Effect of Supplemental Nutrition Assistance Program on Food Consumption among Low-Income Workforce in the United States

This research explores the effect of Supplemental Nutrition Assistance Program (SNAP) on weekly food consumption in the United States’ low-income workforce using Regression Discontinuity Design. Findings show a positive relationship between SNAP and food consumption. The study aims to answer research questions such as whether there exists a relationship between the treatment and control variables as well as the interaction effect and if it does, what kind of relationship. The study also examines the effect of SNAP state wise.

Executive Summary
In order to understand the effect of social programs on the target population it is often crucial to
determine the effect of the program on the population before and after administration of the program.
Using a Regression Discontinuity Design, this research explores the effect of Supplemental Nutrition
Assistance Program (SNAP) on weekly food consumption in the United States’ low-income workforce.
The research aims to answer the question as to whether there exists a relationship between the treatment
and control variables as well as the interaction effect and if it does, what kind of relationship
From the results, at 0.05 level of significance, we reject the null hypothesis of no difference between the
means of expenditure on food between the beneficiaries of SNAP and Non-SNAP beneficiaries and
conclude that there is significant relationship between the overall effect of SNAP in the United States and
when using the multiple regression model we realize that there exists a positive relationship where an
increase in $1 million expenditure in the SANP program leads to an increase weekly food consumption
by approximately $1914.36 implying that the SNAP program influences both the taste of food
consumption among recipients of the social program and amount of food consumed from the consumer
stores. However, when analyzing the effect of SNAP state wise, some states food consumption is not
influenced by SNAP participation.

Contents
Executive Summary.........................................................................................................................1
Contents...........................................................................................................................................2
Introduction......................................................................................................................................3
SNAP...........................................................................................................................................4
Eligibility.....................................................................................................................................4
Research Objective......................................................................................................................4
Research questions...................................................................................................................5
Limitations...................................................................................................................................5
Methodology....................................................................................................................................5
Regression Discontinuity Design............................................................................................5
Assumptions of RDD...............................................................................................................6
Instruments..................................................................................................................................7
Data..............................................................................................................................................7
Hypotheses...................................................................................................................................8
Null Hypothesis 1....................................................................................................................8
Alternative Hypothesis............................................................................................................8
Results and Discussion....................................................................................................................9
Results..........................................................................................................................................9
Regression Discontinuity.........................................................................................................9

Discussion......................................................................................................................................11
Conclusion.....................................................................................................................................11
References......................................................................................................................................12
Introduction
Among the priority objectives by the United States Department of Agriculture (USDA) are: increment of
food security, reduction of hunger though enabling more access to food, promoting consumption of
healthy diets, as well as conducting nutrition education among low-income Americans (Ratcliffe,
McKernan, & Zhang, 2011). To realize the department’s objective, some of the programs designed by
USDA include: Supplemental Nutrition Assistance Program (SNAP), Special Supplemental Nutrition
Program for Women, Infants, and Children (WIC); the National School Lunch and School Breakfast
(School Meals) Programs, etcetera (Fraker and Devaney, 2013). Over the recent past, in the analysis of
the performance of the programs for instance when considering the Supplemental Nutrition Assistance
Program during the financial year 2011, the program served approximately over 46 million Americans
with a cost of approximately $75 billion (Senal and West, 2015). In comparison with the WIC and NLSP
programs, the WIC program has enabled approximately 8.2 million families to access healthy foods,
health care, nutrition education and advice (Schmier, 2015) whereas the NLSP in 2014 alone served about
30 million school children with the government allocating approximately $12.7 billion (CBO, 2015).
Today, the SNAP program is implemented through issuance of monthly benefits to beneficiaries i.e. in
the form of Electronic Benefit Transfer (EBT) which are used when making purchases at selected
consumer stores (Carlson et al., 2017).
In this section, an overview of the SNAP program, eligibility, foods included in the program and the
researches objective are explored.

SNAP
According to Hoynes et al. (2014) SNAP is, “…one of the largest cash or near-cash means tested,
universal safety net programs in the United States.” In particular, in 2013 there was an estimated $275 per
benefits for each household per month, that is $133 per person (Schmier, 2015). However, over time the
program’s implementation and benefits have changed a little relatively over time with just the same
framework adopted approximately 50 years ago being in adopted in the program today (Kim, 2015). The
foods included in the SNAP program include: breads and cereals, fruits and vegetables; meats, fish and
poultry, dairy products, seeds and plants that produce food for the household consumption.
Eligibility
Falk (2014) on the study of categorical eligibility of SNAP notes that, for one to be eligible for the food
stamp they must fall into the following criterion, i.e.: their maximum gross income per month should be
130% that of the federal poverty level, their resource count should not exceed $2,250 unless there is a
disabled or elderly person in which case the countable resources should not exceed $3, 500, the
employment status is also a consideration except for the pregnant, elderly and the disabled who are
exempted from the employment requirement. Other participants who contribute to the workforce are
lawfully admitted non-citizens (Rosenbaum, 2013) hence are suitable for the study’s analysis.
Research Objective
The objective of this study is to examine how the Supplemental Nutrition Assistance Program influences
food consumption among the low-income work force in the United States. As such, using historical data,
a regression model is used to examine whether there exists a relationship between the expenditure of
SNAP program and expenditure of food consumption by persons eligible for the SNAP program.
Moreover, the research adopts a Regression Discontinuity Design Methodology (RDD) which upon
proper design, implementation and analysis, provides an unbiased estimate of the local treatment effect
thence it is approximately as good as a randomized experiment when trying to measure a treatment effect.

Research questions
After addressing the research objective, we hope to answer the following questions:
i. Is there a difference between the expenditure of the SNAP work force and the Non-SNAP work
force?
ii. Does SNAP affect the beneficiaries’ food consumption?
Limitations
Our research was prone with a number of limitations which include the credibility of the data
since it was collected from secondary sources. The model used in this study has estimated effects
which are particularly only unbiased as long as the functional form of the relationship between
the treatment and outcome is correctly modelled otherwise the results are prone to biasness.
Moreover, other treatments might contaminate the chosen treatment variable if different
treatments occur on the same cut-off points of the original assignment variables.
Methodology
Regression Discontinuity Design
In order to explore the success of the SNAP program among the labor force in the United States, a RD
design is adopted. In which case, it is viewed as a pretest-posttest design for two different groups i.e. a
control and treatment group. Consequently, since the concern of this paper is to examine the effect of
SNAP on food consumptions, the data is collected is such a way that it will allow for the definition of a
control and treatment group whereby the treatment variables is “number of children and employment
status” and the post-treatment variable is chosen from the sample for the year 2009 and above. The RDD
model can be depicted as in figure 1 where each group is described by a single line:

Where:
C: groups assigned by means of cutoff scores
O: Administration of measure to group
X: Program Implementation
Assumptions of RDD
The main assumption in RDD is the assumption of Regression: E [ Y 0i | X i, D i] = E [ Y 0i | X i]. In that,
after controlling for confounder Xi we assume that the treatment assignment has been completely
randomized.
The post-treatment which is a variable selected from the years greater or equal 2009 will enable us
to determine whether there is an effect on being a participant in the SNAP program on food expenditure
by members of the U.S workforce and labor force.
The regression model for the RDD is therefore:
Where:
C is the treatment cut-off, D is a binary variable equal to one if X>c. Therefore, letting h be the bandwidth
we obtain c-h≤X≤c+h. In the RDD model, various slopes and intercepts can fit the data on either side of
the cutoff.
Figure 1: RDD depiction

In implementing the regression model, we use “Total expenditures on food last week” as the
dependent variable while a treatment group is generated from the sample using the variables
number of children and employment status both of which were equated to zero (treatment = 0,
replace treatment = 1 if nchild ==0&empstat==0). The post-treatment sample is generated from
the years greater or equal 2009 (post = (year>=2009)). Additionally, the interaction variable is
generated through multiplication of the treatment and post variables (interaction = treatment
*post).
Instruments
For this assignment the STATA statistical software will be used for analysis. To perform RDD in STATA
one can use just a normal regression model which in our case will be regressing expenditure against, the
treatment variable, post-treatment variable and the interaction variable together and independently to
examine their relationship with expenditure as well as develop a predictive model. After defining all of
our variables, our multiple regression model will be:
Y= α0 +α1Treatment+α2Interaction+ α3post
Data
Given the research design which requires the use of historical data, secondary data is collected from
https://fred.stlouisfed.org website which was collected by the U.S. Bureau of the Census for use in
Census Small Area Income and Poverty estimates (SAIPE). The sample dataset was cleaned to contain 12
variables i.e.:
i. Year (1961-2017)
ii. Serial
iii. Month
iv. Household weight, Basic Monthly
v. CPSID, household record
vi. State
vii. Total expenditures on food last week
viii. Person number in sample unit

ix. Final Basic Weight
x. CPSID, person record
xi. Number of own children in household
xii. Employment status
The research data has 1,897,845 entries obtained from the 50 States in the US. For further documentation
on the dataset go to data documentation
Hypotheses
Two hypotheses were formulated in order to enable us to address the research question that is:
Null Hypothesis 1
There is no relationship between SNAP and food expenditures
Alternative Hypothesis
There is a relationship between SNAP and food expenditures
Null Hypothesis 2
There is no relationship between the interaction term and expenditure on food
Alternative hypothesis
There is a relationship between the interaction term and expenditure on food
Null Hypothesis 3
There is no relationship between, educational level, number of children, employment status and
participation in SNAP and food expenditure.
Alternative Hypothesis
There is a relationship between, educational level, number of children, employment status and
participation in SNAP and food expenditure.

Results and Discussion
Results
Regression Discontinuity
Regression model
The p-value of SNAP participation on the expenditures on food is 0.000. Therefore, at 0.05 level
of significance, reject the null hypothesis and conclude that there is a relationship between
participation on SNAP and total expenditures on food. Other factors that indicated a significant
relationship with expenditure are: being in the labor force, number of children and the level of
education all of which had a p-value of 0.000 which is less than 0.05. However, only
participation in the SNAP program significantly reduces food expenditure given that its
coefficient is -13.46 implying that, holding all other factors constant, one unit change in the
expenditure of SNAP program reduces the food expenditure by -13.46 units.

Table 1: Regression diagnostics
From table 3, we note that at 95% confidence interval, the p-value for the post-treatment (0.000<0.05)
hence reject the null hypothesis that there is no relationship between the SNAP and food consumption. In
addition, when controlling for the treatment variable, the P-value for the post treatment variable is 0.000
which is less than 0.05. Reject the null hypothesis of no relationship between the response and
explanatory variables interaction, post (control) variables are 0.000 for all the three variables at a
confidence interval of 95%.
Additional analysis to examine whether SNAP eligibility influences food consumption across all the
states indicates a p-value of approximately 0.000 for most of the sates except for states such as New
Mexico, Pennsylvania, Utah, Virginia, Michigan, Louisiana, Arizona, Oregon and Virginia (table 4)
whose p-value are greater than 0.05 hence SNAP program does not influence food expenditure in this
states.

Table 4

Table 5
From table 5, the interaction effect negatively influences the total expenditures on food for the
past week with a p-value of 0.000 at 95% confidence interval.
Discussion
From the results section, the study’s multiple regression model assumes the form: Y= α0
+α1Treatment+α2Interaction+ α3post, when applying the model to the data, the model becomes
Y=1853.896 - 217.1615Treatment + 103.6082Interaction + 174.9092Post. Therefore, from the
model, an increase in $1 million unit of the expenditure on the SNAP leads to a change in food
consumption by 1914.3559 units which is relatively high. One can infer that food expenditure which
greatly influences food consumption is influenced by one’s participation in the SNAP program.

Individually, the interaction effect between the treatment and post treatment variables also has a negative
effect on the food expenditure unlike when regressed alongside the treatment and post-treatment
variables. Individual analysis of the study variables indicate that only SNAP and interaction variables
display a negative relationship with food expenditure with other variables such as treatment, post-
treatment, employment status, and number of children have a positive relationship with food expenditure.
When applying the Neoclassical theory which suggests that, “focuses on how the perception of efficacy
or usefulness of products affects market forces: supply and demand” (Piana, 2014), to social programs
more so in this case Supplemental Nutrition Assistance Program. It is evident that the efficacy of the
SNAP program affects the market forces of demand and supply such that we can assume that the demand
for food given the SNAP program increases due to the ability of the participants to spend less on more
food, especially the fact that SNAP has a negative relationship with food expenditure indicating that an
increase in snap leads to a decrease in food expenditure except for the employed and educated members
of the workforce where falling into any of these categories leads to an increase on food expenditure given
the positive relationship.
Nevertheless, despite the positive contribution of SNAP towards food security, when analyzing the effect
of SNAP on the country’s economic growth, foods included in the SNAP program generally contribute
lower towards the National GDP compared to commodities not include in the social welfare program
(Carlson et al., 2017).
Conclusion
SNAP to a large extent has contributed to food security in the United States as it is evidenced in
this research’s study. Additionally, it also leads to relative social and economic equality given
the fact that it subsidizes the cost of food for persons who can be defined as socially
disadvantaged i.e. the disabled, poor, the elderly etcetera. From our research it is evident that
SNAP affects the consumption of food in the American low-income workforce and hence the

social program is relevant in the United States society. Further research should be considered to
examine the effect of decreasing budget allocation to the SNAP program on the Country’s GDP.
References
Carlson, A., Lino, M., Juan, W., Hanson, K., and Basiotis, P. (2017). Thrifty Food Plan.
Retrieved
from: http://www.cnpp.usda.gov/Publications/FoodPlans/MiscPubs/TFP2006Report.pdf
Fraker, T. & Devaney, B. (2013). The Effect of Food Stamps on Food Expenditures: An
Assessment of Findings from the Nationwide Food Consumption Survey. 71(1), pp. 124-
130. DOI: DOI: 10.2307/1241778
Hoynes, H., McGranahan, L., Schanzenbach, K. & Diane. A. (2014). SNAP and Food
Consumption. Five Decades of Food Stamps, 3(2), pp 2-17.
Kim, K. (2015). Three Essays on the Impact of Government Assistance Programs on Economic
Behaviors of Vulnerable Households. Family and Economic Issues, 30(4), pp.357-371
Piana, V. (2014). Consumer Theory: The Neoclassical Model and Its Opposite Evolutionary
Alternative. Economics, 4(1), pp. 1-7.
Rosenbaum, D. (2013). The Relationship Between SNAP and Work Among Low-Income
Households. Policy Priorities, 1(4).
Ruth, M. & Cheryll, R. (2017). Supplemental Nutrition Assistance Program. Nutrition and Food
Access. Retrieved from:

Senal, W. Reimer, J. & West, T. (2015). How Does the Supplemental Nutrition Assistance
Program Affect the U.S. Economy? Agricultural and Resource Economics Review, 4(3),
pp. 233-252. DOI: /10.1017/S1068280500005049
Ratcliffe, C., McKernan, S. & Zhang, S. (2011). How Much Does the Supplemental Nutrition
Assistance Program Reduce Food Insecurity? Agricultural Economics, 93(4), pp. 1082–
1098
Verick, S. & Islam, I. (2010). The Great Recession of 2008-2009: Causes, Consequences and
Policy Responses. Employment Analysis and Research, 1(4934), pp. 236-251.