Summary

• People in Arizona eat fruit 1.4 times and vegetable 1.6 times in average per day
• Low-income household subgroup eats statistically significant fewer vegetables than non-low-income subgroup
• SNAP does not affect on fruit and vegetable consumption for low-income households
• The effect of SNAP-Ed is marginally significant, however, further research is required
• Marital status has a different effect on fruit and vegetable consumption by gender. Especially single men eat fewest fruit and vegetable.

Introduction

Background

Supplemental Nutrition Assistance Program (SNAP, formerly know as Food Stamp Program) is the largest domestic hunger safety net program in the United States. SNAP has proven power to increase households' purchasing power. However, it is not clear that if the SNAP participants eat nutritious meals that support their health considering only a quarter of Americans follow recommended healthy eating patterns. This research addresses a question whether the additional income from SNAP transfers to healthier eating or not.

This is my thesis research. The research is funded by The Arizona SNAP-Ed Evaluation team working under Arizona Nutrition Network, patnering with ADHS, DES, and USDA. This README simplified the actual work of my thsis and focused only on significant results.

SNAP and SNAP-Ed

SNAP is designed to permit low-income households to access more nutritious and healthier diet through providing additional purchasing power. SNAP-Ed works with partners to provide food and nutrition education to support SNAP’s role in addressing food insecurity. While SNAP eligibility depends on individual's socioeconomic status, SNAP-Ed targets widely at people in low-income neighborhood through classes and projects.

Goals

There are three goals in this research.

(1) Identify fruit and vegetable consumption patterns of people in Arizona

(2) Evaluate the effect of SNAP and SNAP-Ed on fruit and vegetable consumption

(3) Figure out the characteristics of SNAP participants

• Fruit and vegetable consumption is used to measure the healthy eating pattern
• Fruit and vegetable consumption is measured by frequency (how many times you eat ped day?), not amount
• (3) is not directly related with the evaluation, however, the result would be important information to design SNAP-Ed outreach

Data

• Behabioral Risk Factor Surveillance System (BRFSS) - Survey Data collected in Arizona in 2011 and 2013.
• Grocery store information: Census Bureau #CB1300CZ21 - Number of grocery store in each zip code area in 2011 and 2013. Link is 2013.
• DES Office Location - extracted from DES office locator
• SNAP-Ed contractor - AZ SNAP-Ed contractors by county in 2011 and 2013, internal data

Analysis and Results

Analysis Setting

• Observations without SNAP status and control variables are omitted
• Observations with outliers in fruit and vegetable consumption: who answered to eat fruits more than 16 times and vegetables 23 times a day
• Created low-income subgroup based on USDA SNAP eligibility criteria
• All analysis is conducted with sample weighting using SAS

(1) Fruit and Vegetable consumption patterns in Arizona

histogram

ods graphics on;
proc surveymeans plots=none data=snaped3.snapdata mean stderr clm min max;
	STRATA X_STSTR; CLUSTER X_PSU; WEIGHT X_LLCPWT;
	title "F&V: eligible sample";
	var x_frutsum x_vegesum servings;
	run;			
ods graphics off;

The values in table are multipled by 100. ex. 300 = eat fruits 3 times a day

• Fruit

Quartile	Min	Q1	Median	Q3	Max	Mean
Value	0	0.498	1.006	1.965	14	1.42

• Vegetable

Quartile	Min	Q1	Median	Q3	Max	Mean
Value	0	1.033	1.664	2.508	18.07	2.01

• People in Arizona consume fruits 1.42 times and vegetables 2.01 times as daily average

t-test for the difference between low-income and non-low-income subgroups

proc surveyreg data=snaped3.snapdata ;
	STRATA X_STSTR; CLUSTER X_PSU; WEIGHT X_LLCPWT;
	title "T-test x_frutsum: elig=inelig, for whole sample";
	class eligbroad1;
	model x_frutsum=eligbroad1 /solution vadjust=none;
	run;
proc surveyreg data=snaped3.snapdata;
	STRATA X_STSTR; CLUSTER X_PSU; WEIGHT X_LLCPWT;
	title "T-test x_vegesum: elig=inelig, for whole sample";
	class eligbroad1;
	model x_vegesum=eligbroad1 /solution vadjust=none;
	run;	

• There is a significant difference in mean value of vegetable between low-income and non-low-income subgroups

(2) The effect of SNAP and SNAP-Ed on fruit and vegetable consumption

linear regression model (OLS)

(total of fruit and vegetable consumption)

= control (age/education/employment/race/marital status x gender) + SNAP + other variables

/* control variables + SNAP variable */
proc surveyreg data=snaped3.eligp; STRATA X_STSTR; CLUSTER X_PSU; WEIGHT X_LLCPWT;
	model servings=&pdemo &pinteraction fs; title "base model";
	run;
/* control variables + SNAP + # of grocery store */	
proc surveyreg data=snaped3.eligp; STRATA X_STSTR; CLUSTER X_PSU; WEIGHT X_LLCPWT;
	model servings=&pdemo &pinteraction fs allestabli; title "base + # of grocery store";
	run;	
/* control variables + SNAP + # of SNAP-Ed contractors per 125% poverty level population */	
proc surveyreg data=snaped3.eligp; STRATA X_STSTR; CLUSTER X_PSU; WEIGHT X_LLCPWT;
	model servings=&pdemo &pinteraction fs contracte1; title "base + SNAPEd per person";
	run;		

macro for group of variables is shown in code files

Sample is narrowed down to low-income subgroup to avoid bias based on income

Results

• SNAP has no significant effect of fruit and vegetable consumption

(Original intention was to run selection model after finding significance of SNAP, but it didn't happen..)

• SNAP-Ed variable is marginally significant, which suggests SNAP-Ed program is effective for more consumption of fruit and vegetable

Further research with individual level data (such as this research) is needed for more accurate results

• Single men consume fruit and vegetable at the lowest frequency

Compared with married women, separated male (-2.2 times), divorced male (-1.7 times), and never married male (-1.5 times) eat total of fruit and vegetable less time in a day.

• The lower the education is, the lower frequency the person eat fruit and vegetable

Compared with college grad, high school dropout (-1.1 times), high school graduate (-0.7 times), and college dropout (-0.8) eat total of fruit and vegetable less time in a day.

(3) Probit model

probit model

Prob (SNAP=1) = control (age/education/employment/race/marital status x gender) + distance

*Distance is a variable which takes the distance between the closest DES office that authorize SNAP participation and Zipcode of survey respondent

proc surveylogistic data=snaped3.eligp;
	STRATA X_STSTR; CLUSTER X_PSU; WEIGHT X_LLCPWT;
	model fs(event="1.000000000000000") = &pdemo distance / link=probit;
	output out=model23 predicted=p23;
	run;

Results

• The farther the respondent lives from DES office, the more likely s/he participates in SNAP
• The younger s/he is, the more likely s/he participates in SNAP
• The less the education is, the more likely s/he participates in SNAP
• Single men participates in SNAP than married women

Limitation

• Technical problems: how to build machine learning model with sample weighting?
• Data availability: individual level data for SNAP-Ed participation shoud be used for analysis
• Data quality: does frequency of eating measure the healthy eating?