Post on 26-Jun-2015
SNAP and Diet Quality: A Treatment EffectsApproach
Christian A. Gregory* Shelly Ver Ploeg Margaret AndrewsAlisha Coleman-Jensen
Economic Research Service, USDA*contact author: cgregory@ers.usda.gov.
Annual Meetings AAEASeattle, WA
August 13, 2012
The views expressed are those of the authors and shouldnot be attributed to ERS or USDA.
Background & Motivation
Background: Intent of Program
• SNAP authorizing legislation: “To alleviate such hunger andmalnutrition, a supplemental nutrition assistance program isherein authorized which will permit low-income households toobtain a more nutritious diet through normal channels of tradeby increasing purchasing power ...”
• food security and nutrition declared goals of SNAP
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Background & Motivation
Background: Public Perceptions
• “As I look at what this card is paying for in the orders beingscanned at the register, I see T-bone steaks, thick-cut sirloins,thick-cut pork chops (all expensive cuts of meat). I see crablegs, bags of shrimp, and box after box of pastries, cakes anddoughnuts from the bakery department, and bagged candy,chips and cookies from the snack aisles. Then come the sodas,energy drinks and Starbucks coffee drinks... The people usingthis card are eating better than most families that have twoincomes.” -Letter to Frederick News Post
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Background & Motivation
Background: SNAP & Food Security
• recent research: SNAP ⇓ food insecurity
• Yen et al. (2008); DePolt et al. (2009); Shaefer and Gutierrez(2012); Nord and Golla (2009); Nord and Prell (2011);Ratcliffe et al. (2011)
• estimates suggest SNAP participation ⇓ food insecurity 33 -40 percent
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Background & Motivation
Background: SNAP & Diet Quality
• recently–a good deal of concern
• many expensive chronic illnesses associated with low-incomepopulations
• public bears sizable fraction of cost
• policy suggestions:
– restrict foods eligible for SNAP (as in WIC)– Wholesome Wave Double Coupon– Healthy Incentives Pilot
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Background & Motivation
Motivation
• large extant literature (detail below)
• some–improved intakes (Devaney and Moffitt, 1991; Wildeet al., 1999)
• some–poorer intakes (Butler and Raymond, 1996; Yen, 2010)
• difficult to identify treatment effects
selection on unobservables
• selection: adverse or beneficial?
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Background & Motivation
Our Contribution
• use individual data (NHANES) matched to state-level dataidentify SNAP selection
• estimate treatment effects by isolating unobservables in SNAPand diet
• show that marginal effect of SNAP is positive and significantfor some HEI components; adverse selection accounts forworse diet outcomes
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Background & Motivation
Preview of Results
• as measured by HEI total and component scores
– SNAP participants comparable diets– average treatment effect of SNAP (ATE): slightly lower HEIscores
– economically significant?– selection is adverse for many components– effect of SNAP on marginal participant is positive– in particular, SNAP gets participants to consume somewhole fruit and whole grains
• results corroborated by nutrient intakes
• robust to specification choice?
• suggest policy caution: tradeoff improving nutritional quality,changing selection into the program
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Previous Research
Previous Research
• comprehensive review of literature (Fox et al., 2004)
• wrt intakes, few find significant impact ↑, ↓• highlight Gleason et al. (2000)–array of outcomes includingHEI–rule out large effects in either direction
• studies that find positive effects: Wilde et al. (1999);Kramer-LeBlanc et al. (1997); Basiotis et al. (1998)
• more recent studies: Cole and Fox (2008); Yen (2010)
• Waehrer and Deb (2012) used latent factor model/IV–SNAPparticipants ↑ caloric sweetened beverages ↓ fruits/vegetables
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Data
Data: NHANES 2003-08
• individual: NHANES 2003-04, 2005-06, 2007-08
• dependent variable: Healthy Eating Index Score (HEI) (day 1), total andcomponent
– total = sum of 12 elements– total fruit, whole fruit, total veg, dark green and orange veg, total
grains, whole grains, milk, meat and beans, oils, sat fat, sodium,SoFAAS
– for food groups and oils: zero intake = score of zero; meet/exceeddietary recommendation = perfect score; linear interpolation b/w
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Data
Data: NHANES 2003-08 (cont)
• dependent variable: Healthy Eating Index Score (HEI) (day 1), total andcomponent (continued)
• how to score “moderation” components? (i.e. things you should eat lessof)
– 85th pctile of consumption = score of zero; meet Dietary Guidelinesrecommendation = score of 8; meet somewhat higher standard, belowdietary rec = score of 10; linear interpellation b/w amounts at 0 and 8,8 and 10.
– example: sat fat. – fraction of total energy (2001-2002 NHANES data)
• 85th pctile: 15 % : score of 0• DG: less than 10 %: score of 8• below 7% : score of 10
– weights: milk, meat/beans, oils, sat fat, sodium = 10; total fruit, wholefruit, total veg, dark green and orange veg, total grains, whole grains=5 ; SoFAAS = 20
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Data
Data: NHANES 2003-08
• independent variable of interest: HH SNAP participation
– 2003, 2005 waves: 2 questions HH SNAP participation: number ofpersons authorized to receive SNAP, whether HH receive SNAP 12 mos.
– 2007 wave: HH receive SNAP 12 mos– we use whether HH receive SNAP 12 mos 2003, 2005, 2007– robustness check: sample person currently receiving SNAP
• other rhs variables: race/ethnicity, income, education, SR weight 1 yearago, age, marital status, employment status, vigorous ex./week, nutritioned per poor person, hh size, state fixed-effects
• 200% FPL
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Data
Data: SNAP Policy Database
• in model (following) we need exogenous variables to identifyparticipation in SNAP
– state-month level variation in three policies:– expanded categorical eligibility–relaxed asset and/orincome requirements
– biometric info needed to enroll–usually a fingerprint– certification period–median certification period forhouseholds with earnings calculated from the QC data
• valid: the policies affect SNAP participation but not dietquality/HEI
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Methods
Selection Model
• one might begin with
HEIi = Xiβ + SNAPiδOLS + ϵi (1)
• problem: SNAP is endogenous to HEI• another way to proceed
HEIi = Xiβ + SNAPiδZ + ϵi (2)
SNAP∗i = Ziγ + Xiθ + υi (3)
• Z exogenous variables for SNAP• SNAP∗ latent index of SNAP participation• X other variables correlated w/ SNAP, HEI• ϵ and υ bivariate normal w/covariance matrix
V =
[σ2 ρσρσ 1
]Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Methods
Identification & Marginal Effects
• model is theoretically identified by functional form imposed bydistribution of ϵ and υ.
• we use exogenous policy variables to identify SNAPparticipation
• ATE of SNAP :
µi = δZ + ρσ
[ϕ(Ziγ + Xiθ)
Φ(Ziγ + Xiθ) ∗ [1− Φ(Ziγ + Xiθ)]
](4)
this is what δOLS will estimate
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Methods
Identification & Marginal Effects
• without selection: µi = δOLS ; with selection δZ + difference inexpected value of errors conditional on participation (SeeGreene, 2011)
• unconditional on selection, δZ measures marginal affects ofSNAP on participants
• standard errors (of total effects) (ν) by delta method: letα = [γ, θ]
νµ =
√∂µ
∂αM
∂µ
∂α
′, (5)
where M is the covariance matrix of the selection equation
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Descriptive
51.8
47.8
4950
5152
53H
EI S
core
No SNAP SNAP Participants
Data: NHANES, 2003−08
SNAP Participation Status
HEI Score and SNAP Participation
Figure : Differences in HEI over SNAP Participation
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Descriptive
2094
2124.3
2044
2074
2104
2134
To
tal E
ner
gy
Inta
ke
No SNAP SNAP Participants
Data: NHANES, 2003−08
SNAP Participation Status
Total Food Energy and SNAP Participation
Figure : Differences in Energy over SNAP Participation
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Descriptive
Table : Means of HEI Components by SNAP Participation
HEI Component No SNAP SNAP Difference
TotalFruit 2.11 1.73 -0.38***(0.07) (0.07) (0.12)
WholeFruit 1.93 1.39 -0.54***(0.06) (0.06) (0.10)
TotalVeg 3.00 2.63 -0.37***(0.04) (0.07) (0.08)
DkGOrVeg 1.17 0.83 -0.34***(0.05) (0.05) (0.08)
TotGrain 4.27 4.07 -0.20***(0.03) (0.04) (0.06)
WholeGrain 0.93 0.66 -0.27***(0.04) (0.03) (0.05)
N 5,105
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Descriptive
Table : Means of HEI Components by SNAP Participation, cont’d
HEI Component No SNAP SNAP Difference
Milk 4.77 4.39 -0.38**(0.09) (0.11) (0.15)
Sodium 4.12 4.52 0.40***(0.07) (0.09) (0.11)
SoFAAS 9.47 7.96 -1.51***(0.20) (0.25) (0.41)
N 5,105
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
ATE of SNAP
Table : ATE of SNAP on HEI/Components: 200% FPL
HEI TotalFruit WholeFruit TotalVeg DkGOrVeg
µ -1.241*** -0.144*** -0.520*** -0.069*** -0.103***νµ (0.049) (0.016) (0.082) (0.009) (0.005)
TotGrain WholeGrain Milk MeatBeans Oils
µ -0.094*** -0.307*** 0.004 -0.340*** 0.039**νµ (0.005) (0.078) (0.004) (0.000) (0.017)
SatFat Sodium SoFAAS
µ 0.0290*** 0.376*** -0.388***νµ (0.009) (0.001) (0.039)N 5,105
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Correlation, IV Strength
Table : Selection Paramter: ρ
HEI TotalFruit WholeFruit TotalVeg DkGOrVeg
ρ 0.082 -0.107 -0.648*** 0.071 0.040νρ (0.169) (0.223) (0.203) (0.129) (0.301)
TotGrain WholeGrain Milk MeatBeans Oils
ρ -0.059 -1.032*** -0.017 -0.000 0.066νρ (0.048) (0.069) (0.096) (0.084) (0.106)
SatFat Sodium SoFAAS
ρ -0.035 0.003 0.082νρ (0.127) (0.117) (0.169)
• All F-tests of instruments > 15.
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Marginal Effects of SNAP
Table : Marginal Effects of SNAP=δZ
HEI TotalFruit WholeFruit TotalVeg DkGOrVeg
δ -1.429 0.270 1.981*** -0.301 -0.236νδ (1.916) (0.757) (0.624) (0.382) (0.870)
TotGrain WholeGrain Milk MeatBeans Oils
δ 0.041 1.940*** 0.116 -0.338 -0.425νδ (0.133) (0.095) (0.598) (0.392) (0.697)
SatFat Sodium SoFAAS
δ 0.273 0.357 -1.429νδ (0.908) (0.670) (1.916)
N 5,105
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Questions
• δs seem too large to be believed
• δwf = 1.98, x̄ = 1.39
• δwg = 1.94, x̄ = .66
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Distribution of Components
0.1
.2.3
.4.5
Den
sity
0 1 2 3 4 5Score
Data: NHANES 2003−08, 200% FPL Kernel Density WholeFruit Component Score
0.5
11.
5D
ensi
ty
0 1 2 3 4 5Score
Data: NHANES 2003−08, 200% FPL Kernel Density WholeGrain Component Score
Figure : Distribution of Whole Fruit, Whole Grain Components
• modewf = 0, modewg = 0
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Distributional Concerns
• need to address the violation of distributional assumptions
• GMM, 2SLS, larger std errs, size of δZ still a concern
• finite mixture model (latent class model) – probabilities asfunction of SNAP participation (in process)
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Solution: Bivariate Probit
Table : Bivariate Probit: Effect of SNAP on Score >0
Whole Fruit Whole Grain
Parameter Marginal Effect Parameter Marginal Effect
SNAP 0.672** 0.409 .699*** 0.409(0.29) (0.22)
N 5,105
• effect on SNAP is to increase by 40 percentage points pointsprob of eating any whole fruit or whole grains
• too large? less than 30% of sample eat any whole fruit orwhole grain
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Total Effects: Current Recipients
Table : Total Effects of SNAP (Current) on HEI/Component Scores
HEI TotalFruit WholeFruit TotalVeg DkGOrVeg
µ -2.371*** -0.301*** -0.570*** -0.059*** -0.019νµ (0.601) (0.093) (0.137) (0.013) (0.017)
TotGrain WholeGrain Milk MeatBeans Oils
µ -0.089*** -0.357*** 0.0570*** -0.352*** -0.076***νµ (0.007) (0.102) (0.004) (0.019) (0.005)
SatFat Sodium SoFAAS
µ 0.179*** 0.337*** -0.712***νµ (0.007) (0.028) (0.139)
N 5,105
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Marginal Effects: Current Recipients
Table : Marginal Effect of SNAP (Current) = δZ
HEI TotalFruit WholeFruit TotalVeg DkGOrVeg
δ 5.245 0.897 2.981*** -0.690 -0.674***νdelta (11.316) (1.102) (0.200) (0.514) (0.180)
TotGrain WholeGrain Milk MeatBeans Oils
δ 0.053 1.984*** 0.554 -0.264 -0.277νdelta (0.158) (0.073) (0.614) (0.302) (0.934)
SatFat Sodium SoFAAS
δ 0.108 -0.313 0.203νdelta (0.951) (0.542) (2.326)
N 5,105
• similar marginal effects of SNAP on score > 0.
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Results
Robustness: Nutrient Intake
Table : Total Effects of SNAP on Nutrient Intake
Energy (Kcal) Protein Total Fat Sat Fat Carbs
µ -19.78*** -0.047*** -1.810*** -0.221*** 0.711***νµ (1.87) (0.02) (0.31) (0.05) (0.129)
Vitamin C Niacin Folate Sodium Frac FAFH
µ 8.220*** 0.166*** -0.063*** -0.208*** -0.029***νµ (0.08) (0.02) (0.01) (0.00) (0.00)
N 5,105
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Discussion
Discussion
• Results
– SNAP participants slightly lower HEI scores thancomparable non-participants
– ATE statistically significant, though not economically so– ATE for current recipients somewhat larger–same directions– corroborated by nutrient intake results– however: adverse selection into SNAP– SNAP has positive effect on whole fruit and whole grainconsumption of SNAP participants ⇑ in P(Score) > 0.
– but participants in general have slightly less healthy dietscompared to similar non-participants
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Discussion
Discussion
• Further Questions
– controlled for endogeneity fully?– distribution of error terms–alternative distributions– how might SNAP improve DQ w/o adversely affectingselection/effectiveness?
– subsidies instead of restrictions? (Wholesome Wave, HealthyIncentives)
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012
Discussion
Further Discussion?
Thank You
Gregory, Ver Ploeg, et al. SNAP and Diet Quality Aug. 13, 2012