures.

3 In this study “family” and “household” are used interchangeably to refer to CES Consumption Units (CUs). Within each dwelling, CUs are individuals related by blood, marriage, or adoption; single individuals who do not share resources with others; or two or more unrelated individuals who pool incomes to make joint expenditure decisions.

4 We do not make adjustments to the poverty line to account for family size and composition, other than those already imbedded in the official federal poverty line (FPL). The FPL was originally computed separately for each family type (by number of adults, children and elderly). Citro and Michael (1995) compute the marginal adjustments for an additional person implicit in the current FPL and argue that adjusting for family size according to S = (adults + 0.7 kids)^0.7 should be preferred. We performed this alternative adjustment and found no notable differences in our results (estimates available upon request).

5 Sensitivity analyses (available upon request) show that restricting the sample to households with unsuppressed state and full income reporting produces similar results and conclusions as when using the full sample. Poverty statistics and SNAP participation are similar between the empirical analysis sample and the full sample.

6 Excluding health expenditures may lead to understating differences in well-being across households, as higher healthcare expenditures may be a legitimate proxy for well-being. However, higher health expenditures due to negative health shocks are empirically indistinguishable from positive shocks to well-being in this study. We exclude health expenditures to protect against the latter possibility, but our findings are not sensitive to this choice. Poverty rates are slightly smaller but similar when health expenses are not excluded. Multivariate regression results are nearly identical.

7 Two step estimators are also available, but they are less efficient.

8 Beatty and Tuttle (2012) take a similar approach to examine the SNAP effect on food expenditures. Also, Nord and Prell (2011) find that the ARRA benefit change is associated with higher food spending and lower food insecurity among SNAP eligible households, but not marginally ineligible ones.

9 95% CI (±) reported in brackets in this table instead of standard errors, to make the CES figures comparable to Census reports. Standard errors are presented in all other instances in the document.

10 The FGT2 index does not have a natural interpretation, thus, to give a sense of the effect size, the marginal effects associated with the Tobit estimates were first computed as ME = {[E(Y|Y > 0, FS = 1] − [E(Y|Y > 0, FS = 0]}. These are divided by the sample mean severity of poverty to express the program impact as a percent of average poverty.

11 The only notable difference is that for the endogenous dummy Tobit estimates that use two times the FPL as a threshold, the impact of SNAP on annual poverty is negative and statistically significant.

12 Most poverty indices compress continuous household well-being measures, such as income or consumption, into categorical variables or censored, as in the case of the FGT2 index used in this paper. This is based on poverty lines, and there is no widespread agreement on the best method to compute poverty lines. These robustness checks indicate that our results are not driven by choice of a poverty line or by model specification used to measure consumption-poverty.