What can we learn about the effects of food stamps on obesity in the presence of misreporting?

Transcription

1 What can we learn about the effects of food stamps on obesity in the presence of misreporting? Lorenzo Almada Ian McCarthy Rusty Tchernis Columbia University Emory University Georgia State University IZA & NBER Federal Reserve Bank of Dallas November 16, 2015 AMT (Intent vs. Impact) SNAP and Misreporting 1 / 37

2 Background Substantial increase in obesity in recent decades, currently over 20% of adults in every state and 35% in some states (Ogden et al., 2014) Negative effects from obesity on personal health, healthcare expenditures, and wage growth are well documented (Rosin, 2008; Bhattacharya & Bundorf, 2009; Finkelstein et al., 2009; Cawley & Meyerhoefer, 2012) Much higher obesity rates among lower income households (Rosin, 2008) AMT (Intent vs. Impact) SNAP and Misreporting 2 / 37

3 Obesity Maps AMT (Intent vs. Impact) SNAP and Misreporting 3 / 37

4 Obesity Trend AMT (Intent vs. Impact) SNAP and Misreporting 4 / 37

5 SNAP Trend AMT (Intent vs. Impact) SNAP and Misreporting 5 / 37

6 Role of SNAP Supplemental Nutrition Assistance Program (SNAP) is the largest federal nutrition assistance program in the US Increase in expenditures from $23 billion in 2000 (17.2 million recipients) to $74 billion in 2014 (46.5 million recipients) Large body of literature documents positive relationship between SNAP and obesity (Townsend et al., 2001; Gibson, 2003; Chen et al., 2005; Kaushal, 2007; Ver Ploeg & Ralston, 2008; Meyerhoefer & Pylypchuk, 2008; Baum, 2011) Almada and Tchernis (2015) show negative effects of SNAP on adult obesity at the intensive margin of SNAP benefit available per adult AMT (Intent vs. Impact) SNAP and Misreporting 6 / 37

7 Empirical Problem 1 In addition to self-selection, SNAP participation is heavily mismeasured in self-reported survey data (Bollinger & David, 1997; Meyer et al., 2009; Vassilopoulos et al., 2011; Kreider et al., 2012) 2 Are findings of a positive effect of SNAP on adult obesity robust to both self-selection and misreported participation? AMT (Intent vs. Impact) SNAP and Misreporting 7 / 37

8 Is misreporting a problem? 1 Meyer, Mok, and Sullivan (2009) document substantial levels of under-reporting of program participation in survey data compared to administrative data 2 Lewbel (2007) shows that if misreporting is random and participation is exogenous it results in attenuation bias 3 Nguimkeu, Denteh, and Tchernis (2015) model endogenous misreporting, show possible sign reversal of OLS and Lewbel s estimator, expansion bias of IV, and develop a consistent estimators AMT (Intent vs. Impact) SNAP and Misreporting 8 / 37

9 Current Study Revisit role of SNAP on adult obesity with several alternative estimators 1 IV-based point estimates 2 Point estimates with parametric misreporting 3 Nonparametric bounds AMT (Intent vs. Impact) SNAP and Misreporting 9 / 37

10 What is SNAP? Largest federal nutrition assistance program Eligibility based on gross income < 130% of FPL, along with TANF or SSI recipients Benefits distributed differently across states Mandated EBT cards nationally in 2002 AMT (Intent vs. Impact) SNAP and Misreporting 10 / 37

11 Food Stamp Cycle Participants exhaust benefits before end of month (80% have no funds left on EBT after two weeks) Periods of undereating and potential overeating Weight gain for both children and adults Some evidence of this in SNAP (Wilde & Ranney, 2000; Hastings & Washington, 2010) AMT (Intent vs. Impact) SNAP and Misreporting 11 / 37

12 Types of Foods Purchased Poor health effects if SNAP participants purchase more but unhealthy foods Positive health effects if SNAP participants purchase healthier (more expensive) foods Ambiguous effect on obesity AMT (Intent vs. Impact) SNAP and Misreporting 12 / 37

13 Types of Foods Purchased Poor health effects if SNAP participants purchase more but unhealthy foods Positive health effects if SNAP participants purchase healthier (more expensive) foods Ambiguous effect on obesity AMT (Intent vs. Impact) SNAP and Misreporting 12 / 37

14 In-kind Nature of SNAP Benefits If surplus, participants must spend more on food or have benefits go unused Some evidence that SNAP participants spent more on food than with equivalent cash transfer (Devaney & Moffitt, 1991; Fraker et al., 1995) Ambiguous effect on obesity AMT (Intent vs. Impact) SNAP and Misreporting 13 / 37

15 In-kind Nature of SNAP Benefits If surplus, participants must spend more on food or have benefits go unused Some evidence that SNAP participants spent more on food than with equivalent cash transfer (Devaney & Moffitt, 1991; Fraker et al., 1995) Ambiguous effect on obesity AMT (Intent vs. Impact) SNAP and Misreporting 13 / 37

16 Empirical Evidence Ultimately an empirical question Large nutrition and public health literature showing positive effect Taking selection into SNAP more seriously, economics literature still finds positive effect for women (Gibson, 2003; Meyerhoefer & Pylypchuk, 2008; Baum, 2011) Policy briefs describe positive effect of SNAP on obesity for adult women (Ver Ploeg & Ralston, 2008) AMT (Intent vs. Impact) SNAP and Misreporting 14 / 37

17 Self-reported SNAP Participation 12% under-reporting at the household level (Bollinger & David, 1997) 22% under-reporting at the individual level (Marquis & Moore, 2010) Increasing under-reporting over time in the CPS, from 21% in 1991 to 36% in 1999 (Cody & Tuttle, 2002) 35% under-reporting in 2001 ACS, 50% in the CPS (Meyer et al., 2011) AMT (Intent vs. Impact) SNAP and Misreporting 15 / 37

18 Available Methods for Misreporting Mahajan (2006) and Lewbel (2007) achieve point identification of ATE with instrument-like variable that is correlated with the true treatment status but conditionally independent of measurement error Frazis & Loewenstein (2003) develop moment estimator, but under-identified with both endogenous and misreported treatment Brachet (2008), following Hausman et al. (1998), proposes two-stage estimator with parametric misreporting specification Nonparametric bounds (Manski & Pepper, 2000; Kreider & Pepper, 2007; Kreider et al., 2012; Gundersen et al., 2012) AMT (Intent vs. Impact) SNAP and Misreporting 16 / 37

19 Data Sources National Longitudinal Survey of Youth Cohort (NLSY79) for State level SNAP policies from Food Stamp Program Rules Database AMT (Intent vs. Impact) SNAP and Misreporting 17 / 37

20 NLSY-79 SNAP participation in past year (self-reported) Estimated SNAP eligibility (based on FPL thresholds using income and household size) BMI from self-reported height (in 1985 or 1982) and weight Other demographics including number of children, age, race, gender, marital status, employment, and education AMT (Intent vs. Impact) SNAP and Misreporting 18 / 37

21 State SNAP Policies Biometric identification technology (fingerprint scanning) % of benefits issued by direct mail AMT (Intent vs. Impact) SNAP and Misreporting 19 / 37

22 SNAP Eligibility Formally restricted to 130% of FPL Households may be temporarily eligible during the year despite annual incomes > 130% of FPL Focus on 250% of FPL (consider 130% and 185%) AMT (Intent vs. Impact) SNAP and Misreporting 20 / 37

23 Parametric Estimators Fixed Effects Fixed Effects IV Misreporting-adjusted Fixed Effects AMT (Intent vs. Impact) SNAP and Misreporting 21 / 37

24 Misreporting Denote true treatment status by Ti = 1 (z i δ + ν i > 0), and observed treatment status by T i. Then misreporting probabilities are denoted α 0 Pr (T i = 1 T i = 0) α 1 Pr (T i = 0 T i = 1). AMT (Intent vs. Impact) SNAP and Misreporting 22 / 37

25 Misreporting-adjusted Fixed Effects 1 Estimate misreporting probabilities using NLS, including state IVs (Hausman et al., 1998): 1 N {T i α 0 (1 α 0 α 1)F ν(z i δ)} 2, N i=1 where F ν( ) is the cumulative distribution function (CDF) of ν i. 2 Form the predicted true treatment status, ˆP it = F ν(z it ˆδ t), for each year. 3 Estimate a standard linear fixed effects model, y it = γ ˆP it + x it β + φ i + ε it. Identification derives from nonlinearity in F ν( ) and requirement that α 0 + α 1 < 1. AMT (Intent vs. Impact) SNAP and Misreporting 23 / 37

26 Misreporting-adjusted Fixed Effects 1 Estimate misreporting probabilities using NLS, including state IVs (Hausman et al., 1998): 1 N {T i α 0 (1 α 0 α 1)F ν(z i δ)} 2, N i=1 where F ν( ) is the cumulative distribution function (CDF) of ν i. 2 Form the predicted true treatment status, ˆP it = F ν(z it ˆδ t), for each year. 3 Estimate a standard linear fixed effects model, y it = γ ˆP it + x it β + φ i + ε it. Identification derives from nonlinearity in F ν( ) and requirement that α 0 + α 1 < 1. AMT (Intent vs. Impact) SNAP and Misreporting 23 / 37

27 Non-parametric Bounds Produce bounds on the magnitude of the treatment effects under variety of assumptions about the nature of selection and misreporting Monotone treatment selection: P[Y (t) = 1 D = 0] P[Y (t) = 1 D = 1], t {0, 1} Monotone instrumental variable: for u 1 > u > u 2, P[Y (t) = 1 ν = u 1 ] P[Y (t) = 1 ν = u] P[Y (t) = 1 ν = u 2 ] AMT (Intent vs. Impact) SNAP and Misreporting 24 / 37

28 Misreporting Year(s) Full Sample Females Males All Years ˆα *** 0.064*** 0.056*** ˆα *** 0.397*** 0.696*** 1996 ˆα *** 0.108*** 0.079** ˆα *** 0.312*** 0.568** 1998 ˆα *** 0.045*** 0.042** ˆα *** 0.439*** 0.667*** 2000 ˆα ** 0.038** ˆα *** 0.657** 2002 ˆα ** 0.098*** 0.041* ˆα *** 0.719*** 2004 ˆα *** 0.090*** 0.077*** ˆα *** 0.465*** 0.577*** AMT (Intent vs. Impact) SNAP and Misreporting 25 / 37

29 Point Estimates for P(Obese) FE-IV FE Biometric Direct Mail Both IVs MA-FE Full Sample, N = 12, ** (0.010) (0.212) (0.280) (0.170) (0.017) Females, N = 6, (0.012) (0.194) (0.289) (0.162) (0.017) Males, N = 5, *** (0.016) (0.693) (0.689) (0.510) (0.024) AMT (Intent vs. Impact) SNAP and Misreporting 26 / 37

30 Point Estimates for P(Overweight) FE-IV FE Biometric Direct Mail Both IVs MA-FE Full Sample, N = 12, * * * (0.010) (0.245) (0.286) (0.196) (0.019) Females, N = 6, (0.012) (0.206) (0.302) (0.181) (0.017) Males, N = 5, (0.017) (1.009) (0.676) (0.588) (0.022) AMT (Intent vs. Impact) SNAP and Misreporting 27 / 37

31 Bounds for P(Obese) Error Rates Arbitrary No False Misreporting Positives Exogenous Selection 0 [ 0.063, 0.063] [ 0.063, 0.063].05 [ , 0.226] [ , 0.226].1 [ , 0.459] [ , 0.358] Worst-case Selection 0 [ , 0.614] [ , 0.614].05 [ , 0.664] [ , 0.664].1 [ , 0.714] [ , 0.714] Negative Monotone Treatment Selection (MTSn) 0 [ , 0.063] [ , 0.063].05 [ , 0.226] [ , 0.226].1 [ , 0.459] [ , 0.358] Monotone Instrumental Variable with MTSn 0 [ , ] [ , ].05 [ , 0.128] [ , 0.142].1 [ , 0.225] [ , 0.235] AMT (Intent vs. Impact) SNAP and Misreporting 28 / 37

32 Bounds for P(Overweight) Error Rates Arbitrary No False Misreporting Positives Exogenous Selection 0 [ 0.008, 0.008] [ 0.008, 0.008].05 [ , 0.268] [ , 0.088].1 [ , 0.381] [ , 0.152] Worst-case Selection 0 [ , 0.393] [ , 0.393].05 [ , 0.443] [ , 0.443].1 [ , 0.493] [ , 0.493] Negative Monotone Treatment Selection (MTSn) 0 [ , 0.008] [ , 0.008].05 [ , 0.268] [ , 0.088].1 [ , 0.381] [ , 0.152] Monotone Instrumental Variable with MTSn 0 [ , ] [ , ].05 [ , 0.105] [ , 0.030].1 [ , 0.222] [ , 0.078] AMT (Intent vs. Impact) SNAP and Misreporting 29 / 37

33 Summary of Estimated Effects FE FE-IV MA-FE MTSn MIV-MTSn Obese - Full Sample 0.020** [ , 0.358] [ , 0.235] (0.010) (0.170) (0.017) Obese - Females [ , 0.531] [ , 0.339] (0.012) (0.162) (0.017) Obese - Males 0.054*** [ , 0.769] [ , 0.643] (0.016) (0.510) (0.024) Overweight - Full Sample * * [ , 0.152] [ , 0.078] (0.010) (0.196) (0.019) Overweight - Females [ , 0.292] [ , 0.173] (0.012) (0.181) (0.017) Overweight - Males [ , 0.343] [ , 0.251] (0.017) (0.588) (0.022) AMT (Intent vs. Impact) SNAP and Misreporting 30 / 37

34 Sensitivity Alternative dataset with persistent eligibility and balanced panel (Baum, 2011) BMI adjustment Cawley (2004) Alternative eligibility criteria AMT (Intent vs. Impact) SNAP and Misreporting 31 / 37

35 Takeaways 1 Evidence of high rates of misreporting in SNAP, variable over time and across gender 2 Usual IV approach problematic 3 No adjustments or only misreporting adjustment appear better than IV adjustment 4 Findings of positive effect of SNAP on obesity among women do not persist in presence of misreporting 5 Work in Progress: Nguimkeu, Denteh and Tchernis (2015) endogenous misreporting estimator. AMT (Intent vs. Impact) SNAP and Misreporting 32 / 37

36 Simulation Results from NDT AMT (Intent vs. Impact) SNAP and Misreporting 33 / 37

37 Thank You AMT (Intent vs. Impact) SNAP and Misreporting 34 / 37

38 Bibliography I Baum, Charles L The effects of food stamps on obesity. Southern Economic Journal, 77(3), Bhattacharya, J., & Bundorf, M.K The incidence of the healthcare costs of obesity. Journal of health economics, 28(3), Bollinger, Christopher R, & David, Martin H Modeling discrete choice with response error: Food Stamp participation. Journal of the American Statistical Association, 92(439), Brachet, Tanguy Maternal smoking, misclassification, and infant health. Working Paper. University of Pennsylvania. Cawley, John The impact of obesity on wages. Journal of Human Resources, 39(2), Cawley, John, & Meyerhoefer, Chad The medical care costs of obesity: an instrumental variables approach. Journal of health economics, 31(1), Chen, Zhuo, Yen, Steven T, & Eastwood, David B Effects of food stamp participation on body weight and obesity. American Journal of Agricultural Economics, 87(5), Cody, Scott, & Tuttle, Christina The Impact of Income Underreporting in CPS and SIPP on Microsimulation Models and Participating Rates. Washington, DC: Mathematica Policy Research, Inc, July, 24. Devaney, Barbara, & Moffitt, Robert Dietary effects of the food stamp program. American Journal of Agricultural Economics, 73(1), Finkelstein, Eric A, Trogdon, Justin G, Cohen, Joel W, & Dietz, William Annual medical spending attributable to obesity: payer-and service-specific estimates. Health affairs, 28(5), w822 w831. Fraker, Thomas M, Martini, Alberto P, & Ohls, James C The effect of food stamp cashout on food expenditures: An assessment of the findings from four demonstrations. Journal of Human Resources, Frazis, Harley, & Loewenstein, Mark A Estimating linear regressions with mismeasured, possibly endogenous, binary explanatory variables. Journal of Econometrics, 117(1), Gibson, Diane Food stamp program participation is positively related to obesity in low income women. The Journal of nutrition, 133(7), Gundersen, Craig, Kreider, Brent, & Pepper, John The impact of the National School Lunch Program on child health: A nonparametric bounds analysis. Journal of Econometrics, 166(1), AMT (Intent vs. Impact) SNAP and Misreporting 35 / 37

39 Bibliography II Hastings, Justine, & Washington, Ebonya The First of the Month Effect: Consumer Behavior and Store Responses. American Economic Journal: Economic Policy, 2(2), Hausman, Jerry A, Abrevaya, Jason, & Scott-Morton, Fiona M Misclassification of the dependent variable in a discrete-response setting. Journal of Econometrics, 87(2), Kaushal, Neeraj Do food stamps cause obesity?: Evidence from immigrant experience. Journal of Health Economics, 26(5), Kreider, Brent, & Pepper, John V Disability and employment: reevaluating the evidence in light of reporting errors. Journal of the American Statistical Association, 102(478), Kreider, Brent, Pepper, John V, Gundersen, Craig, & Jolliffe, Dean Identifying the effects of SNAP (Food Stamps) on child health outcomes when participation is endogenous and misreported. Journal of the American Statistical Association, 107(499), Lewbel, Arthur Estimation of average treatment effects with misclassification. Econometrica, 75(2), Mahajan, Aprajit Identification and estimation of regression models with misclassification. Econometrica, 74(3), Manski, Charles F, & Pepper, John V Monotone instrumental variables: with an application to the returns to schooling. Econometrica, 68(4), Marquis, Kent H, & Moore, Jeffrey C Measurement errors in SIPP program reports. Survey Methodology, 01. Meyer, Bruce D, Mok, Wallace KC, & Sullivan, James X The under-reporting of transfers in household surveys: its nature and consequences. Working Paper. National Bureau of Economic Research. Meyer, Bruce D, Goerge, Robert M, & Assistance, Food Errors in survey reporting and imputation and their effects on estimates of food stamp program participation. Food Assistance and Nutrition Research Program. Meyerhoefer, Chad D, & Pylypchuk, Yuriy Does participation in the food stamp program increase the prevalence of obesity and health care spending? American Journal of Agricultural Economics, 90(2), Ogden, CL, Carrol, l MD, Kit, BK, & Flega, l KM Prevalence of childhood and adult obesity in the united states, JAMA, 311(8), Rosin, O The economic causes of obesity: a survey. Journal of Economic Surveys, 22(4), AMT (Intent vs. Impact) SNAP and Misreporting 36 / 37

40 Bibliography III Townsend, Marilyn S, Peerson, Janet, Love, Bradley, Achterberg, Cheryl, & Murphy, Suzanne P Food insecurity is positively related to overweight in women. The Journal of nutrition, 131(6), Vassilopoulos, Achilleas, Drichoutis, Andreas, Nayga, Rodolfo, & Lazaridis, Panagiotis Does the Food Stamp Program Really Increase Obesity? The Importance of Accounting for Misclassification Errors. Working Paper. MPRA Paper Ver Ploeg, Michele L, & Ralston, Katherine L Food Stamps and obesity: what do we know? Economic Information Bulletin. Wilde, Parke E, & Ranney, Christine K The Monthly Food Stamp Cycle: Shooping Frequency and Food Intake Decisions in an Endogenous Switching Regression Framework. American Journal of Agricultural Economics, 82(1), AMT (Intent vs. Impact) SNAP and Misreporting 37 / 37