48167_err167

United States Department of Agriculture
Economic
Research
Service
Prevalence of U.S. Food Insecurity Is
Economic
Research
Related to Changes in Unemployment,
Report
Number 167 Inflation, and the Price of Food
June 2014
Mark Nord, Alisha Coleman-Jensen, and Christian Gregory

Economic Research Service

www.ers.usda.gov
Access this report online:

www.ers.usda.gov/publications/err-economic-research-report/err167
Download the charts contained in this report:

• Go to the report’s index page www.ers.usda.gov/publications/
err-economic-research-report/err167
• Click on the bulleted item “Download err167.zip”
• Open the chart you want, then save it to your computer
Recommended citation format for this publication:

Nord, Mark, Alisha Coleman-Jensen and Christian Gregory. Prevalence of U.S. Food
Insecurity Is Related to Changes in Unemployment, Inflation, and the Price of Food,
ERR-167, U.S. Department of Agriculture, Economic Research Service, June 2014.
Cover image: Thinkstock.
Use of commercial and trade names does not imply approval or constitute endorsement by USDA.
The U.S. Department of Agriculture (USDA) prohibits discrimination in all its programs and activities on
the basis of race, color, national origin, age, disability, and, where applicable, sex, marital status, familial
status, parental status, religion, sexual orientation, genetic information, political beliefs, reprisal, or because
all or a part of an individual’s income is derived from any public assistance program. (Not all prohibited
bases apply to all programs.) Persons with disabilities who require alternative means for communication of
program information (Braille, large print, audiotape, etc.) should contact USDA’s TARGET Center at (202)
720-2600 (voice and TDD).
To file a complaint of discrimination write to USDA, Director, Office of Civil Rights, 1400 Independence
Avenue, S.W., Washington, D.C. 20250-9410 or call (800) 795-3272 (voice) or (202) 720-6382 (TDD). USDA
is an equal opportunity provider and employer.
Economic
Research Prevalence of U.S. Food Insecurity Is
Service
Economic
Research
Inflation, and the Price of Food
Report
Number 167
June 2014
Abstract
This report examines the extent to which year-to-year changes in the prevalence of U.S. house-
hold food insecurity can be explained by changes in the national unemployment rate, inflation,
and the price of food relative to other goods and services. Data are from the 2001-12 Current
Population Survey Food Security Supplements and statistics on employment and prices from
the U.S. Bureau of Labor Statistics. Understanding the strength and consistency of these asso-
ciations can broaden understanding of how national and household economic conditions affect
food insecurity. As an example, the report sheds light on why food security has remained essen-
tially unchanged since the 2007-09 recession. Falling unemployment from early post-recession
(2009-10) to 2012, absent any other changes, would suggest a modest decline in the prevalence
of food insecurity. However, this report finds that potential improvement was almost exactly
offset by the effects of higher inflation and the higher relative price of food in 2012.
Keywords: Food security, food insecurity, food prices, unemployment, food price inflation,
SNAP, food assistance
About the Authors

Mark Nord (retired) was a sociologist in the USDA, Economic Research Service, when this
research was conducted. Alisha Coleman-Jensen is a sociologist, and Christian Gregory is an
economist, both in the USDA, Economic Research Service.
Acknowledgments
The authors thank Karen Cunnyngham, Senior Analyst, Mathematica Policy Research; Judi
Bartfield, Professor, University of Wisconsin; Kathryn Law and Sangeetha Malaiyandi, USDA,
Food and Nutrition Service; and Shelly Ver Ploeg, David Smallwood, Jean Buzby, and Ephraim
Leibtag, USDA, Economic Research Service, for their comments on the report. The authors
also thank Dale Simms for editing and preparing the report for publication and Curtia Taylor
for the design and layout of the report.
Contents
Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Related Research. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Data and Measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Analytic Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Findings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
National Economic Factors in Food Insecurity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

Household Characteristics Associated With Food Insecurity. . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Robustness of Findings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Appendix A: Associations of National Economic Conditions

With Very Low Food Security and Children’s Food Insecurity . . . . . . . . . . . . . . . . . . . . . . . . 24
Appendix B: Measures of Household Demographic, Economic,

and Geographic Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Appendix C: Alternative Measures of National Economic Conditions. . . . . . . . . . . . . . . . . . . 29

A report summary from the Economic Research Service June 2014

Find the full report
at www.ers.usda. Mark Nord, Alisha Coleman-Jensen, and Christian Gregory
gov/publications/err-
economic-research-
report/err167
What Is the Issue?

The U.S. Department of Agriculture (USDA) monitors the food security of the Nation’s
households—their consistent access to adequate food for active healthy lives—through
an annual, nationally representative food security survey. Household-level determinants
of food insecurity, like education levels and household income, are well documented in
earlier research. But the effects of national-level economic factors such as inflation and
food prices on food security—and the extent to which changes in both household and
macroeconomic factors account for year-to-year changes in the national prevalence of
food insecurity—are not known. The extent to which nationally aggregated economic
measures such as the unemployment rate can proxy for household-level employment
and labor force data in explaining year-to-year changes in the prevalence of food inse-
curity is also not known.
The association of food insecurity with household characteristics and national economic
conditions over 2001-12 provides insight into why food insecurity remained at about
the same level in 2012 as shortly after the recession (2009-10), despite an improvement
in the national unemployment rate. Providing an explanation of these relationships
can increase public understanding of food insecurity and its causes and may aid in the
development of food and nutrition assistance policies and programs to support the food
security of the Nation’s households.
What Did the Study Find?

Three national-level economic measures taken together accounted for 92 percent of the
year-to-year variation in the national prevalence of food insecurity from 2001 to 2012.
ERS is a primary source
of economic research and
analysis from the U.S.
Department of Agriculture,
providing timely informa-
tion on economic and policy
issues related to agriculture,
food, the environment,and
rural America. www.ers.usda.gov
• An increase of 1 percentage point in the unemployment rate (measured as the highest
monthly unemployment rate in the past calendar year) was associated with an increase of 0.5
percentage point in the prevalence of food insecurity.
• An increase of 1 percentage point in annual inflation, as measured by the Consumer Price

Index (CPI-U), was associated with a 0.5-percentage-point increase in the prevalence of food
insecurity.
• An increase of 1 percent in the annual relative price of food (i.e., the ratio of food price to
the price of all goods and services) was associated with a 0.6-percentage-point increase in the
prevalence of food insecurity.
These associations help explain why the prevalence of food insecurity in 2012 remained essentially
unchanged from the early post-recession period (2009-10). Based on the associations estimated in
the study:
• The observed decline of 1.65 percentage points in the highest monthly unemployment rate
over that period, absent other changes, would have reduced the prevalence of food insecurity
by 0.9 percentage point.
• However, this potential decline was almost exactly offset by effects of higher annual inflation
and higher annual relative price of food in 2012.
Given the associations estimated in this study, the expected prevalence of food insecurity in 2012
was 14.7 percent, nearly the same as the observed prevalence of 14.5 percent.
This study has potential implications for SNAP (USDA's Supplemental Nutrition Asistance Program,
formerly food stamps). The associations of food insecurity with inflation and the relative price of
food suggest that timely and adequate adjustment of SNAP benefits for increases in food prices may
be important for food security.
How Was the Study Conducted?

The study used data on household food security and other household characteristics from the
Current Population Survey Food Security Supplement (CPS-FSS) and national statistics on unem-
ployment and price inflation from the U.S. Bureau of Labor Statistics. The CPS-FSS is an annual,
nationally representative survey of U.S. civilian households sponsored by USDA and administered
by the U.S. Census Bureau. It is the source of data for USDA’s series of annual reports on the food
security of U.S. households.
Associations of the prevalence of food insecurity from 2001-12 with national-level economic
measures were estimated using multivariate linear regression methods. Joint effects of national-level
and household-level characteristics were estimated similarly. The extent to which changes in unem-
ployment, inflation, and the relative price of food may have contributed to changes in the prevalence
of food insecurity from pre-recession (2005-07) and early post-recession (2009-10) to 2012 was esti-
mated based on results of the regression models.
www.ers.usda.gov
Introduction
A household’s ability to consistently put adequate food on the table—its food security—is primarily
a function of its available resources, competing demands for those resources, and the cost of
acquiring food. Food insecurity, by definition, results from a household’s lack of adequate resources
for food, and in any given period the adequacy of those resources is affected by national economic
conditions. It is likely, then, that year-to-year changes in the prevalence of food insecurity are asso-
ciated with changes in both U.S. economic conditions and household circumstances. This report
examines the strength of those associations at both national and household levels.
The U.S. Department of Agriculture (USDA) monitors the food security of the Nation’s households
through an annual, nationally representative food security survey conducted by the U.S. Census
Bureau with support from USDA. USDA’s Economic Research Service (ERS) analyzes the survey
data; publishes an annual report on food security at the national level, for key subpopulations, and
for each State; and conducts research on determinants of food insecurity. Previous research by ERS
and others provides considerable insight into household-level factors that affect food insecurity.
However, the effects of national economic factors such as inflation and food prices on the national
prevalence of food insecurity are unknown. Also, the extent to which changes in household and
macroeconomic factors account for year-to-year changes in the prevalence of food insecurity—and
how well aggregate economic measures such as the unemployment rate proxy for household-level
employment and labor force data—is not known.
This report is an attempt to fill those gaps. We first examine the extent to which year-to-year differ-
ences in the prevalence of food insecurity are associated with differences in national-level economic
conditions. We then examine the joint effects of household-level and national economic conditions
to ascertain the degree to which national economic conditions alone (as opposed to these condi-
tions in combination with household-level factors) explain year-to-year changes in the prevalence of
food insecurity. We investigate why the prevalence of food insecurity has remained at a higher level
during the years immediately following the 2007-08 recession than during the years immediately
preceding it. Following USDA’s release of its annual report, Household Food Security in the United
States in 2012 (Coleman-Jensen et al., 2013), policy officials and media asked USDA why food secu-
rity had not improved in 2012, considering that the unemployment rate had fallen since the end of
the recession and that some other economic indicators had improved (fig. 1). Detailed understanding
of national and household-level determinants of food insecurity may provide an answer to this
question, increase public understanding of food insecurity and its causes, and inform policies and
programs to support the food security of the Nation’s households.
1
Prevalence of U.S. Food Insecurity Is Related to Changes in Unemployment, Inflation, and the Price of Food, ERR-167
Economic Research Service/USDA
Figure 1
Prevalence of food insecurity, very low food security, and the national unemployment rate,
1999-2012
Percent
16
14
Prevalence rate, food insecurity
12
10
8
Annual average unemployment rate
6
2 Prevalence rate, very low food security
0
1999 2000 01 02 03 04 05 06 07 08 09 10 11 12
Note: Shaded areas indicate recessions.

Source: Calculated by USDA, Economic Research Service using food security data from the Current Population Survey
Food Security Supplement and unemployment statistics from the Bureau of Labor Statistics.
2
Related Research
Selection of national economic indicators and household-level characteristics that may explain
changes in U.S. food insecurity was based on previous studies of factors associated with food inse-
curity at the household level and the prevalence of food insecurity at the State level. Tapogna et al.
(2004) found that the prevalence of food insecurity across States was positively associated with
State measures of poverty, housing-cost burden, residential mobility, and the proportion of house-
holds with children. A multivariate model with these variables—plus State unemployment rate
(highest monthly rate during the study period) and the share of the population that was non-Hispanic
White—accounted for 70 percent of the cross-State variance in the prevalence of food insecurity
and 59 percent of the variance in the prevalence of very low food security—the more severe range of
food insecurity in which eating patterns of some household members were disrupted and their food
intake reduced below levels they consider appropriate.
Bartfeld et al. (2006) estimated the joint effects of household and State-level factors on food insecu-
rity. Household characteristics associated with higher probability of food insecurity included:
• Low income—income less than or up to 185 percent of the Federal poverty line
• Low education of adults—especially less than high-school education
• Minority status (Black, Hispanic, and Native American)
• Renting rather than owning the home
• Living in a central city of a metropolitan area
• Having three or more children
• Household headed by a single female
• No adult employed
• No elderly in the household
• Having a household member with a disability
• Having a noncitizen head of household.
Several State-level characteristics were also associated with higher probability of a resident house-
hold being food-insecure, even after adjusting for the household characteristics:
• Low average wages

• High cost of rental housing
• High unemployment rate
• High rate of residential instability
• Low participation in the Supplemental Nutrition Assistance Program (SNAP, formerly the Food
Stamp Program)
• High tax burden on low-income households.
Taken together, these household and State characteristics accounted for 86 percent of the differences
in the prevalence of food insecurity across States during 1998-2001 (Bartfeld et al., 2006).
3
Other studies have modeled household-level factors incidental to the main purpose of the studies
(Nord and Coleman-Jensen, 2010; Nord and Prell, 2011; Nord 2013a). These studies confirm the
consistent associations of food insecurity with low income, unemployment (including part-time
employment by workers who want full-time work) and disability, and the inverse association of food
insecurity with elderly household members. However, in some of these models, with income and
employment more thoroughly specified, associations of food insecurity with low education, minority
status, and residence in central cities became statistically insignificant or reversed.
Food prices have been found to be positively associated with food insecurity at the household level.
The association appears to be strongest for SNAP recipients (Nord 2013a; Gregory and Coleman-
Jensen, 2013; Nord and Prell, 2011), but it is likely that other low-income households are also
affected (Zhang et al., 2013).
4
Data and Measures
Data on household food security are from the Current Population Survey Food Security Supplement
(CPS-FSS) from 2001 to 2012. The CPS-FSS is an annual, nationally representative survey
conducted for USDA by the U.S. Census Bureau. The CPS-FSS provides the data for USDA’s annual
reports on household food security in the United States (Coleman-Jensen et al., 2013). The survey
included an average of 46,149 households each year during the study period.
One individual in each household responds to a series of 10 questions (plus an additional 8 if chil-
dren are in the household) about behaviors and conditions that characterize households when they
are having difficulty meeting their food needs (see box, “Questions Used To Assess Households’
Food Security,” on page 6). Each question asks about occurrences in the previous 12 months and
specifies a lack of money as the reason for not having enough food so as not to count voluntary
fasting or dieting as food insecurity.
Following standard procedures, households were classified as food insecure if they reported three
or more behaviors or conditions indicating food insecurity (Bickel et al., 2000). Households were
further classified as having very low food security if they reported six or more food-insecure behav-
iors or conditions (eight or more in households with children, counting responses to both adult and
child items). Households with children were classified as having food insecurity among children
if they reported two or more potentially food-insecure conditions in response to the eight child-
referenced questions. The primary analyses were based on food insecurity at the household level and
on the annual national prevalence rate of food insecurity based on that measure, i.e., the percentage
of households that were classified as food-insecure. Key analyses (reported in Appendix A) were
repeated for very low food security and for food insecurity among children.
U.S. monthly unemployment rates (seasonally adjusted) were accessed from the Bureau of Labor
Statistics website. The national unemployment rate was hypothesized to be associated with food
insecurity because it is aggregated from household-level employment status, which is known to
be strongly associated with food security at the household level. The measure of unemployment
expected to be most closely associated with the prevalence of food insecurity in a given year is the
highest monthly unemployment rate during that calendar year (fig. 2); households’ measured food
insecurity generally reflects the most food-insecure conditions encountered during the year. Each
question in the food security measure asks whether the condition or behavior ever occurred “in the
past 12 months.” Food insecurity tends to reflect the worst economic conditions over the year and
these conditions are better captured by the highest monthly unemployment rate than by average
annual unemployment.
Measures of inflation and prices were also accessed from the Bureau of Labor Statistics. Inflation
was measured as the percentage change in the annual average Consumer Price Index (CPI-U) for
all goods and services (fig. 3). With rapid price inflation, price increases may outpace increases in
the incomes of some households. Inflation in food prices may pose particular challenges to food
security, and food security may be affected by the current level of food prices compared with prices
for other goods and services, as well as by any recent change in food prices (Nord and Prell, 2011;
Nord, 2013a; Zhang et al., 2013; Gregory and Coleman-Jensen, 2013). The relative price of food
was measured as the annual average CPI-U for food divided by the annual average CPI-U for all
goods and services (fig. 4). Because food is relatively price inelastic, and because households on the
5
Questions Used To Assess the Food Security of Households
in the CPS Food Security Survey
1. “We worried whether our food would run out before we got money to buy more.” Was that often,
sometimes, or never true for you in the last 12 months?
2. “The food that we bought just didn’t last and we didn’t have money to get more.” Was that often, some-
times, or never true for you in the last 12 months?
3. “We couldn’t afford to eat balanced meals.” Was that often, sometimes, or never true for you in the
last 12 months?
4. In the last 12 months, did you or other adults in the household ever cut the size of your meals or skip
meals because there wasn’t enough money for food? (Yes/No)
5. (If yes to question 4) How often did this happen—almost every month, some months but not every
month, or in only 1 or 2 months?
6. In the last 12 months, did you ever eat less than you felt you should because there wasn’t enough money
for food? (Yes/No)
7. In the last 12 months, were you ever hungry, but didn’t eat, because there wasn’t enough money for
food? (Yes/No)
8. In the last 12 months, did you lose weight because there wasn’t enough money for food? (Yes/No)
9. In the last 12 months did you or other adults in your household ever not eat for a whole day because
there wasn’t enough money for food? (Yes/No)
(Questions 11-18 were asked only if the household included children age 0-17)
11. “We relied on only a few kinds of low-cost food to feed our children because we were running out of
money to buy food.” Was that often, sometimes, or never true for you in the last 12 months?
12. “We couldn’t feed our children a balanced meal because we couldn’t afford that.” Was that often,
13. “The children were not eating enough because we just couldn’t afford enough food.” Was that often,
14. In the last 12 months, did you ever cut the size of any of the children’s meals because there wasn’t
enough money for food? (Yes/No)
15. In the last 12 months, were the children ever hungry but you just couldn’t afford more food? (Yes/No)
16. In the last 12 months, did any of the children ever skip a meal because there wasn’t enough money for
food? (Yes/No)
18. In the last 12 months did any of the children ever not eat for a whole day because there wasn’t enough
money for food? (Yes/No)
Determination of Household Food Security Status
Households are classified as food insecure if they responding affirmatively to 3 or more questions. They are
classified as having very low food security if they respond affirmatively to 6 or more questions (8 or more
in households with children, including responses to the child-referenced items).
6
Figure 2
Highest monthly unemployment rate in the calendar year, 2001-12
Percent
12.0
10.0
8.0
6.0
4.0
2.0
0.0
2001 02 03 04 05 06 07 08 09 10 11 12
Source: Calculated by USDA, Economic Research Service using monthly unemployment statistics from the Bureau of
Labor Statistics.
Figure 3
Annual inflation, Consumer Price Index for all goods and services (CPI-U), 2001-12
Percent
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
-0.5
-1.0
2001 02 03 04 05 06 07 08 09 10 11 12
Source: United States Department of Commerce, Bureau of Labor Statistics.
7
Figure 4
Price of food relative to the price of all goods and services, 2001-12*
Percent
103
102
101
100
99
98
97
96
2001 02 03 04 05 06 07 08 09 10 11 12
*The relative price of food was calculated for each year as the annual average Consumer Price Index (CPI-U) for food
divided by the CPI-U for all goods and services, and expressed as a percentage. The base for both indices was the
1982-84 average, so the relative price of food for that period would have been 1.0.
Source: Calculated by USDA, Economic Research Service using Consumer Price Index statistics from the Bureau of
Labor Statistics.
margins of food insecurity spend a higher share of their income on food, such households may be
particularly sensitive to changes in the relative price of food.
Household-level characteristics likely to affect food security were identified from previous research,
and variables representing those characteristics were calculated from CPS-FSS data. These included
variables representing income, employment and labor-force status of adults in the household; house-
hold structure and composition; race and Hispanic ethnicity of the household reference person;
education of the most highly educated adult in the household; recent household moves; and metro-
politan residence (see Appendix B).
8
Analytic Methods
National-level associations of the annual prevalence of food insecurity with three annual economic
variables were estimated using ordinary least-squares regression:
(National-level regression model)
FIy = a + b1*Unempy + b2*Inflationy + b3*Relative-price-of-foody + ey
All variables are annual statistics measured at the national level. FIy is the percentage of households
that were food insecure in year y. Unempy is the highest monthly unemployment rate (seasonally
adjusted) in calendar year y; Inflationy is the annual average rate of CPI inflation for the year; and
Relative-price-of-foody is the relative price of food for the year. Alternative specifications of the
explanatory variables were explored, but none were found to perform better than the initial, theoreti-
cally based specifications (see appendix C). 1
The contributions of changes in the explanatory variables to changes in food insecurity from pre-
recession (average 2005-07) and early post-recession (average 2009-10) to 2012 were calculated
based on the regression coefficients estimated in the national model. Because the model was linear,
the contribution of each explanatory variable was calculated as the coefficient multiplied by the
change in the explanatory variable over the period of interest. The sum of these contributions equals
the change in the dependent variable predicted by the model.
Household-level variables were then incorporated into the model by estimating a two-level random-
intercept model in which the first level comprised households and the second level (for which sepa-
rate intercepts were estimated) comprised survey years:
(Combined household-national-level regression model)
FIh,y = a + BhXh,y + GyMy + eh + fy
In this model, the dependent variable, FIh,y, is a binary variable with value 1 if the household is
food insecure and 0 if the household is food secure. Household-level variables (Xh,y) were assumed
to have the same effect (estimated as Bh) in all years. The intercept for each year was assumed to
depend on the values of the national economic variables for that year (My) with effects estimated
as Gy. Errors at each level were assumed to be distributed normally, with a mean of zero. The two-
level random intercept model takes account of the limited degrees of freedom of the national-level
variables (12 years versus more than half a million households), thus providing theoretically unbi-
ased estimates of the variance of the coefficient estimates on which assessments of statistical signifi-
cance are based.
The combined household/national model was estimated as a linear probability model. While a
logistic or probit specification might be more precise and avoid certain biases, the linear probability
model supports straightforward disaggregation of predicted change into components corresponding
1The national poverty rate, as a measure of income, was also strongly correlated with the prevalence of food insecurity
(r = 0.91), as expected, based on the associations reported in household-level analyses. However, including the poverty
rate in the national-level regression model did not improve the model performance, and the coefficient was weak and not
nearly statistically significant (p = 0.76).
9
to each explanatory variable. Given the range of observed rates of food insecurity (from 10.7 percent
to 14.9 percent), the linear probability model was expected to introduce relatively little estimation
bias.2 Replication of the model using logistic regression confirmed this expectation.
In a preliminary estimation of the two-level model that included the same three national economic
variables as in the national model, the coefficient on the unemployment variable was found to be
weak and not statistically significant (analysis not shown). This was not unexpected, because the
model included a rich set of variables representing employment, unemployment, and labor force
status at the household level. In the final specification of the two-level model, therefore, the national-
level unemployment variable was omitted.
The contributions of changes in explanatory variables to changes in food insecurity from pre-reces-
sion and early post-recession periods to 2012 were calculated as described for the national model,
except that both household- and national-level variables were included. Contributions of household-
level variables were then aggregated to groups of variables representing three general categories:
income, employment/labor-force status, and “other.” The latter category includes a large number
of household-level variables, but they are variables for which the means do not change greatly over
periods of a few years, such as race/ethnicity and metropolitan/nonmetropolitan residence. As a
result, those variables contribute little to the year-to-year changes in food insecurity.
The national-level model has few degrees of freedom and a limited number of independent variables.
Future research using this model would benefit from additional years of data. A concern is that the
three national-level variables may be proxying for other unmeasured factors not accounted for in
the model. However, alternative variables and specifications were explored, along with robustness
checks. Other key analyses were repeated with very low food security at the household level as the
dependent variable and with children’s food insecurity as the dependent variable (see Appendix A).
Data manipulation and model estimations were carried out in SAS 9.2 (SAS Institute Inc., Cary,
North Carolina). The national-level model was estimated using PROC REG. The two-level random-
intercept model was estimated using PROC MIXED.3 Robustness of the results of the linear
two-level model was assessed in a random-effects logistic regression model using the XTLOGIT
procedure in STATA 11.2 (www.stata.com).
Supplement weights provided by the Census Bureau indicate how many households are represented
by each interviewed household and were used for all calculations involving household-level data.4
2Predicted values from the regression model ranged from -.15 to .84, and about 18 percent of observations had values
less than 0. However, these theoretically impossible probability values appear not to have biased the annual prevalence
estimates substantially; the annual prevalence rates predicted by the linear and logistic models were essentially identical.
3The MIXED procedure model was specified as random intercept with survey year as the “subject.” The “degrees of
freedom” was specified as “between-within.” Under this specification, SAS assigned degrees of freedom based on the
number of years to variables that did not differ within the same year, thus to national economic variables.
4Food security weights multiplied by 0.75 were used rather than supplement weights for the 2007 CPS-FSS. This
ensured that prevalence rates for that year were consistent with published rates, while taking into account the relatively
small sample size in that year (USDA, Economic Research Service, 2008).
10
Findings
National Economic Factors in Food Insecurity

About 92 percent of the year-to-year variation in the national prevalence of household food inse-
curity from 2001 to 2012 was associated with changes in the national unemployment rate, inflation
over the previous year, and the relative price of food (table 1). On average, the observed prevalence
rate differed from the rate predicted by the model by about a third of a percentage point (mean abso-
lute error 0.341). Holding other modeled conditions constant, a 1-percentage-point increase in the
unemployment rate or in annual CPI inflation was associated with an increase of 0.523 percentage
point in the prevalence of food insecurity. A 1-percentage-point increase in the price of food rela-
tive to the price of all goods and services was associated with an increase of 0.583 percentage point.
The standardized regression coefficients, which take into consideration the extent to which each
of the economic predictors vary from year to year, indicate that changes in the unemployment rate
and relative price of food accounted for more of the year-to-year change in food insecurity than the
change in CPI inflation. Given the relationships among the model variables over the entire period,
the observed prevalence of food insecurity in 2012 was very near the predicted level (fig. 5).5
The regression results suggest why the prevalence of food insecurity remained above pre-recession
levels in 2012 (table 2). The highest monthly unemployment rate was 3.23 percentage points higher
in 2012 than in the period immediately before the recession (2005-07). The prevalence of food inse-
curity is estimated to have increased 1.69 percentage points as a result. The price of food relative to
Table 1
Linear regression of annual prevalence rate of food insecurity (percent of households) on
national unemployment, inflation, and relative price of food, 2001-12
Coefficient Std.
Variable (%-age points) coefficient P
Constant -50.010
Highest monthly unemployment rate in year (percent) .523 .542 .062
CPI inflation, annual average (percent) .471 .288 .024
Relative price of food (CPI food as a percent of total
.583 .569 .041
CPI)
Number of years 12
Adjusted R2 .921
Dependent mean (prevalence of food insecurity) 12.602
Standard error of model-predicted prevalence .513
Mean absolute error of model-predicted prevalence .341
Food Security Supplement and statistics on unemployment and inflation from the Bureau of Labor Statistics.
5There was little evidence that regression results were distorted by effects of serial autocorrelation. The Durbin-Watson
statistic for first-order autocorrelation was 1.60 (p = 0.16 for positive autocorrelation, and p = 0.84 for negative autocor-
relation.)
11
Figure 5
Prevalence of food insecurity, 2001-12, and predicted prevalence based on highest monthly
national unemployment rate, annual inflation, and relative price of food
Percent
18
Observed food insecure
16
Predicted food insecure
14
12
10
0
2001 03 05 07 09 11
Food Security Supplement and statistics on unemployment and inflation from the Bureau of Labor Statistics. Predicted
prevalence rates are based on regression model in table 1.
Table 2
Sources of regression-predicted change in the national prevalence rate of food insecurity
National economic Prevalence of food
determinants insecurity
Relative price
Unemployment1 CPI inflation2 of food3 Predicted4 Observed
Coefficient (from table 1) 0.523 0.471 0.583
Level 2012 8.30 2.07 101.82 14.71 14.51
Level 2005-07 average 5.07 3.15 97.45 10.98 11.02

Change 2005-07 to 2012 3.23 -1.09 4.37 3.73 3.49
Coefficient x change 2005-
1.69 -.51 2.55 3.73
07 to 2012
Level 2009-10 average 9.95 .64 101.16 14.51 14.60

Change 2009-10 to 2012 -1.65 1.43 .66 .20 -.09
-.86 .67 .39 .20
10 to 2012
Notes:
1Highest monthly unemployment rate during the year (percent)
2Percentage change in annual average Consumer Price Index (CPI-U)
3CPI for food as a percentage of CPI for all goods and services (CPI-U)
4Predicted prevalence based on regression model in table 1.
12
the price of all goods and services was 4.37 percentage points higher in 2012 than in 2005-07, and
this factor is estimated to have increased the prevalence of food insecurity by 2.55 percentage points.
Changes in those two factors were partially offset by a lower rate of CPI inflation in 2012, which
is estimated to have reduced the prevalence of food insecurity by 0.51 percentage point. Given the
changes in these three factors, the prevalence of food insecurity was predicted to be 3.73 percentage
points higher in 2012 than in 2005-07—slightly more than the observed increase of 3.49 percentage
points.
Comparing 2012 with the early post-recession period (2009-10) provides additional insight into why
food insecurity rates have changed so little since the recession. The highest monthly unemployment
rate fell from 2009-10 to 2012 and, absent any other changes, it is estimated that food insecurity
would have fallen by 0.86 percentage point as a result (table 2). However, that potential decline was
offset by the effects of higher overall inflation and higher relative price of food in 2012 compared
with the early post-recession period. The net result was a predicted increase in food insecurity of
0.20 percentage point from 2009-10 to 2012, compared with an observed decline of 0.09 percentage
point.
Household Characteristics Associated With Food Insecurity

Associations of food insecurity with household characteristics were consistent with those found in
previous studies. The coefficients in table 3 represent the difference in probability of food insecu-
rity that is associated with a one-unit difference in the covariate. With the exception of income and
numbers of children in two age ranges, the covariates are all binary (zero/one) variables, so the
difference represented by the coefficient is the difference in probability of food security between
households with and without the characteristic that is identified by the variable. Or, for a variable
that is one of a set of several binary variables representing a multiple-category classification (such
as household composition) the coefficient represents the difference in probability of food insecu-
rity between households with the identified characteristic and households in the reference category
for that set of variables. Multiplying a coefficient by 100 expresses these differences in percentage
points. The average prevalence of food insecurity over the 12-year period was 12.6 percent, which
provides a context for interpreting these percentage-point differences.
Food insecurity was strongly associated with household income and employment. For example,
households with incomes 1.5 times the poverty line were about 6 percentage points less likely to
be food insecure than households with incomes at half the poverty line.6 Food insecurity was also
associated strongly with household employment and labor force status. Holding income and all
other covariates constant, compared with households where all adults were out of the labor force for
reasons other than retirement and disability, the following associations were estimated (principal
labor force status):
• Households with an adult employed full-time were 4.13 percentage points less likely to be food
insecure.
• Households with no adult employed full-time, but with one or more adults retired, were 6.44
percentage points less likely to be food insecure.
6The complex specification of income in this model complicates interpretation. In the example described, the differ-
ence in food insecurity associated with a difference of 1 “poverty level” unit—from 0.5 to 1.5—takes into account both
the coefficient on income and the coefficient on income-squared.
13
Table 3
Coefficients from the combined household-national-level regression model of food
insecurity and means of the model variables for three periods
Mean Mean Mean
Coefficient1 p 2005-07 2009-10 2012
Intercept -0.4794 <.001
Income (ratio to poverty line) -.0675 <.001 3.8145 3.6647 3.6215
Income (ratio to poverty line)
.0041 <.001 21.1760 19.5522 18.9516
squared
Income less than 50% poverty line .0382 <.001 .0392 .0445 .0474
Income in lowest reported category -.0500 <.001 .0245 .0270 .0267
Income not reported .0013 .260 .2000 .2155 .2140
Labor force status, principal (refer-
ence: out of labor force, not retired
or disabled)
Employed full time -.0413 <.001 .6942 .6478 .6538
Out of labor force—retired -.0644 <.001 .1790 .1874 .1920
Part-time for non-economic reasons -.0285 <.001 .0422 .0451 .0438
Part-time for economic reasons .1196 <.001 .0085 .0221 .0177
Unemployed .1306 <.001 .0186 .0367 .0278
Out of labor force—disabled .1486 <.001 .0365 .0386 .0421
Labor force status of other adults
(omitted: out of labor force, not re-
tired or disabled, or no other adult)
Employed full time -.0139 <.001 .2770 .2393 .2477
Out of labor force—retired -.0031 .071 .1165 .1199 .1298
Part-time for non-economic reasons -.0140 <.001 .1255 .1129 .1095
Part-time for economic reasons .0694 <.001 .0150 .0352 .0341
Unemployed .0775 <.001 .0369 .0757 .0634
Out of labor force—disabled .0780 <.001 .0518 .0581 .0583
Household composition (reference:
married couple with child/children)
Single male with child/children .0318 <.001 .0224 .0247 .0247
Single female with child/children .0960 <.001 .0828 .0835 .0853
Other household with child/children .0248 <.001 .0055 .0052 .0048
Two or more adults, no child .0011 .457 .3870 .3958 .4079
Male living alone -.0021 .290 .1191 .1219 .1223
Female living alone .0066 <.001 .1523 .1490 .1473
Number of children ages 5-9 years .0112 <.001 .1784 .1771 .1713
Number of children ages 10-17
.0125 <.001 .2901 .2799 .2713
years
One or more elderly in the house-
-.0490 <.001 .2330 .2463 .2627
hold
Race and Hispanic ethnicity (refer-
ence: White non-Hispanic)
Black non-Hispanic .0443 <.001 .1209 .1229 .1232
—continued
14
Table 3
Coefficients from the combined household-national-level regression model of food inse-
curity and means of the model variables for three periods—continued
Mean Mean Mean
Coefficient1 p 2005-07 2009-10 2012
Hispanic .0354 <.001 .1092 .1146 .1257
American Indian or Alaska Native .0430 <.001 .0073 .0084 .0094
Educational attainment of most
highly educated adult (reference:
high school or GED)
Less than high school .0374 <.001 .0857 .0777 .0695
Some college, no four-year degree -.0026 .020 .2965 .3006 .3024
Bachelor degree -.0346 <.001 .2220 .2265 .2322
Professional or advance degree -.0361 <.001 .1401 .1513 .1607

Household moved recently .0216 <.001 .0495 .0448 .0459
Metropolitan/nonmetropolitan
residence (reference: metropolitan,
not in principal city)
Metropolitan, principal city .0061 <.001 .2781 .2761 .2794
Metropolitan, not specifically identified -.0021 .098 .1428 .1430 .1412
Not in metropolitan area -.0155 <.001 .1685 .1686 .1651
National-level economic measures

Inflation in CPI-U since previous
.0037 .044 3.1799 .6444 2.0690
year
Price of food relative to all goods
.0079 <.001 97.4090 101.1539 101.8200
and services
Number of cases 541,687 127,786 89,955 43,792
1Coefficientsrepresent probabilities; multiply by 100 to represent associations in percentage points.
• Households with an adult employed part-time for economic reasons (i.e., underemployed), those
with an adult unemployed (looking for work), and those with an adult out of the labor force
because of disability were 12 to 15 percentage points more likely to be food insecure (house-
holds in these categories had no adult employed full-time or retired or employed part-time for
non-economic reasons (i.e., voluntarily part-time)).
Employment and labor-force status of other adults in the household was also associated with food
insecurity. The associations were in the same direction as those for principal labor force status, but
the coefficients were smaller. For example, households with another adult employed full-time were
1.39 percentage points less likely to be food insecure than households with no other adults or with
other adults out of the labor force for reasons other than retirement and disability.
Even after taking into account income and employment, several other household characteristics were
associated with food insecurity.
15
• Single parents with children were more likely to be food insecure than other households (3.18
percentage points for single men with children and 9.60 percentage points for single women
with children).
• Households with elderly were less likely (4.90 percentage points) to be food insecure than
households with no elderly.
• Households with Black, Hispanic, American Indian, or Alaska Native reference persons were
more likely to be food insecure.
• Households in which no adult graduated high school or completed a GED were more likely
to be food insecure than those with a high school graduate. Households with an adult who
completed a 4-year college degree or higher were less likely to be food insecure than those in
which the most highly educated adult held only a high school diploma or GED.
• Households in the “principal cities” of metropolitan areas (i.e., within the incorporated area of
the largest city or cities comprising the area) were 0.61 percentage point more likely to be food
insecure than those in suburban and exurban areas around metropolitan areas.
• Households outside of metropolitan areas were 1.55 percentage points less likely to be food
insecure than those in suburban and exurban areas around metropolitan areas. Apparently, the
nonmetro disadvantage in food insecurity reported for most years in USDA’s annual food secu-
rity reports (Coleman-Jensen et al., 2013) is accounted for entirely by lower household income
and differences in other household characteristics included in this regression model.
The coefficients on the two national-level economic variables in the two-level model remained
statistically significant with the richer set of controls for household-level income, employment, and
other characteristics. The coefficient on CPI-U inflation was smaller in the two-level model (0.37
percentage point) than in the national model (0.471 percentage point), while the coefficient on the
relative price of food was larger in the two-level model (0.79 percentage point) than in the national
model (0.583 percentage point). Considering the estimation errors in both models, however, the
differences between these coefficients in the two models were not statistically significant.
The size of the coefficient on the relative price of food in both models, but especially in the two-level
model, is larger than expected given other evidence for the association of food insecurity with food
spending (Nord and Prell, 2011). Examination of the relative price of food over the study period
reveals that the variable changed relatively little prior to the recession, increased during the reces-
sion, and remained relatively unchanged subsequently (fig. 4). Thus, in the regression analyses, the
coefficient could be representing effects not only of the relative price of food, but also of any other
variables not controlled in the model that changed coincident with the recession. With a dummy
variable added to the model to identify recession and post-recession years, the coefficient on relative
price of food was about one-fourth as large as that in table 3 (0.18 compared with 0.79, analysis not
shown). Nonetheless, the findings on the importance of food prices to food insecurity are consistent
with previous research on food security and food prices/spending (Gregory and Coleman-Jensen,
2013; Zhang et al., 2013; Nord and Prell, 2011; Nord, 2013).
The predicted changes in food insecurity from pre-recession (2005-07) to 2012 based on the
two-level model were similar to those based on the national model, although the relative size
of the factors’ contribution differed. The two-level model suggests that from 2005-07 to 2012,
lower income and less favorable employment accounted for an increase in food insecurity of 1.24
percentage points (table 4), compared with 1.69 percentage points for the unemployment rate in
the national-level model (table 2). The two-level model assigned a larger increase in food insecu-
16
Table 4
Sources of regression-predicted change in the national prevalence rate of food insecurity
based on the combined household-national-level model
2005-07 2009-10
to to
Statistic 2012 2012
Percent
Prevalence of food insecurity
Base period (2005-07 average or 2009-10 average) 11.01 14.60
2012 14.51 14.51
Change from base period to 2012 3.50 -.09
Accounted for by changes in explanatory variables

in the model*
Income .40 .05
Employment and labor force status .84 -.28
Other household characteristics -.25 -.11
CPI-U inflation, all goods and services -.41 .52
Price of food relative to price of all goods and services 3.50 .53
Total accounted for by model 4.08 .71
Not explained by model* -.58 -.80

*Based on coefficients and means in table 3.
rity to the higher relative price of food (3.50 percentage points) than did the national-level model
(2.55 percentage points). In both models, the effects of lower inflation in 2012 partially offset these
increases (-0.41 percentage point in the two-level model and -0.51 percentage point in the national
model). The reduction in food insecurity of 0.25 percentage point accounted for by “other household
characteristics” in the two-level model reflects primarily an increase in the number of households
with elderly members and improvements in educational attainment from 2005 to 2012.
The two-level analysis comparing the early post-recession (2009-10) to 2012 was also similar to
the national model. The two-level model assigned a smaller positive role to income and employ-
ment (a decline in food insecurity of 0.23 percentage point, compared with 0.86 percentage point in
the national model). In both models, this improvement was more than offset by an increase in the
prevalence of food insecurity associated with higher inflation and the higher relative price of food
in 2012. The increase in food insecurity associated with higher overall inflation was estimated at
0.52 percentage point by the two-level model and 0.67 percentage point by the national model. The
increases associated with the higher relative price of food were 0.53 percentage point in the two-
level model and 0.39 percentage point in the national model.
Predicted annual rates of food insecurity based on the two models were similar (fig. 6). Both indi-
cate that the observed prevalence of food insecurity in 2012 was near or slightly below the predicted
17
Figure 6
Prevalence of food insecurity, 2001-12, and predicted prevalence based on two
regression models
Percent
18
16
14
12
10
8
Observed prevalence of food insecurity
6 Predicted based on national economic variables only
Predicted based on combined household and national variables
4
0
2001 03 05 07 09 11
Table 5
Comparisons of model fit and accuracy of annual estimated prevalence rates of food
insecurity by the national-level regression model and the combined household-national-
level regression model
National-level Combined household-
Statistic model national-level model1
R-squared .943 .945
Adjusted R-squared1 .921 .940

Standard error (percentage points) .513 .450
Mean absolute error (percentage points) .341 .318
Largest overestimate (percentage points) .653 .596
Largest underestimate (percentage points) .953 .900
N 12 12
1Statisticsfor the combined household-national-level model were calculated from a second regression in which survey
year was the unit of analysis, the observed prevalence of food insecurity was the dependent variable, and the predicted
prevalence based on the combined household-national-level model was the single independent variable. The adjusted
R-squared for this model is overstated because it does not take account of the degrees of freedom lost due to the two
national-level variables in the combined model.
18
level given household and national conditions in that year and the associations observed over the
12-year period.
Estimates from the two-level (household and national) model were, on average, nearer the observed
values than the estimates from the national-level model, but the differences were not large (table 5).
Total R-squared was nearly the same for the two models.7 Mean absolute error was smaller for the
two-level model (0.318 percentage point) than for the national model (0.341), as were the largest
positive and negative errors.
7Adjusted R-squared was larger for the two-level model (0.940) than for the national-level model (0.921), but the unad-
justed R-squared, which was essentially the same for the two models, may provide a more appropriate comparison for as-
sessing future performance. Although the secondary regression (of observed prevalence by year on predicted prevalence
from the two-level model) included only that single independent variable, this understates the loss of degrees of freedom
due to the model because two national-level variables were included in the two-level model on which the single indepen-
dent variable in the secondary regression was based.
19
Robustness of Findings
The combined household/national model was replicated using logistic regression (results not shown).
Given the binary character of the dependent variable (food insecurity), logistic regression should
provide more accurate coefficient estimates. The linear model was preferred for the purpose of this
study because it is more understandable to a broad audience and it more readily supports decompo-
sition of year-to-year changes into changes due to each of the explanatory variables. But use of the
linear model is appropriate only if it does not substantially distort coefficients. The logistic regres-
sion analysis assessed the extent of those distortions and verified that they were not substantial.
Coefficients from the logistic regression (converted to marginal effects) were essentially identical to
the coefficients from the linear model, and the two national-level variables in the two-level model
(inflation and relative price of food) remained statistically significant in the logistic model.
Variance estimates in the two-level regression analysis could be understated because most house-
holds are in the CPS-FSS twice. Households in the monthly CPS are interviewed in each of 4
successive months, then again in the same 4 months a year later; the possible lack of independence
between these two observations was not accounted for in the model. To assess possible bias on vari-
ance estimates of regression coefficients that could result from this survey schedule, the analysis
was repeated twice, first with the subsample of households that were in 1 of their first 4 months in
the CPS sample, then with households that were in 1 of their final 4 months (analysis not shown).
In each of these subsamples, no household was included more than once. In the subsample analyses,
the size of the coefficients on the two national-level variables—those of primary interest for the
analysis—differed by less than 5 percent from the main-model estimates. The coefficients on rela-
tive cost of food remained highly statistically significant in both subsamples (p < 0.0001). The
coefficients on annual inflation remained statistically significant with single-tailed p < 0.05 in both
subsamples.
20
Conclusion
The association of food insecurity with household characteristics and national economic conditions
over 2001-12 provides insight into why food insecurity remained at about the same level in 2012 as
shortly after the recession. Although employment and income had improved somewhat by 2012, the
predicted reduction in food insecurity associated with those improvements was only in the range
of 0.23 to 0.86 percentage point. However, that potential improvement was offset by the effects of
higher inflation and higher prices of food relative to other goods and services in 2012.
The relationships estimated in this study may inform future queries about why food security
improves, worsens, or remains unchanged from prior years. Effects of the relative price of food may
be overestimated in this report because of the co-occurrence of the recession and the increase in that
measure. (Other factors not accounted for in the models may have changed at the time of the reces-
sion and remained unchanged since.) Monitoring this association in the next few years might better
capture the effect of the relative price of food on food insecurity. The effect of food prices estimated
in this study were in the same direction but larger than in related studies.
Differences in the accuracy of food insecurity estimates between the national model (with 3 inde-
pendent variables) and the household/national model (with 37 household-level variables and 2
national-level variables) were relatively small. Contributions of the major determinants to year-to-
year changes in food insecurity were also similar between the two models. The two-level model
will likely provide more accurate estimates of food insecurity prevalence in future years and a more
detailed and nuanced assessment of the factors underlying changes in rates. On the other hand, the
national-level model is simpler, relies on familiar national-level statistics, and is likely easier to
explain to policy officials, media, and the general public. Further, the national unemployment rate
appears to serve as a proxy for more difficult to obtain household economic characteristics. Thus,
both models may contribute to increasing public understanding of important determinants of U.S.
national food security.
The national model is only based on 12 years of data with 9 degrees of freedom. It will be neces-
sary to monitor results of this model in coming years with more data. While the model appears to
explain food insecurity rates very well, the limitation of the model is that it may be prone to outliers
and idiosyncratic conditions.
Previous research has established that employment and other factors affecting household resources
are important determinants of food security. The findings of this study reinforce those findings and
suggest, additionally, that food security is affected by recent inflation and the price of food relative
to other goods and services. Over the period studied, differences in the national annual prevalence of
food insecurity were associated as strongly with inflation and food prices, taken together, as with the
unemployment rate.
The associations of food insecurity with inflation and relative food prices warrant further research
because of their potential policy implications for SNAP. If the associations reflect causal relation-
ships, and if the effects are as strong as suggested by the results, then they point to the importance
of timely and adequate adjustment of SNAP benefits for changes in food prices. Currently, SNAP
benefits are adjusted for inflation in October of each year, based on food-price data from June of that
year. Thus, by the end of the fiscal year in the following September, any further inflation that may
have occurred over a period of 15 months is not reflected in the SNAP benefit amount. If the effects
21
of food-price inflation on food insecurity are as strong as suggested by results of this study, then this
lag in adjustment could weaken the food security of SNAP recipients, especially when inflation is
high. For every 1 percentage point of food-price inflation not accounted for in SNAP benefits, food
insecurity may be expected to increase by nearly a half percentage point.
22
References
Bartfeld, Judi, Rachel Dunifon, Mark Nord, and Steven Carlson. 2006. What Factors Account
for State-to-State Differences in Food Security? Economic Information Bulletin No. 20.
U.S.Department of Agriculture, Economic Research Service.
Bickel, Gary, Mark Nord, Christofer Price, William L. Hamilton, and John T. Cook. 2000. Guide to
Measuring Household Food Security, Revised 2000. USDA, Food and Nutrition Service. http://
www.fns.usda.gov/fsec/files/fsguide.pdf/
Coleman-Jensen, Alisha, Mark Nord, and Anita Singh. 2013. Household Food Security in the United
States in 2011, Economic Research Report No. 155. U.S. Department of Agriculture, Economic
Research Service.
Edin, Kathryn, Melody Boyd, James Mabli, Jim Ohls, Julie Worthington, Sara Greene, Nicholas
Redel, and Swetha Sridharan. 2013. “SNAP Food Security In-Depth Interview Study: Final
Report” Family Programs Report. U.S. Department of Agriculture, Food and Nutrition Service.
Gregory, Christian A., and Alisha Coleman-Jensen. 2013. “Do High Food Prices Increase Food
Insecurity in the United States?” Applied Economic Perspectives and Policy, Oct., doi: 10.1093/
aepp/ppt024.
Nord, Mark. 2013a. Effects of the Decline in the Real Value of SNAP Benefits from 2009 to 2011.
Economic Research Report No. 151. U.S. Department of Agriculture, Economic Research
Service.
Nord, Mark. 2013b. “Youth Are Less Likely To Be Food Insecure Than Adults in the Same
Household,” Journal of Hunger and Environmental Nutrition 8(2):146-163.
Nord, Mark, and Alisha Coleman-Jensen. 2010. “Food Insecurity after Leaving SNAP,” Journal of
Hunger and Environmental Nutrition 5(4): 434-453.
Nord, Mark, and Mark Prell. 2011. Food Security Improved Following the 2009 ARRA Increase in
SNAP Benefits. Economic Research Report No. 116. U.S. Department of Agriculture, Economic
Research Service.
U.S. Department of Agriculture, Economic Research Service. 2008. December 2007 Microdata
File: Technical Documentation. http://www.ers.usda.gov/datafiles/Food_Security_in_the_
United_States/Current_Population_Survey/2007/notes1207.pdf.
Tapogna, John, Allison Suter, Mark Nord, and Michael Leachman. 2004. “Explaining Variations in
State Hunger Rates,” Family Economics and Nutrition Review 16(2): 12-22. www.cnpp.usda.gov/
Publications/FENR/V16N2/fenrv16n2.pdf
Zhang, Qi, Sonya Jones, Christopher J. Ruhm, and Margaret Andrews. 2013. “Higher Food Prices
May Threaten Food Security Status Among American Low-Income Households with Children,”
Journal of Nutrition 143(10): 1659-1665.
23
Appendix A: Associations of National Economic
Conditions With Very Low Food Security and Children’s
Food Insecurity
The national-level model (table 1) was repeated with very low food security as the dependent vari-
able (table A-1). The annual prevalence of very low food security from 2001 to 2012 was less
strongly associated with unemployment, inflation, and the relative price of food than was overall
food insecurity. About 81 percent of the year-to-year variation in the national prevalence of very low
food security was accounted for by the model, compared with 92 percent for overall food insecurity.
On average, the observed very low food security rate differed from the rate predicted by the model
by 0.307 percentage point, smaller than for overall food insecurity (0.341 percentage point). But the
mean prevalence of very low food security (5.5 percent) was less than half that of overall food inse-
curity (12.6 percent), so the proportional error was greater for very low food security.
The association of the highest monthly unemployment rate was weaker with very low food secu-
rity than with food insecurity overall, and the coefficient was not statistically significant (p=.492).
However, the associations of very low food security with annual CPI inflation and the relative price
of food were statistically significant. Tabled values indicate significance with 90-percent confidence,
but those values are based on two-tailed tests. If statistical significance is assessed based on single-
tailed tests—which is appropriate considering that the hypothesized associations are directional—
the associations were significant with 95-percent confidence. Holding other modeled conditions
constant, an increase in annual CPI of 1 percentage point was associated with an increase of 0.28
percentage point in the prevalence of very low food security. A 1-percentage-point increase in the
price of food relative to the price of all goods and services was associated with an increase of 0.445
percentage point in very low food security. The standardized regression coefficients, which take into
consideration the extent to which each of the economic predictors vary from year to year, indicate
Table A-1
Linear regression of annual prevalence rate of very low food security (percent of house-
holds) on national unemployment, inflation, and relative price of food, 2001-12
Variable Coefficient Std. coefficient P
Constant -41.278
Highest monthly unemployment rate in year (percent) .150 .279 .492
CPI inflation, annual average (percent) .280 .308 .093
Relative price of food (CPI food as a percent of total CPI) .445 .780 .064
Number of years 12
Adjusted R2 .811
Dependent mean (prevalence of very low food security) 4.528
Standard error of model-predicted prevalence .443
Mean absolute error of model-predicted prevalence .307

24
that the relative price of food was more strongly associated with the prevalence of very low food
security than was annual CPI inflation. Given the relationships among the model variables over the
entire period, the prevalence of very low food security in 2012 was very near the expected level (fig.
A-1).
The decomposition of sources of change in very low food security from pre-recession and early post-
recession to 2012 are similar to those for food insecurity (table A-2). From pre-recession (2005-07
average) to 2012, the higher relative price of food was associated with an increase in very low food
security of 1.95 percentage points. Higher unemployment was associated with an additional increase
of 0.49 percentage point. (This is the in-sample association, but because it is not statistically signifi-
cant, there is some uncertainty about the causal relationship and about the likelihood of a similar
relationship in future years.) These effects were partially offset by lower overall inflation (in 2012),
which is estimated to have reduced the prevalence of very low food security by 0.30 percentage
point. Thus, the observed increase of 1.75 percentage points in the prevalence of very low food
security from 2005-07 to 2012 was slightly less than expected based on the regression model (2.13
percentage points).
Comparing 2012 with the early post-recession period (2009-10) may help explain the lack of
improvement in very low food security since the recession. Unemployment fell from 2009-10 to
2012, and absent any other changes, it is estimated that very low food security might have fallen by
0.25 percentage point as a result. However, that potential decline was offset by the effects of higher
overall inflation and the higher relative price of food in 2012 compared with 2009-10. The net result
was a predicted increase in the prevalence of very low food security of 0.45 percentage point from
2009-10 to 2012, compared with an observed increase of 0.19 percentage point.
Figure A-1
Prevalence of very low food security, 2001-12, and predicted prevalence based on
highest monthly national unemployment rate, annual inflation, and relative price of food
Percent
7
Observed very low food security Predicted very low food security
6
0
2000 02 04 06 08 10 12 14
Food Security Supplement and statistics on unemployment and inflation from the Bureau of Labor Statistics. Predicted
prevalence rates are based on regression model in table A-1.
25
Table A-2
Sources of regression-predicted change in the national prevalence rate of very low food
security
Prevalence of food
National economic determinants insecurity
Relative price
Unemployment1 CPI inflation2 of food3 Predicted4 Observed
Coefficient (from table A-1) 0.150 0.280 0.445
Level in 2012 8.30 2.07 101.82 5.83 5.72
Level 2005-07 average 5.07 3.15 97.45 3.71 3.97
Change 2005-07 to 2012 3.23 -1.09 4.37 2.13 1.75

.49 -.30 1.95 2.13
07 to 2012
Level 2009-10 average 9.95 .64 101.16 3.38 5.53

Change 2009-10 to 2012 -1.65 1.43 .66 .45 .19
-.25 .40 .30 .45
10 to 2012
Notes:
1Highest monthly unemployment rate during the year (percent)
2Percentage change in annual average Consumer Price Index (CPI-U)
3CPI for food as a percentage of CPI for all goods and services (CPI-U)
4Predicted prevalence based on regression model in table A-1.
A two-level (household/national) model was also estimated for very low food security following the
procedures described for overall food insecurity (results not shown). As was the case in the food
security model, the national unemployment variable was not significant and was omitted from the
final model. Coefficients for both the annual inflation and relative price of food variables were statis-
tically significant with values that differed little from those in table A-1. Predicted prevalence rates
for each year were essentially identical to those from the national-level model shown in figure A-1.
Given the similarities of the coefficients and predicted prevalence rates, a decomposition of sources
of change was not conducted for the prevalence of very low food security based on the two-level
model.
The national-level model was repeated once more with food insecurity among children as the depen-
dent variable (results not shown). Model fit was weaker than for the household-level measures.
Adjusted R-squared was 0.53 with all three national economic variables in the model, and none of
the coefficients was statistically significant. Bivariate associations with the prevalence of children’s
food insecurity were statistically significant and of nearly the same strength for both the highest
monthly unemployment rate (coefficient=0.39, adjusted R-squared=.57, p=.003) and the relative
price of food (coefficient=.42, adjusted R-squared=.60, p=.002). With both variables in the model,
however, neither was statistically significant (p=.59 for unemployment and p=.37 for relative price
of food). There remains, therefore, considerable uncertainty about the causal relationships of the two
variables with children’s food insecurity.
26
Appendix B: Measures of Household Demographic,
Economic, and Geographic Characteristics
This appendix provides descriptions of each of the household-level variables in the two-level model
in table 3. These were calculated from variables in the labor force core section of the Current
Population Survey.
Income (ratio to poverty line)—Annual household income is reported in ranges in the core labor
force portion of the CPS. Income for each household was approximated as the center of the reported
range. The poverty line for each household was assigned from the Census Bureau’s table of poverty
thresholds for the year of the survey, based on the number of adults and the number of children in
the household and whether the household reference person was younger or older than 65. Income
was specified as quadratic (i.e., including Income Squared) because the association of income
with food insecurity is substantially nonlinear. A binary variable indicating Income Less Than
50 Percent of the Poverty Line was also included to account for further nonlinearity in the lowest
income range.
Income in lowest reported category—A binary variable indicates whether the household reported
annual income in the lowest income (less than $5,000) category. This category may include a
mixture of households that have very low income and low resources and other households that have
temporarily low income but have other resources to draw on. There is evidence of this in the present
study in that the coefficient on this variable is negative and statistically significant, indicating that
households with incomes reported in the lowest category have better food security than expected
given their reported income and other characteristics.
Income not reported—A binary variable indicating that the household did not report annual
income. (About 20 percent of households did not report annual income in the CPS.)
Labor force status, principal—The labor force status of each adult (age 18 or older) was assigned
in one of six categories, based on the variables monthly labor force recode (PEMLR) and full/part-
time work status (PRWKSTAT) in the core labor force portion of the CPS. The six categories are:
1. Employed full time

2. Out of the labor force—retired
3. Employed part time for noneconomic reasons (the individual was not seeking full-time work)
4. Employed part time for economic reasons (the individual wanted to work more hours but could
not find a full-time job)
5. Unemployed (looking for work)
6. Out of the labor force—disabled
7. Out of the labor force—not retired or disabled.
Principal labor force status for the household was assigned as the lowest-numbered category of any
adult in the household. Thus, a household in category 5 would have had no adult employed or out of
the labor force due to retirement and at least one adult unemployed and looking for work. The set of
categories was represented in the multivariate model as a set of six dummy variables, with category
7 (out of the labor force—not retired or disabled) as the reference category.
27
Labor force status of other adults—The labor force status of any additional adults in the household
was represented by a series of six binary variables representing the categories described above. More
than one of these variables could have value 1 if there were more than two adults in the household.
All six variables were zero if there was only one adult present or if any other adults were in the
omitted category (out of the labor force—not retired or disabled).
Household composition was represented by a set of six dummy variables. The reference category
was married couple with child/children.
Number of children ages 5-9 and 11-17 adjusts both for the difference in expenses for older versus
younger children, and also for the greater extent to which adults shield younger children from food
insecurity even if the adults are food insecure (Edin et al., 2013; Nord, 2013b). The presence of chil-
dren younger than age 5 only is accounted for by the household composition variables together with
values of zero for these two variables.
One or more elderly in the household is a binary variable indicating the presence of at least one
household member age 65 or older.
Race and Hispanic ethnicity of the household reference person is represented by a set of three
dummy variables for Black non-Hispanic, Hispanic, and American Indian or Alaska Native. The
reference category is White non-Hispanic and Asian non-Hispanic. Those who reported their race as
American Indian or Alaska Native and also reported Hispanic ethnicity were classified as American
Indian or Alaska Native.
Educational attainment of most highly educated adult is represented by a set of three dummy
variables with just high school or GED as the reference category.
Household moved recently is a binary variable indicating that the household has moved into the
sampled address since that address was first in the survey sample—a period that could have been
from 1 to 15 months.
Metropolitan/nonmetropolitan residence is represented by a set of three dummy variables with

metropolitan, not in principal city as the reference category. These would generally be households
in the suburban and exurban areas outside the incorporated city or cities at the heart of the metro-
politan area.
28
Appendix C: Alternative Measures of National Economic
Conditions
The following alternative and additional specifications of national-level income, employment, and
labor force conditions were explored.
• The annual poverty rate was examined as a representation of income and income distribution.
In the national-level model, the coefficient was weakly negative (i.e., opposite in sign to the
expected direction of association) and not nearly significant (p = 0.76).
• The annual average unemployment rate was examined as an alternative to the highest monthly
unemployment rate. As expected, it was not as strongly associated with food insecurity as the
highest monthly unemployment rate. Food-insecure conditions tend to be based on the worst
economic conditions during the year, which is better captured by the highest monthly unem-
ployment rate.
• The highest monthly unemployment rate for workers with less than a high school education
might better represent labor market conditions faced by less educated and less skilled workers,
who are more likely to be vulnerable to food insecurity. In fact, the association of this measure
with the prevalence of food insecurity was essentially identical to that of the highest monthly
unemployment rate for all workers, and the two unemployment measures were highly corre-
lated over the 12-year period (Pearson correlation = 0.997). We elected to use the more compre-
hensive measure.
• The annual average labor force participation rate for persons age 16 and older was included in
an exploratory model to account for the fact that the unemployment rate does not directly repre-
sent discouraged workers who drop out of the labor force. The coefficient for the labor-force
participation variable was weak and not nearly significant, so the variable was not included in
the final models.
• Two measures of long-term unemployment were assessed. The number of workers unemployed
for more than 26 weeks was entered as a percent of total labor force and (in a separate analysis)
as a percent of unemployed. Neither measure contributed to overall model fit, and neither coef-
ficient approached statistical significance.
Households with discouraged workers who drop out of the labor force and workers who are unem-
ployed for long periods probably are more likely to be food insecure than otherwise similar house-
hold without such members. However, at the national level, those conditions are, apparently, so
strongly associated with the unemployment rate that the model accounts adequately for the effects of
those associated conditions on food insecurity.
Two alternative specifications of inflation and prices were also explored.
• The price of food at home relative to the prices of other goods and services was considered as
an alternative to the corresponding measure based on the price of all food. This measure did
not perform as well as the broader measure, possibly because even low-income households rely
in part on food away from home.
• A similar measure of the relative price of foods in the Thrifty Food Plan was calculated as
the cost of the Thrifty Food Plan in December of the survey year divided by the CPI-U for all
goods and services in the same month. Inflation in the cost of the Thrifty Food Plan is based
29
on changes in prices of more basic, less processed foods, which may more closely represent
the foods typically purchased by lower income households. Although this measure performed
somewhat better than the relative price of all food, we retained the more general measure out of
concern that discontinuities could occur in the Thrifty Food Plan-based measure periodically
when the methodology for calculating the cost of the Thrifty Food Plan is revised.
30

48167_err167

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

48167_err167

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

48167_err167

Uploaded by

Copyright:

Available Formats

United States Department of Agriculture

Mark Nord, Alisha Coleman-Jensen, and Christian Gregory

Economic Research Service

Access this report online:

Download the charts contained in this report:

Recommended citation format for this publication:

Cover image: Thinkstock.

About the Authors

Data and Measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

National Economic Factors in Food Insecurity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

Appendix A: Associations of National Economic Conditions

Appendix B: Measures of Household Demographic, Economic,

Appendix C: Alternative Measures of National Economic Conditions. . . . . . . . . . . . . . . . . . . 29

A report summary from the Economic Research Service June 2014

Prevalence of U.S. Food Insecurity Is

What Is the Issue?

What Did the Study Find?

• An increase of 1 percentage point in annual inflation, as measured by the Consumer Price

How Was the Study Conducted?

2 Prevalence rate, very low food security

Note: Shaded areas indicate recessions.

• Low average wages

Source: United States Department of Commerce, Bureau of Labor Statistics.

(National-level regression model)

FIy = a + b1*Unempy + b2*Inflationy + b3*Relative-price-of-foody + ey

(Combined household-national-level regression model)

FIh,y = a + BhXh,y + GyMy + eh + fy

National Economic Factors in Food Insecurity

Level 2012 8.30 2.07 101.82 14.71 14.51

Level 2005-07 average 5.07 3.15 97.45 10.98 11.02

Level 2009-10 average 9.95 .64 101.16 14.51 14.60

Household Characteristics Associated With Food Insecurity

Some college, no four-year degree -.0026 .020 .2965 .3006 .3024

Bachelor degree -.0346 <.001 .2220 .2265 .2322

Professional or advance degree -.0361 <.001 .1401 .1513 .1607

Metropolitan, not specifically identified -.0021 .098 .1428 .1430 .1412

Not in metropolitan area -.0155 <.001 .1685 .1686 .1651

National-level economic measures

Accounted for by changes in explanatory variables

Not explained by model* -.58 -.80

Adjusted R-squared1 .921 .940

Mean absolute error (percentage points) .341 .318

Largest overestimate (percentage points) .653 .596

Largest underestimate (percentage points) .953 .900

Highest monthly unemployment rate in year (percent) .150 .279 .492

CPI inflation, annual average (percent) .280 .308 .093

Dependent mean (prevalence of very low food security) 4.528

Standard error of model-predicted prevalence .443

Mean absolute error of model-predicted prevalence .307

Level in 2012 8.30 2.07 101.82 5.83 5.72

Level 2005-07 average 5.07 3.15 97.45 3.71 3.97

Change 2005-07 to 2012 3.23 -1.09 4.37 2.13 1.75

Level 2009-10 average 9.95 .64 101.16 3.38 5.53

1. Employed full time

Metropolitan/nonmetropolitan residence is represented by a set of three dummy variables with

Two alternative specifications of inflation and prices were also explored.

You might also like

FIy = a + b1Unempy + b2Inflationy + b3*Relative-price-of-foody + ey