Dynamics of the food stamp program as reported in the Survey of Income and Program Participation

              m
United States
Department of
Agriculture
Food and
Nutrition
Service
Office ot
Analysis and
Evaluation
Current Perspectives on Food Stamp Program Participation
Dynamics of the
Food Stamp Program as
Reported in the
Survey of Income and
Program Participation
r
United States Food and 3101 Park Center Drive
Department of Nutrition Second Floor
fj Agriculture Service Alexandria. VA 22302
Dynamics of the
Food Stamp Program as
Reported in the
Survey of Income and
Program Participation
Nancy R. Burstein
A product ot
Abt Associates Inc.
55 Wheeler Street
Cambridge, MA 02138
Under Subcontract to
Mathematica Policy Research, Inc.
January 1993
ACKNOWLEDGMENTS
The author gratefully acknowledges the following individuals:
• Katie Merrell. for her assistance with the programming;
• Alice Robbin and Martin David of SIPP ACCESS at the University of
Wisconsin-Madison, Alberto Martini of MPR, and David McMillen of the
Bureau of the Census, for their help in using and interpreting the SIPP;
• Pat Doyle of MPR, for the creation and documentation of an easily
accessible version of the SIPP Full Panel Research File;
• Christine Kissmer, of the Food and Nutrition Service, for her guidance in
relating the analysis to current policy concerns;
• Steven Carlson, also of the Food and Nutrition Service, for his many
excellent comments and suggestions:
• Jean Wood, for her skilled supervision of the research project: and
• William L. Hamilton, for his insightful contributions to all aspects of this
report.
FNS Contract Number: 53-3198-9-31
FNS Project Officer: Christine Kissmer
■ •
TABLE OF CONTENTS
Chapter Bags
ONE
TWO
THREE
FOUR
FIVE
EXECUTIVE SUMMARY ix
INTRODUCTION 1
CIRCUMSTANCES SURROUNDING FOOD STAMP
SPELL BEGINNINGS 5
Analysis of Trigger Events 5
Definition of Trigger Events and the Population at Risk 6
Population Subgroups 9
Overall Probability of Opening 12
Occurrence of Trigger Events: All Recipients IS
Occurrence of Trigger Events: Earners vs. Nonearners 19
Occurrence of Trigger Events: Education Subgroups 19
Occurrence of Trigger Events: Demographic Subgroups 24
Summary 32
DURATION OF RECEIPT 35
Length of Completed Spells for Individuals 35
Length of Completed Spells for Households 40
Summary 46
CIRCUMSTANCES SURROUNDING EXITS FROM THE FOOD
STAMP PROGRAM 47
Definition of Trigger Events 48
Overall Probability of Exit 51
Occurrence of Trigger Events: All Recipients 53
Occurrence of Trigger Events: Earners vs. Nonearners 55
Occurrence of Trigger Events: Education Subgroups 58
Occurrence of Trigger Events: Demographic Subgroups 58
Recidivism 63
Summary 69
PATTERNS OF FOOD STAMP RECEIPT 73
All Recipients 74
Earners vs. Nonearners 74
Education Subgroups 77
Demographic Subgroups , 79
Summary 84
REFERENCES 85
iii
APPENDIX A The Data 87
APPENDIX B Details of Specification of Trigger Events 97
APPENDIX C Methodological Issues in Estimating the Distribution and Mean of
Completed Spell Lengths 105
APPENDDC D The Welfare History Topical Module 119
APPENDDC E Distribution of Length of Completed Spells for Subgroups of Individuals ... 125
APPENDDC F Distribution of Length of Completed Spells for Subgroups of Households ... 137
APPENDDC G Sources of Differences Between Individual-Level and Household-Level
Distributions 147
LIST OF TITLES IN THIS SERIES 157
iv
LIST OF EXHIBITS
Page
Exhibit 1.1 Analysis Samples 3
Exhibit II. 1 Definitions of Recipient Subgroups 11
Exhibit II.2 Overall Probability of Entering the Food Stamp Program Between
Two Consecutive Four-Month Periods 13
Exhibit II.3 Occurrence of Trigger Events for Openings: All Individuals 16
Exhibit II.4 Occurrence of Trigger Events for Openings: Individuals in
Households with Earnings in Baseline Wave 20
Exhibit II.5 Occurrence of Trigger Events for Openings: Individuals in
Households with No Earnings in Baseline Wave 21
Exhibit II.6 Occurrence of Trigger Events for Openings: Individuals in
Households Containing High School Graduates in Baseline Wave 22
Exhibit II.7 Occurrence of Trigger Events for Openings: Individuals in
Households Containing Only High School Dropouts in Baseline Wave 23
Exhibit II.8 Occurrence of Trigger Events for Openings: Abie-Bodied,
Childless Individuals 25
Exhibit II.9 Occurrence of Trigger Events for Openings: Aged and Disabled
Individuals 26
Exhibit II. 10 Occurrence of Trigger Events for Openings: One Adult with
Children 27
Exhibit II. 11 Occurrence of Trigger Events for Openings: Multiple Adults
with Children 28
Exhibit II. 12 Occurrence of Trigger Events for Openings: Children Living
with One Adult 19
Exhibit II. 13 Occurrence of Trigger Events for Openings: Children Living
with Multiple Adults 30
Exhibit n. 14 Distribution of Trigger Events for Openings: All Subgroups 33
Exhibit III. 1 Distribution of Length of Completed Spells: All Individuals 36
Exhibit III.2 Length of Food Stamp Spells for Subgroups of Individuals 37
Exhibit m.3 Distribution of Length of Completed Spells: AH Households 42
LIST OF EXHIBITS
(continued)
Page
Exhibit III.4 Distribution of Lengths of Spells for Households and Individuals 44
Exhibit III.5 Length of Food Stamp Spells for Subgroups of Households 45
Exhibit IV. 1 Overall Probability of Exiting from the Food Stamp Program
Between Two Consecutive Four-Month Periods 52
Exhibit IV.2 Occurrence of Trigger Events for Closures: All Recipients 54
Exhibit IV.3 Occurrence of Trigger Events for Closures: Earners 56
Exhibit IV.4 Occurrence of Trigger Events for Closures: Nonearners 57
Exhibit IV.5 Occurrence of Trigger Events for Closures: High School Graduates 59
Exhibit IV.6 Occurrence of Trigger Events for Closures: High School Dropouts 60
Exhibit IV.7 Occurrence of Trigger Events for Closures: Abie-Bodied,
Childless Adults 61
Exhibit IV.8 Occurrence of Trigger Events for Closures: Aged and Disabled 62
Exhibit IV.9 Occurrence of Trigger Events for Closures: One Adult with
Children 64
Exhibit IV. 10 Occurrence of Trigger Events for Closures: Multiple Adults
with Children 65
Exhibit IV. 11 Occurrence of Trigger Events for Closures: Children with One
Adult 66
Exhibit IV. 12 Occurrence of Trigger Events for Closures: Children with
Multiple Adults 67
Exhibit IV.13 Recidivism 68
Exhibit IV.14 Distribution of Trigger Events for Closings: All Subgroups 70
Exhibit V.l Patterns of Food Stamp Participation: All Individuals 75
Exhibit V.2 Patterns of Food Stamp Participation for Earners and Nonearners 76
Exhibit V.3 Patterns of Food Stamp Participation by Education 78
VI
LIST OF EXHIBITS
(continued)
Page
Exhibit V.4 Patterns of Food Stamp Participation for Members of
Childless Households 80
Exhibit V.5 Patterns of Food Stamp Participation for Adults With and
Without Children 82
Exhibit V.6 Patterns of Food Stamp Participation for Children 83
Exhibit A.l Summary of Food Stamp Program Participation in 1984 and 1985 95
Exhibit B.l Relationship Between Decrease in Household Income and
Probability of Entering Food Stamp Program Between Two
Consecutive Four-Month Periods 100
Exhibit B.2 Relationship Between Increase in Household Income and
Probability of Entering Food Stamp Program Between Two
Consecutive Four-Month Periods 102
Exhibit C. 1 Effects on Estimated Distribution of Completed Spell Lengths
of Excluding Known Recidivists 110
Exhibit C.2 Use of Right- and Left-Censored Spells to Calculate Mean
Spell Length 116
Exhibit D.l Food Stamp History Section of Fifth Topical Module 122
Exhibit D.2 Length of First Completed Food Stamp Spell, As Reported
in Fifth Topical Module 124
Exhibit E. 1 Distribution of Length of Completed Spells for Individuals
in Households with Earnings in First Month of Receipt 127
Exhibit E.2 Distribution of Length of Completed Spells for Individuals
in Households with No Earnings in First Month of Receipt 128
Exhibit E.3 Distribution of Length of Completed Spells for Individuals
in Households Containing High School Graduates in First
Month of Receipt 129
Exhibit E.4 Distribution of Length of Completed Spells for Individuals
in Households Containing High School Dropouts Only in First
Month of Receipt 130
Exhibit E.5 Distribution of Length of Completed Spells for Individuals
Who Are Abie-Bodied and Childless in First Month of Receipt 131
vii
LIST OF EXHIBITS
(continued)
Page
Exhibit E.6 Distribution of Length of Completed Spells for Individuals
Who Are Aged or Disabled in First Month of Receipt 132
Exhibit E.7 Distribution of Length of Completed Spells for Adults Living
With Children But No Other Adults in First Month of Receipt 133
Exhibit E.8 Distribution of Length of Completed Spells for Adults Living
With Children and Other Adults in First Month of Receipt 134
Exhibit E.9 Distribution of Length of Completed Spells for Children Living
With One Adult in First Month of Receipt 135
Exhibit E. 10 Distribution of Length of Completed Spells for Children Living
With More Than One Adult in First Month of Receipt 136
Exhibit F. 1 Distribution of Length of Completed Spells for Households
Containing Earners in First Month of Receipt 139
Exhibit F.2 Distribution of Length of Completed Spells for Households
Containing No Earners in First Month of Receipt 140
Exhibit F.3 Distribution of Length of Completed Spells for Households
Containing High School Graduates in First Month of Receipt 141
Exhibit F.4 Distribution of Length of Completed Spells for Households
Containing High School Dropouts Only in First Month of Receipt 142
Exhibit F.5 Distribution of Length of Completed Spells for Households
Consisting of Abie-Bodied Adults Only in First Month of Receipt 143
Exhibit F.6 Distribution of Length of Completed Spells for Households
Consisting of Aged and Disabled Individuals in First Month
of Receipt 144
Exhibit F.7 Distribution of Length of Completed Spells for Households
Consisting of One Adult and Children in First Month of Receipt 145
Exhibit F.8 Distribution of Length of Completed Spells for Households
Consisting of Multiple Adults and Children in First Month of Receipt 146
Exhibit G. 1 Implications of Change in Food Stamp Receipt Status of
Individuals and Longitudinal Households on Relative Spell Lengths 150
Exhibit G.2 Source of Differenc s Between Individual and Household
Level Spell Lengths 153
V1I1
EXECUTIVE SUMMARY
Food stamp administrators have an ongoing need for information about what kinds of
people participate in the Food Stamp Program, what conditions motivate them to apply for
benefits, how long they will participate, and what circumstances allow them to become
independent of assistance. Such knowledge is important not only in establishing budgets and
staffing levels, but also in designing policies to help food stamp recipients achieve self-sufficiency.
The analysis reported here is intended to contribute to the growing body of research on
the dynamics of food stamp participation. The data source is me Survey of Income and Program
Participation (SIPP), a national longitudinal survey. The SIPP collects ■cH ly data on a sample
of households over a period of nearly three years, through interviews conducted at four-month
intervals. The present research uses the 1984 SIPP panel, which covers a period from late 1983
to early 1986. The analysis uses respondents' reports of whether they received food stamps
during each four-month interview interval, together with selected demographic characteristics
of individuals and their households.
Highlights
People that enter the Program tend to receive food stamps for
relatively brief periods. Of all recipients that enter the Food Stamp
Program, half leave the program in six months or less and two-thirds
within one year. Averaging in some people who stay for very long spells,
the mean length of time that people receive food stamps is somewhat less
than two years.
Many people stop receiving food stamps for a period and then return
to the program. Somewhat more than one-third of all recipients who
stopped receiving food stamps began receiving them again within one
year.
Earned income is a dominant factor in participation patterns. Most
new food stamp households had some earnings shortly before entering the
program. A decline in a household member's earnings is the most
common event associated with beginning a food stamp spell, and an
increase in earnings most often accompanies the end of the spell.
Households that have earnings when they begin receiving food stamps are
able to leave the program more quickly. Households that have earnings
when they leave the program are less likely to return.
The food stamp recipient population is made up of groups with quite
distinct participation patterns.
— Most new food stamp recipients are in households that contain
at least two adults and at least one child. Participation patterns
for the food stamp population as a whole (cited above) largely
reflect this group's experiences, because it includes 71 percent of
all new recipients.
One-adult households with children show the most persistent
dependency patterns. This group, accounting for 14 percent of
new recipients, has the longest food stamp spells and the highest
recidivism rate.
Able-bodied, childless adults have the shortest spells of food
stamp participation and among the lowest recidivism rates.
This group is especially likely to begin participating after a drop
in earnings and to stop after an earnings gain. Fewer than one in
ten new recipients are in this group.
The aged and disabled have relatively long food stamp spells,
but one. they leave the program they ire least likely to return.
This group accounts for just seven percent of new food stamp
recipients.
Among people not receiving food stamps, children and high school
dropouts are especially likely to participate. Children are more than
twice as likely to start receiving food stamps as able-bodied childless
adults, and four times as likely as elderly and disabled childless adults.
Members of households with no high school graduates are nearly three
times as likely to begin receiving food stamps as people with at least one
high school graduate (or equivalency degree) in the household.
Trigger Events for Food Stamp Spells
Why do people enter the Food Stamp Program? One way to address this question is to
examine changes in household circumstances that occur just before people begin receiving food
stamps. This approach is not definitive, however. For example, a household may gain a new
infant and shortly afterward begin receiving food stamps, but one cannot be certain that the new
arrival, rather than some other factor, caused the family to apply for assistance. Nonetheless,
this approach has proven useful in studying the onset of dependency on food stamps and Aid to
Families with Dependent Children (AFDC).
Three kinds of events are hypothesized as "triggers" for a spell of food stamp
participation:
• Loss of household income. A household is considered to have lost
income if its total income for a four-month SIPP reporting period has
declined at least $400 from the prior period. A loss may result from a
decline in earnings or unearned income for one or more household
members, or from the departure of a household member with income.
• Increase in needs. A household is considered to have increased needs if
it gains a member who has no income. The new member may be an
infant, normally representing a new birth, or may be any other person
added to the household.
• New receipt of cash assistance. A household might apply for food
stamps not because its circumstances changed, but because it obtained new
information about the program or about the household's possible
eligibility. Because a new AFDC or General Assistance recipient might
be given such information, the beginning of such an assistance spell
without any reported loss of income or increase in needs is a potential
trigger event.
Overall, 82 percent of all individuals who began a food stamp spell experienced one or
more of the three kinds of trigger events. The frequency of the events is summarized in
Exhibit 1.
A sharp decline in earnings was by far the most common event. This occurred for
S3 percent of all persons beginning a food stamp spell. Another 18 percent lost income in some
XI
Exhibit 1
Incidence of Trigger Events for New Food
Stamp Recipients
Earnings decrease for household member
Unemployment Insurance benefits ended for household member
Other unearned income decrease for household member
Departure of member with earnings
Departure of member with other income
New infant in household
Other new household member without income
Begin cash assistance spell, no income loss or new member
]
]
+ +
10 20 30 40 SO
Percent of Spells with Trigger Event
60
Exhibit 2
Median Food Stamp Spell Length for
Subgroups
AD recipients
Earners
No earnings
Able-bodied, childless
Multiple adults with children
Aged and disabled
One adult with children
., -<$*?$&•
'
H 1 h
4 6 3
Months
———. .
-
10 12
xii
other way, most commonly through a reduction in unearned income or the departure of a
household member with earnings.
Fewer new food stamp recipients had experienced a recent increase in needs. Ten
percent had a new infant, and 8 percent had seen some other person without income added to
the household. Five percent of the new food stamp recipients did not have an observed loss in
income or increase in needs, but had recently begun receiving cash assistance.
This general characterization applies well to households with two or more adults and at
least one child. Other subgroups show some interesting differences in trigger events, however.
• Most new food stamp recipients (79 percent) were in households with
some earnings in the period before entering the program. Among these
people, nearly two-thirds experienced a decline in earnings just before
getting food stamps.
• A recent decline in unearned income was relatively common among
new recipients in households with no earnings during the pre-food
stamp period, with 23 percent experiencing this event. About 9 percent
of the new recipients had just begun receiving cash assistance. Overall,
however, trigger events were found for only 54 percent of those without
earnings in the pre-food stamp period.
• Among households made up entirely of aged or disabled adults, only 50
percent experienced any of the trigger events. Many of these
households are presumably responding to factors that are either not
measured in SIPP or occurred before the 8-month time frame considered
here.
• Single-adult families with children were the group most likely to have
a new infant in the household, with this event occuring for 17 percent
of the recipients. Even in this group, however, a decline in earnings
occurred for more than half of the new recipients.
Trigger events do not automatically lead to food stamp participation. Among the
population examined here (individuals with incomes below 300 percent of the poverty line), just
three percent of those who experienced a trigger event began receiving food stamps shortly
therafter. Some groups seem particularly vulnerable, however. Members of households with
Xlll
■81
no high school graduates, one-adult households with children, and households with no
earnings were more likely to begin receiving food stamps after a trigger event. These
groups may be living closer to the financial margin and be less able to cope with the strain
imposed by the trigger event.
Duration of Food Stamp Spells
Once individuals begin receiving food stamps, how long do they participate? We address
this question by examining the number of consecutive months1 of food stamp receipt reported
in the SIPP.
The median food stamp spell in the SIPP data is six months long -- that is, half of all
new recipients stop receiving food stamps in six months or less. Two-thirds of the spells end
within one year, while one-fifth last more than two years. A mean spell length cannot be
calculated directly from the SIPP data because the time frame is too short to observe the longest
spells in their entirety. Based on the available data, however, the mean spell length is estimated
at 22 months.
Different subgroups participate for dramatically different lengths of time, as illustrated
in Exhibit 2. Among the striking patterns:
•
•
Individuals in households that have some earnings when they begin
receiving food stamps have comparatively short spells. Their median
spell is just five months, and their mean spell is estimated at 14 months.
Households with no earnings at the time they enter the program
receive food stamps for more than twice as long as those with
earnings. Their median is about 10 months, and the mean stay on the
program is 30 months.
One-adult families with children stay on food stamps the longest. The
median spell for these new recipients is 11 months, while the mean is 38
months.
1 Certain analytic adjustments are made to the data as reported in the SIPP. In particular,
one-month gaps in the reported food stamp receipt are assumed to be reporting error, and it is
assumed that the household participated in the missing month.
xiv
• Able-bodied, childless adults have the shortest spells. Nearly half leave
the program within four months. The median is 5 months and the mean
stay is under 14 months.
These patterns, which reflect the diversity of the food stamp population, have important
implications for initiatives aimed at helping recipients attain self-sufficiency, such as employment
and training programs. For example, most recipients who begin with earnings will leave in a
very few months; a cost-effective program for these people would have to operate quickly and
be relatively inexpensive. In contrast, a program aimed at single-adult families with children
could operate over a longer period at a higher cost and still potentially be cost-effective.
Trigger Events for Food Stamp Closures
Why do people leave the Food Stamp Program? To address this question, we again
consider trigger events-that is, changes in peoples' household circumstances that occur just
before they stop receiving food stamps. The prevalence of these trigger events is summarized
in Exhibit 3.
• Increased earnings of household members is the single most common
trigger event. An earned income increase of $400 or more between two
four-month periods was reported for 57 percent of the recipients whose
cases closed.
• In comparison, other trigger events were rarely associated with food
stamp closures. The departure of a household member without income,
which reduces the family's need, occurred for about 12 percent of the
individuals leaving food stamps. Increases in unearned income occurred
for 11 percent. Only occasionally does a closure occur after a new person
with income enters the household (5 percent). Death, institutionalization,
emigration, or entry into the armed services (events which remove the
individual from the sample as well as from the Food Stamp Program)
accounted for about 4 percent of program exits.
These patterns generally characterize the experiences of multiple-adult households with
children and of able-bodied childless adults. Other subgroups show different patterns, however:
xv
Exhibit 3
Incidence of Trigger Events for
Individuals Ending Food Stamp Spell
Earnings increase for household member
Unearned income increase for household member
New household member with earnings
New household member with unearned income
Departure of household member without income
Death, institutionallzation, etc.
+ + +
10 20 30 40 50
Pareent of Spall with Triggar Event
60
Exhibit 4
Recidivism to Food Stamps
Within One Year
Afl recipients
Earners
No timings
Aged and disabled
Able-bodied, childless
Multiple adults with children
One adult with children
%•■%«*£¥c*.'-:
■.**> 1
»?;&-
%8*#*i ■ Pf* i&i:
H—I—I—h -H 1—I—I—I
0 5 10 15 20 25 30 35 40 45 50
Percent Re-Opening In One Year
xvi
Aged and disabled persons are the only group for which most case
closures are not accompanied by an increase in earnings. This group's
closures are typically associated with death, institutionalization and related
events (26 percent), or with an increase in unearned income such as Social
Security (24 percent).
For one-adult households with children, 12 percent of the closures
followed the entry of a new household member with earnings. This
was much higher than the rate for any other group, though still much
lower than the frequency of increased earnings.
Overall, 81 percent of the individuals whose food stamp spells were observed to end in
the SIPP data experienced one or more of these trigger events. This is about the same as the
pattern seen for spell beginnings. As with spell beginnings, many food stamp spells ended with
no observed trigger event, and many trigger events occurred to food stamp recipients who did
not immediately terminate.
Nearly all of the trigger events were more likely to lead to a program exit for
recipients in households with earnings than for recipients without earned income. Those
without earned income, who are presumably farther from self-sufficiency, may require larger
changes to be able to leave the program.
Recidivism
After people stop receiving food stamps, how many return to the program and how
quickly? The data examined here provide information on new spells that began within 16
months of a closure.
More than one-third of all recipients who stopped receiving food stamps (38 percent)
reported receiving benefits again within one year. Twelve percent reported new benefits
within four months, and 44 percent in 16 months.
Recidivism rates differ somewhat across subgroups. The aged and disabled are least
likely to return to the rolls, while one-adult households with children are most likely to do
so (see Exhibit 4). Households that have earnings when they end a food stamp spell are less
xvii
likely to reopen than those without earnings, but the difference is not so dramatic as some other
earnings/non-earnings comparisons.
Overview
From the preceding findings we can draw a picture of the most common type of new
food stamp recipient. This recipient is part of a household that includes at least two adults as
well as one or more children. The household had earnings before applying for food stamps, and
applied for food stamps after those earnings declined sharply. The individual receives food
stamps for six months, at which time an increase in household earnings occurs and the household
leaves the program. The individual does not receive food stamps again for at least a year.
The food stamp recipient population is not monolithic, however, and three other
important recipient types can be identified. One-adult households with children show the
strongest pattern of prolonged and repeated dependency. Childless adult households tend to
leave the program quickly and not return. Aged and disabled recipients, with long spells and
low recidivism, are the only group for which movement on and off the program has little to do
with fluctuations in earned income. These distinctive subgroups establish a complex
environment for the formulation of food stamp policy.
xviii
Current Perspectives on Food Stamp Program Participation
Recent Titles in this Series:
Dynamics of the Food Stamp Program as
Reported in the Survey of Income and Program Participation
. 993 Nancy R Bu'Stem
Food Stamp Program Participation Rates:
January 1989
July '992 Carole Tnppe and Pat Doyle
Trends in Food Stamp Program
Participation Rates: 1976 to 1990
i '992) Carole Tnppe. Pat Doyle and Andrew As"er
Food Stamp Program Participation Rates: January 1988
'February 1992. Carole Tnppe and Pat Doyle
Participation in the Food Stamp Program:
A Multivariate Analysis
" :n'992i Alberto Martini
(For a complete list of titles in this series, see page 157.)
KVX
CHAPTER ONE
INTRODUCTION
The past decade has seen the emergence of a growing body of research on the dynamics
of participation in assistance programs in general, and in the Food Stamp Program in particular.
An important theme of this research is that food stamp recipients form a heterogenous population
with widely varying patterns of participation. An understanding of these patterns is essential for
developing policies that will enable recipients to achieve economic self-sufficiency.
Four research questions of particular interest in this regard are:
• What circumstances lead people to enter the Food Stamp Program?
• How long do households and individuals tend to receive food stamps?
• What circumstances lead people to leave the program?
• How do participation patterns vary by specific demographic characteristics
(e.g., age, education, household composition, attachment to the labor
force)?
These questions have important policy implications. If many recipients of a particular type
normally exit the program after only a few months of food stamps, then it is probably not
efficient to enroll them in employment and training programs. Conversely, it is valuable to
know what types of recipients stay on the rolls for a year or more, and whether their eventual
exits are associated with events that could be influenced by program policy.
These same questions were addressed in a report by Burstein and Visher (1989). That
report used two nationally representative data sources: an administrative data base which
covered a sample of food stamp cases receiving benefits between October 1980 and December
1983; and an extract from the Panel Study of Income Dynamics (PSID), consisting of annual
data on a sample of households from 1973 to 1983. These two data bases had complementary
advantages and shortcomings. The administrative data measured participation on a monthly
basis, which is the appropriate time unit for analyzing the dynamics of a program that pays
monthly benefits. Furthermore, these data were free from recall error (although like most data,
1
they were subject to transcription error). On the other hand, the administrative data pertain only
to households receiving food stamps. Hence the circumstances of households in the months
immediately prior to entry or subsequent to exit could not be observed.
The PSID, in contrast, collects data on recipients and nonrecipients alike. Its primary
disadvantages are that information is available only on an annual basis;1 and that reported
receipt of food stamps is likely to understate actual receipt.
Analyses presented in this report use data from the 1984 panel of the Survey of Income
and Program Participation (SIPP). The SIPP has features in common with both of the data bases
mentioned above.2 Each panel collects monthly data on a sample of households over a period
of nearly three years, in this case running from the latter part of 1983 to the early part of 1986,
interviewing all members aged IS and older every four months. Thus, these data support both
subannual analysis of food stamp receipt and investigations of circumstances surrounding Food
Stamp Program exits and entrances. The disadvantages of the SIPP-which are inherent in this
type of data-are that the time period covered is too short to observe households' participation
for more than two or three years; that the number of food stamp recipients in the sample is
limited to a few thousand; and that the data are subject to some degree of recall error and
systematic underreporting. Despite these negative characteristics the SIPP data are of great
value in adding to our understanding of the dynamics of participation in the Food Stamp
Program.
In the chapters that follow, we present answers to each of the above research questions
based on households' responses to this survey. As shown in Exhibit 1.1, the population
examined varies in a fundamental way among the analyses. For studying circumstances leading
people to enter the Food Stamp Program, the sample consists of poor and near-poor
'Recent waves of PSID data have collected more detailed monthly information. No attempt
was made to use these monthly data because they were only available for the last year or two
of the extract, and because recall error was expected to be a major problem for monthly data
collected from annual retrospectives.
3A detailed description of the SIPP data and the extracts used in this report may be found
in Appendix A.
Exhibit 1.1
ANALYSIS SAMPLES
Research Question Conceptual Sample
What circumstances lead people to
enter the Food Stamp Program?
How long do households and
individuals tend to receive food stamps?
What circumstances lead people to
leave the program?
Poor and near-poor non-recipients
Households and individuals
beginning food stamp spells
Current recipients
nonrecipients. For determining the length of time households and individuals tend to receive
food stamps, the sample consists of new entrants during the observation period. Those who
were already receiving food stamps at the time the survey began are excluded (unless they left
and reentered the piogram). This part of the analysis thus addresses the question, "Of the next
100 persons who walk into a food stamp office, how many will be on the Program for one
month, two months, three months, and so on?" Finally, the analysis of circumstances leading
people to leave the Food Stamp Program focuses on ongoing food stamp recipients, including
those who were receiving benefits at the time the survey began.
CHAPTER TWO
CIRCUMSTANCES SURROUNDING FOOD STAMP SPELL BEGINNINGS
This chapter addresses the question of what circumstances are associated with people
starting to receive food stamps. While loss or decrease of earnings is by far the most common
occurrence, there turn out to be marked variations in patterns from one subgroup of individuals
to another, depending on labor force status, education, and household composition.
Analysis of Trigger Events
In their seminal work on the dynamics of AFDC receipt, Bane and ELIwood (1983) used
the PSID to explore the circumstances that lead families to enter the AFDC program. Their
approach was to examine all households that began a spell of AFDC receipt, and determine how
many had recently experienced a marital dissolution, loss of earnings, and other "trigger events";
that is, changes in household circumstances that could be expected to lead to a spell beginning.
They thus calculated the probability that households beginning a spell of AFDC experienced a
trigger event.
This dynamic approach, which links changes in household circumstances with changes
in recipiency status, was a step forward from earlier work which simply related current receipt
to current household circumstances. The underlying presumption is that a household that
experiences a major change (e.g., a divorce) will either maintain its independence by some
adaptation, or else require welfare almost at once.
While these conditional probabilities provide useful information, their interpretation is
enhanced if they can be compared with the corresponding conditional probabilities for eligible
households that did not enter the Food Stamp Program. When we compare the percentage of
individuals experiencing a trigger event among those who enter the Food Stamp Program to the
percentage of people experiencing the same event among those who did rjoi begin receiving food
stamps, we learn to what extent the trigger event is associated with an entry.
Another way to gauge the importance of the hypothesized trigger event is to calculate the
probability of beginning to receive food stamps conditional on the event occurring. Suppose,
for example, that about 2 percent of all individuals not receiving food stamps in one period begin
a spell of food stamps in the next period. If the proportion of individuals beginning a spell is
much higher than 2 percent for people who have experienced a particular event, then we can
identify the event as a trigger.
It is tempting to interpret trigger events as causes of food stamp beginnings. In general,
this interpretation is not justified. By a cause, we mean a factor which, if it alone were altered,
would change the outcome. But the events precipitating a successful food stamp application are
likely to be a series rather than a single occurrence. For example, a household head may suffer
a work-related injury that causes him or her to lose his job; collect unemployment insurance for
some months; and then apply for food stamps. It is probably a meaningless question whether
the spell of food stamp receipt was "caused" by the injury, the job loss, or the exhaustion of
unemployment benefits. For this reason, it is appropriate to interpret the association of trigger
events with food stamp spell beginnings as descriptive rather than causal.
Definition of Trigger Events and the Population at Risk
The events that will lead to a food stamp spell beginning are of three general types.
First, an individual may have suffered a loss of household income. The lost income may be of
various types, e.g. wages, unemployment insurance benefits, or other unearned income. An
individual may lose income through a decline in his or her own personal income, through
departure from the household of the person who had the income, or through a decrease in
income to other people who are still in the household. A household is defined simply as a group
of people living at one address at a given point in time. For convenience, we say that an earner
has departed from an individual's household whenever it is true that they no ionger live together;
but in fact, it may be the individual who has moved out while the earner stayeo behind. Death
of a household member with income is included as one form of a departure.
The second type of event that could lead to a food stamp spell beginning is an increase
in needs. The instances that we analyze here are the birth of a baby (or to be precise, the
addition of an infant to the household) and the addition of other people to the household who do
not have any income of their own. One can imagine other increases in needs that could lead to
food stamp spell beginnings-such as rent increases, price increases, and medical emergencies--
but the SIPP data are not suited for measuring these.
Yet a third type of trigger event is a gain of information. Individuals may be
circumstantially eligible for food stamps for months or years before applying. Some begin to
receive some form of cash assistance such as AFDC or SSI, and then begin to receive food
stamps at about the same time. It is a plausible inference that these people have received
information or encouragement about applying for food stamps from the administrators of the
cash assistance programs. But again, individuals may gain information about the Food Stamp
Program in ways that are not captured by the SIPP--e.g., through networks of family and
friends, or through outreach programs by the agency or by local advocacy groups.
There are dangers in identifying potential trigger events either too broadly or too
narrowly. A broad definition (e.g., an income loss of any size occurring any time within the
past three years) will be associated with a large number of spell beginnings. Yet the probability
of an opening for individuals experiencing this event may be no higher than the unconditional
probability of opening for all individuals. Such a definition would therefore not be useful.
Conversely, a very narrow definition (e.g., a major income loss within the past few
months) may be associated with a relatively high conditional probability of opening, in that a
relatively large proportion of people who experienced the event began to receive food stamps.
Yet the event may be so rare that it is associated with only a small percentage of all food stamp
spell beginnings. The operational definitions of trigger events must avoid both extremes.
A key decision in this regard was to focus on the four-month data collection period used
in the SIPP, known as a wave, rather than on the individual month, as the unit of analysis. This
decision was influenced by two factors. First, we have more confidence in the food stamp
recipiency data for four-month reference periods than for individual months.1 Second, it seems
'The reliability of the SIPP data is discussed in Appendix A.
plausible that the lags between changes in household circumstances and food stamp recipiency
would be on the order of several months, rather than a single month.
As a consequence, a food stamp opening is defined here as receipt of food stamps in
a four-month reporting period, or wave, when no food stamps were received in the
preceding wave. A person in the sample may contribute as many as five observations to this
analysis, corresponding to the possibilities of a food stamp opening in Waves 4, 5, 6, 7, and 8.'
A trigger event may have occurred either in the wave of opening or in the preceding wave.
Suppose, for example, that a person who loses a job in Wave 5 begins food stamp receipt in
Wave 6. Depending on whether the job was lost near the beginning or near the end of Wave
5, the major decrease in earned income may occur between Waves 4 and S, or alternatively
between Waves S and 6. Hence, a decrease in earnings in either of these time frames is consid-ered
to be a possible trigger for a food stamp opening in Wave 6. The minimum loss of income
between waves that is deemed to be a potential trigger event is $400, corresponding to a change
in income of $100 per month. The relationship between income losses f various sizes and the
probability of beginning a food stamp spell is discussed in detail in Appendix B.
Some individuals are so unlikely to have a food stamp opening that there is little or no
gain from including them in the analysis. Individuals that are already receiving food stamps in
a given wave clearly cannot begin to receive food stamps in the following wave. These person-waves
are therefore excluded from the analysis. In addition, it may reasonably be supposed that
individuals with relatively high household incomes have a sufficient financial cushion that even
a job loss or other major event is not likely to lead to a quick food stamp opening. Retaining
them in the sample would attenuate measured relationships for those households with a
significant probability of beginning a food stamp spell. We have therefore eliminated higher-income
households from the sample as follows. Baseline income is measured in the second prior
'Openings in Wave 3 and earlier cannot be analyzed because to do so would require
comparing household data from Wave 1. An idiosyncrasy of the SIPP is that data collection for
all four months in Wave 1 was based on household composition in Month 5 (the first month of
Wave 2), rather than on household composition in each month of Wave 1. The data are
therefore not comparable with those from other waves.
wave before the wave in which an opening could occur. (For example, for an opening in Wave
6, baseline income is measured in Wave 4.) If the baseline household income exceeds three
times the estimated poverty threshold, then we conclude that a food stamp opening two waves
later has a negligible probability of occurring.' The corresponding person-wave is then dropped
from the sample.
We find that in a given four-month period, nearly half of all persons who did not receive
food stamps live in households that have income over three times the poverty threshold. Less
than two in a thousand of these individuals begin to receive food stamps two waves later, and
they account for less than eight percent of food stamp openings. The next lowest group on the
income scale, those with income between two and three times the poverty line, contribute 13
percent of food stamp openings while comprising less than a quarter of the nonrecipient
population. We retain them in the analysis sample.
The presence of significant assets could also render it virtually impossible for a household
to enter the Food Stamp Program in the near future. The available data on assets are too limited
to use for constructing a cutoff for identifying ineligible households, however.
Population Subgroups
In addition to determining patterns of food stamp participation for the population at large,
it is of interest to see how these patterns vary among subgroups of the population. Some
dimensions on which important variations may occur are:
• presence or absence of earnings;
"The official poverty threshold measure is based on the family, rather than the household;
varies outside the continental United States and according to the presence nf elderly individuals;
and is recalculated for each calendar year. For current purposes, we have simply assigned to
each household month in the sample the average national value of the poverty threshold for
families that are the size of that household. (The time dimension was accommodated by using
the average of the published values of thresholds for 1984 and 1985.) By this rule, the annual
poverty thresholds assigned to households of size 1, 2, 3, 4, and 5, for example, were $5374,
$6880, $8425, $10,799, and $12,787, respectively. The monthly thresholds were these values
divided by 12.
9
• education level of household members who are not disabled or elderly;
• age and disability status; and
• presence or absence of multiple adults in households which contain
children.
Exhibit n. 1 displays the subgroups used in the analyses in this report.
The population has been partitioned in three independent ways.1 For each partition, the
operational definitions are shown both for individuals (the level of analysis used throughout the
report) and for households (a level of analysis used in Chapter Three only). Even at the
individual level, however, subgroup definitions are generally based on characteristics of the
household of which the individual is a member. This is done because we assume that welfare
dynamics for individuals are driven by household circumstances.
The first partition pertains to the presence or absence of earnings. Households are
classified according to whether or not they contain an earner. Individuals are classified
according to whether their household contains an earner. The time dimension in which the
presence of earnings is measured-e.g. current wave, preceding wave, current month-varies by
research question, and is noted each time subgroup results are presented.
The second partition pertains to the education of the members of the household who are
potentially in the labor force-that is, adults under the age of 60 who are not disabled. A
household that contains at least one such adult who has a high school diploma falls in the
category of high school graduates. If there are able-bodied, non-elderly adults present, but none
with a high school diploma, then the household falls in the category of high school dropouts.
The remaining households, in which there are no able-bodied adults under age 60, are excluded
from this partition. Individuals are again classified according to the household to which they
belong. Thus, a child in a household which includes a high school graduate is put in the
graduate subgroup, because the welfare dynamics for the child is determined in part by the
education of adult household members.
'Sample size did not permit that these partitions be interacted.
10
Exhibit IL1
DEFINITIONS OF RECIPIENT SUBGROUPS
Subgroup Households Individuals
Earners
Noneamers
Households with earnings
Households without earnings
Members of such households
Members of such households
High-school graduates
High-school dropouts
Households containing at least one
non-elderly, able-bodied adult with
a high school diploma
Households containing at least one
non-elderly, able-bodied adult, but
none with a high school diploma
Members of such households
Members of such households
Able-bodied
childless adults
Households containing no
children, elderly, or disabled
Members of such households
Elderly and disabled
childless adults
Children living
with one adult
Single adult living
with children
}
Households containing at least
one elderly or disabled individual,
not more than one able-bodied,
non-elderly adult, and no children
Households consisting of one
adult and one or more children {
Elderly and disabled members
of such households
Children living in such households
Adults living in such households
Children living with
more than one adult
Adults living with
other adults and
children
} Households consisting of multiple
adults and one or more children {
Children living in such households
Adults living in such households
11
The final partition pertains to the demographic composition of the household. Four
household types and six individual types have been defined. The first type of household consists
entirely of able-bodied, non-elderly, childless adults. The individuals in this subgroup are the
members of such households. The second type of household also contains no children, but
contains at least one elderly or disabled person. One able-bodied non-elderly adult may also be
present in such a household, e.g., the spouse of an elderly or disabled person. The individuals
in this subgroup are members of such households.
The remaining two household types are single-adult and multiple-adult households with
children. Four individual types have been identified corresponding to these, according to
whether the individual in question is a child or an adult living in such a household. These types
correspond approximately to one- and two-parent families. We have not used the more familiar
terms, however, because the Food Stamp Program, unlike the AFDC program, does not focus
on relationships by blood or marriage. An adult who is living with a dependent child is deemed
to have parental responsibility, although that adult may be the child's aunt, grandparent, or
stepparent. Furthermore, the marital status of adults, which is self-reported, may be ambiguous.
We assume that the dynamics of participation by households with children are determined more
by whether multiple adults are present than by their particular legal and biological relationships
to each other and to the children.
Overall Probability of Opening
Exhibit n.2 shows for the population as a whole and for the various subgroups the
probability that an individual who did not receive food stamps in a particular wave did receive
them in the subsequent wave. The subgroups are defined as of the baseline wave, that is, two
waves before the potential opening. A person is considered to be a member of a household with
earnings if he or she lived in a household with earnings at any time during that wave.
Educational and demographic classifications are determined as of the first month of the baseline
wave. This ensures that the subgroups are defined prior to the occurrence of the putative trigger
events.
12
Exhibit II.2
OVERALL PROBABILITY OF ENTERING THE FOOD STAMP PROGRAM
BETWEEN TWO CONSECUTIVE FOUR-MONTH PERIODS
Percent opening
Percent of in next four Percent of
population months openings
Earners 80.0 2.0 79.2
Nonearners 20.0 2.1 20.8
High school graduates 71.3 1.9 65.7
High school drop-outs 11.6 5.2 29.3
Able-bodied, childless 13.5 1.2 8.1
Elderly/disabled, childless 20.9 0.7 7.3
One adult living with children 2.4 3.6 4.3
Multiple adults living with children 32.9 2.2 36.0
Children living with one adult 3.6 5.5 9.8
Children living with multiple 26.7 2.6 34.5
adults
ALL INDIVIDUALS 100.0 2.0 100.0
Source: 1984 SD?P Panel (June 1983 to June 1986).
Unweighted sample size: 75,161 observations.
Notes: 1. This table includes only individuals whose household income is less than three
times the poverty threshold income in the baseline wave (i.e., two waves before
the food stamp opening).
2. The percentages shown pertain to Waves 3 through 8 combined.
3. For definitions of population subgroups, see Exhibit HI. High school graduate
and dropout subgroups do not sum to 100 percent of the population because
individuals in households containing only elderly or disabled adults are excluded.
13
The first column of the exhibit shows the distribution of the population among the
subgroups at the baseline wave. It is notable that only a few (6.0 percent) of these individuals
live in households consisting of a single adult with children. This is a consequence of the defini-tion
of the population at risk, namely, individuals in households with income under three times
the poverty line who are not currently receiving food stamps. Lower-incomt single-adult
households with children that are not already receiving food stamps are relatively rare.'
For the entire population, the probability of an opening is 2 percent. This varies little
by whether or not households had earnings in the baseline wave. Marked variations are seen
with regard to the other dimensions, however. Excluding those households in which the only
adults are elderly or disabled, individuals in households which contain a high school graduate
are about as likely to commence food stamp receipt as the general population; but those in
households that contain only high school dropouts are two and one half times as likely to do so.
The demographic subgroups also show substantial variation. The presence of children
in a household substantially increases the probability of a food stamp spell beginning: single
adults living with children are three times as likely to begin a spell as able-bodied childless
adults (3.6 versus 1.2 percent), and seven times as likely as elderly and disabled childless adults.
Furthermore, children living with one adult are twice as likely to start receiving food stamps as
children living with multiple adults (5.5 versus 2.6 percent).
The final column shows the percent of all food stamp openings coming from each
subgroup. Thus, for example, members of high school dropout households comprise only 11.6
percent of the population at risk, but because of their high entry rates account for 29.3 percent
of food stamp openings.
'Doyle (1990) calculated a 74.8 percent participation rate (August 1985) among eligible
households consisting of a single female adult with children. The numerator was based on the
Food Stamp Program Statistical Summary of Operations and the denominator on the 1984 and
1985 panels of the SIPP. The participation rate for this household type was substantially higher
than the rate for eligible households in general (59.4 percent).
14
Occurrence of Trigger Events: All Recipients
Exhibit II.3 shows the occurrence of trigger events to all individuals at risk cf an
opening, and the effects of the events on the chances of a food stamp opening occurring. As
can be seen from the final line of the exhibit, over half of the nonrecipient population
experienced a trigger event of one sort or another, and these individuals then had an opening rate
of 3 percent, compared with only 2 percent for the general nonrecipient population at risk.
Looking at it from the opposite perspective, 80 percent of those who began to receive food
stamps experienced one or more of the trigger events.
The first type of trigger event considered is losses of household income. These were
subdivided into six types:
loss or decrease of earnings to a household member;
loss or decrease of unemployment insurance benefits to a household
member;
loss or decrease of other unearned income to a household member;
departure of a household member who had earnings;
departure of a household member who had other income; and
miscellaneous.
For individuals who experienced a drop in household income of at least $400 in either the wave
in which the opening could have occurred or the preceding wave, it was first determined in
which wave the greatest income loss occurred, and then which component of income within that
wave showed the greatest loss. If no single component accounted for $400, the income loss was
classed as miscellaneous. Thus, the income loss types are mutually exclusive and collectively
exhaustive.
By far the most common trigger event is a decrease in earnings to household members.
This event occurred to 38 percent of individuals at risk of a food stamp opening, and accounts
for S3 percent of all food stamp spell beginnings. Yet it is only a moderately good predictor
of a food stamp spell beginning: the probability of an opening among individuals who
15
Exhibit n.3
OCCURRENCE OF TRIGGER EVENTS FOR OPENINGS:
ALL INDIVIDUALS
Event
Percent of all
individuals with
event
Conditional probability of;
opening | event event | opening
Household income decreased significantly,
primarily because of:
Decrease of earnings to household
member
Loss of unemployment insurance
benefits to household member
Decrease of other unearned income to
household member
Departure of member with earnings
Departure of member with other income
Miscellaneous
New household member without income
Infant
Other
Startup of cash assistance, with none of the
above events
ALL EVENTS
38.0
0.8
7.9
3.7
3.0
2.1
55.3
2.8
4.4
2.0
5.5
5.7
5.0
3.0
53.1
1.7
8.0
3.0 4.3 6.4
0.6 4.9 1.5
0.5 2.7 0.6
10.1
8.3
5.1
81.8
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 75,161 observations.
Notes: 1. This table includes only individuals whose household income is less than three times the
poverty threshold income in the baseline wave (i.e., two waves before the food stamp
opening).
2. The overall probability of an opening for all individuals is 2.0 percent.
3. Probability of opening | event: proportion of individuals experiencing the event who enter the
Food Stamp Program within one or two waves. Probability of event | opening: proportion of
individuals entering the Food Stamp Program who experienced the event one or two waves
previously.
16
experienced this event, 2.8 percent, is not dramatically greater than the probability of 2.0
percent for the population as a whole.1
In contrast, loss of unemployment insurance benefits is a rare event, affecting less than
1 percent of these individuals. Yet for those who experience it, the probability of an opening
is over 4 percent. It seems plausible that some households follow a path from a job loss to
receipt of unemployment benefits, and then to entrance into the Food Stamp Program when these
benefits expire.2
Approximately 8 percent of individuals experience a significant drop in other income, but
only 2 percent of these individuals then enter the Food Stamp Program. This is no higher than
the percentage of the entire population that does so.
Two other rare income-related events have relatively high probabilities of triggering a
food stamp spell: the departure of a household member who had been contributing earnings,
and the departure of a household member who had been contributing other income (including,
extremely rarely, unemployment benefits). These events occur to only 3 percent and 1 percent
of individuals, respectively; yet the individuals who experience these events have a 4 to 5
percent chance of beginning to receive food stamps.
An increase in household needs also may trigger a food stamp spell beginning. Four
percent of individuals experience the addition of an infant to their household in a given four-month
period, and 3 percent the addition of another person without income. Of those
experiencing one or both of these events, approximately 6 percent then enter the Food Stamp
Program. These two events have not been defined to be mutually exclusive with each other or
'It should be noted, however, that these statistics are a function of the cutoff that was chosen
to identify a significant loss of income. Choice of a higher cutoff--e.g., a decrease of $800--
would lead to this event occurring less frequently and accounting for fewer spell beginnings, but
predicting openings among individuals who experienced the event with more power.
It is not possible to determine from the SIPP data whether the loss of unemployment
insurance benefits is due to exhaustion of the benefit or some other cause. The event measured
here is simply a decrease in reported income from that source.
17
with income losses. Consequently, some individuals may have experienced both of these events,
and some may have experienced decreases in household income at the same time.
Finally, some individuals who experienced none of the above events began receiving
government transfer payments-Aid to Families with Dependent Children (AFDC) or other public
assistance, Supplemental Security Income (SSI), Social Security, or Unemployment Insurance.
As suggested above, the administrators of these programs may recommend that the household
apply for food stamps as well. It can be seen that this potential trigger event occurred to 2
percent of individuals, 5 percent of whom then began to receive food stamps.
The final line of the exhibit shows the combined effects of all trigger events. As noted
above, fifty-five percent of individuals experienced at least one of these events, and they
collectively had a 3 percent probability of commencing food stamp receipt. In all, 82 percent
of individuals who began to receive food stamps experienced one or more of these events.1
'Burstein and Visher (1989) obtained rather different results from their analysis of the Panel
Study of Income Dynamics (PSID). They found that departures of adults were associated with
nearly 40 percent of food stamp openings, while income losses were associated with only 31
percent of openings. The primary reason for the difference in findings is that Burstein and
Visher's hierarchical definition of trigger events was based on David Ellwood's research on the
AFDC program. Hence all changes which consisted of the departure of a household head or
spouse who had earnings or other income were classified as household composition changes.
The current analysis focuses on Food Stamp Program requirements, which do not depend on the
structure of the household. Departure of a household head or spouse with earnings is therefore
considered an income change. In contrast to the earlier study, if no associated income loss
occurs, departure of an adult from a household is not considered to be a trigger event at all.
An additional source of noncomparabUity is that the proportion of openings that are associated
with income losses is to some extent arbitrary, as it depends on the size of the income loss that
is chosen for a cutoff. The two studies used different cutoffs. Finally, the earlier analysis had
the advantage of a much longer time series to examine-11 years versus two and one-half-but
the disadvantage of only annual interviews. Hence both trigger events and receipt of food
stamps were defined more broadly in the time dimensions. For these reasons, the proportions
of food stamp openings that are associated with changes of various types cannot be compared
between the two reports.
18
Occurrence of Trigger Events: Earners vs. Nonearners
We turn now to an examination of trigger events for the population subgroups defined
earlier. Exhibits II.4 and n.5 indicate some significant differences in the patterns of trigger
events between earners and nonearners.
For individuals in households that had earnings two waves prior to the food stamp
opening, nearly three quarters of openings can be associated with a loss of earnings or departure
of an earner. In contrast, loss of an earner or an ongoing household member's earnings is
naturally a rare event for individuals in households initially without earners; it can occur only
if the household achieves a significant level of earnings in the wave after the baseline, and then
loses the earnings again in the following wave. Twelve percent of food stamp openings for
members of nonearner households are due to this sort of fluctuation.
Another striking feature of this pair of tables is the very high conditional probability of
opening for members of nonearner households that gain new infants or other persons without
income, or experience a startup of cash assistance in the absence of a measured chanp in
resources or needs. These probabilities are in the 11 to IS percent range-contrasted with only
4 to 5 percent for earner households. In fact, all of the conditional probabilities of openings are
greater for nonearner than for earner households, suggesting that they may have fewer resources
than earner households to avert a food stamp spell beginning when circumstances change for the
worse.
Only 54 percent of nonearners who begin to receive food stamps have experienced one
or more of the enumerated trigger events, compared with 90 percent for earners. There are
undoubtedly other changes occurring in nonearner households that these definitions (or possibly
the SIPP data) fail to capture. As noted earlier, these could be medical emergencies, local
agency outreach efforts, and so on.
Occuirence of Trigger Events: Education Subgroups
Variations among individuals by educational status of the adults in their households are
shown in Exhibits n.6 and II.7. Again, some substantial differences can be seen. Among
19
Exhibit II.4
OCCURRENCE OF TRIGGER EVENTS FOR OPENINGS:
INDIVIDUALS IN HOUSEHOLDS WITH EARNINGS IN BASELINE WAVE
Event
Percent of
subgroup
with event
Conditional probability of:
opening | event event | opening
Household income decreased significantly,
primarily because of:
Decrease of earnings to household 46.6
member
Loss of unemployment insurance 0.8
benefits to household member
Decrease of other unearned income to 4.8
household member
Departure of member with earnings
Departure of member with other income
Miscellaneous
New household member without income
Infant 4.4
Other 3.3
Startup of cash assistance, with none of the 2.2
above events
ALL EVENTS 61.7
2.8
4.1
1.9
64.3
1.6
4.6
3.8 4.2 7.9
0.5 5.1 1.2
0.5 2.4 0.6
5.0
5.0
3.7
2.9
11.1
8.2
4.2
89.0
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 59,088 observations.
Notes: 1. This table includes only individuals whose household income is less than three times the
poverty threshold income in the baseline wave (i.e., two waves before the food stamp
opening).
2. The overall probability of an opening for this subgroup is 2.0 percent.
3. Probability of opening \ event: proportion of individuals experiencing the event who enter the
Food Stamp Program within one or two waves. Probability of event | opening: proportion of
individuals entering the Food Stamp Program who experienced the event one or two waves
previously.
20
Exhibit n.5
OCCURRENCE OF TRIGGER EVENTS FOR OPENINGS:
INDIVIDUALS LW HOUSEHOLDS WITH NO EARNINGS IN BASELINE WAVE
Event
Percent of
subgroup
with event
Conditional probability of;
opening | event event | opening
Household income decreased significantly,
primarily because of:
Decrease of earnings to household
member
Loss of unemployment insurance
benefits to household member
Decrease of other unearned income to
household member
Departure of member with earnings
Departure of member with other income
Miscellaneous
New household member without income
Infant
Other
Startup of cash assistance, with none of the
above events
ALL EVENTS
3.6 6.2 10.7
0.8 5.9 2.1
20.5 2.2 21.0
0.2 11.2 1.0
1.2 4.5 2.4
0.2 5 0.7
0.9
1.6
1.4
29.4
14.6
11.2
13.5
3.9
6.4
8.7
8.7
54.4
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 16,073 observations.
Notes: 1. This table includes only individuals whose household income is less than three times the
poverty threshold income in the baseline wave (i.e., two waves before the food stamp
opening).
2. The overall probability of an opening for this subgroup is 2.1 percent.
3. Probability of opening | event: proportion of individuals experiencing the event who enter the
Food Stamp Program within 1 or 2 waves. Probability of event | opening: proportion of
individuals entering the Food Stamp Program who experienced the event one or two waves
previously.
21
Exhibit H.6
OCCURRENCE OF TRIGGER EVENTS FOR OPENINGS:
INDIVIDUALS IN HOUSEHOLDS CONTAINING HIGH SCHOOL GRADUATES
IN BASELINE WAVE
Event
Percent of
subgroup
with event
Conditional probability of:
opening | event event | opening
Household income decreased significantly,
primarily because of:
Decrease of earnings to household 44.6
member
Loss of unemployment insurance 0.9
benefits to household member
Decrease of other unearned income to 5.4
household member
2.4
2.0
2.5
58.0
1.0
7.3
Departure of member with earnings 3.7 3.3 6.6
Departure of member with other income 0.5 4.9 1.3
Miscellaneous 0.5 2.9 0.7
New household member without income
Infant
Other
Startup of cash assistance, with none of the
above events
ALL EVENTS
4.4
3.2
2.2
60.4
3.9
4.7
3.8
2.6
9.4
8.1
4.6
83.3
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 52,602 observations.
Notes: 1. This table includes only individuals whose household income is less than three times the
poverty threshold income in the baseline wave (i.e., two waves before the food stamp
opening).
2. The overall probability of an opening for this subgroup is 1.9 percent.
3. Probability of opening | event: proportion of individuals experiencing the event who enter the
Food Stamp Program within 1 or 2 waves. Probability of event | opening: proportion of
individuals entering the Food Stamp Program who experienced the event one or two waves
previously.
22
Exhibit n.7
OCCURRENCE OF TRIGGER EVENTS FOR OPENINGS:
INDIVIDUALS IN HOUSEHOLDS CONTAINING ONLY HIGH SCHOOL DROPOUTS
IN BASELINE WAVE
Event
Percent of
subgroup
with event
-Conditional probability of;
opening | event event | opening
Household income decreased significantly,
primarily because of:
Decrease of earnings to household
member
43.3 5.9 49.4
Loss of unemployment insurance
benefits to household member
0.9 20.7 3.5
Decrease of other unearned income to
household member
6.4 6.9 8.5
Departure of member with earnings 3.5 10.6 7.1
Departure of member with other income 0.9 10.1 1.8
Miscellaneous 0.4 2.9 0.2
New household member without income
Infant 4.6 14.0 12.5
Other 5.0 9.9 9.6
Startup of cash assistance, with none of the
above events
2.9 9.3 5.1
ALL EVENTS 61.8 7.1 84.4
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 8,396 observations.
Notes: 1. This table includes only individuals whose household income is less than three times the
poverty threshold income in the baseline wave (i.e., two waves before the food stamp
opening).
2. The overall probability of an opening for this subgroup is 5.2 percent.
3. Probability of opening | event: proportion of individuals experiencing the event who enter the
Food Stamp Program within one or two waves. Probability of event | opening: proportion of
individuals entering the Food Stamp Program who experienced the event one or two waves
previously.
23
individuals in high school graduate households who begin to receive food stamps, 58 percent
have experienced a significant loss of earnings to an ongoing household member. The corres-ponding
proportion for individuals in high school dropout households is only 49 percent. In the
dropout households, several other trigger events occur relatively more frequently to individuals
who begin to receive food stamps-e.g. loss of unemployment benefits, acquisition of a new
baby, acquisition of another household member without income.
The most striking contrast between members of the graduate and dropout households,
however, is in the conditional probability of opening given the occurrence of any trigger event:
only 2.6 percent for the former, but 7.1 percent for the latter. Dropout households may be
living nearer the financial margin, such that any shock is more likely to lead them to seek
assistance.
Occurrence of Trigger Events: Demographic Subgroups
Exhibits II. 8 through n. 13 show the occurrence of trigger events for the six demographic
subgroups. As shown in Exhibit n.8, able-bodied, childless individuals are relatively unlikely
to begin to receive food stamps, even if a trigger event occurs. Their pattern of trigger events
is similar to that of the population in general, except that loss of unearned income to a household
member is associated with a large number of openings.
For the aged and disabled, less than half of all openings can be associated with a trigger
event. The dynamics of food stamp participation for this subgroup clearly cannot be explained
simply in terms of changes in needs and resources measured in the SIPP. Furthermore, the
probability of an opening given a trigger event is only 1 percent. It thus appears that these
households are quite stable, and unlikely to begin receipt of food stamps if they are not already
receiving benefits. Loss of earnings or departure of an earner accounts for a quarter of all
openings for this subgroup; it should be recalled that one able-bodied adult may be present in
these households, e.g., as a spouse.
The threefold difference in the likelihood of beginning a food stamp spell between single
adults living with children and able-bodied, childless adults was previously remarked upon.
24
Exhibit H.8
OCCURRENCE OF TRIGGER EVENTS FOR OPENINGS:
ABLE-BODIED, CHILDLESS INDIVIDUALS
Event
Percent of
subgroup
with event
Conditional probability of;
opening | e\aanint | opening
Household income decreased significantly,
primarily because of:
Decrease of earnings to household member
Loss of unemployment insurance
benefits to household member
Decrease of other unearned income to
household member
37.4
0.9
5.2
2.0
0.0
2.4
62.4
0.0
10.4
Departure of member with earnings 3.6 1.6 4.9
Departure of member with other income 0.3 7.5 2.0
Miscellaneous 0.6 5.7 2.7
New household member without income
Infant 3.7 3.3 10.2
Other 4.4 3.1 11.3
Startup of cash assistance, with none of the
above events
2.2 1.2 2.4
ALL EVENTS 53.9 1.9 87.2
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 9,058 observations.
Notes: 1. This table includes only individuals whose household income is less than three times the
poverty threshold income in the baseline wave (i.e., two waves before the food stamp
opening).
2. The overall probability of an opening for this subgroup is 1.2 percent.
3. Probability of opening | event: proportion of individuals experiencing the event who enter the
Food Stamp Program within one or two waves. Probability of event | opening: proportion of
individuals entering the Food Stamp Program who experienced the event one or two waves
previously.
25
Exhibit n.9
OCCURRENCE OF TRIGGER EVENTS FOR OPENINGS:
AGED AND DISABLED INDIVIDUALS
Event
Percent of
subgroup
with event
Conditional probability of:
opening | event event | opening
Household income decreased significantly,
primarily because of:
Decrease of earnings to household
member
Loss of unemployment insurance
benefits to household member
Decrease of other unearned income to
household member
Departure of member with earnings
Departure of member with other income
Miscellaneous
New household member without income
Infant
10.5
0.3
18.3
1.5
0.0
0.5
22.4
0.0
14.2
0.5 3.5 2.7
1.1 0.6 1.0
0.4 1.5 0.9
0.3 4.3 1.6
Other 1.1 2.8 4.5
Startup of cash assistance, with none of the
above events
1.0 4.7 7.1
ALL EVENTS 32.9 1.0 49.8
Source: 1984 SD?P Panel (June 1983 to June 1986).
Unweighted sample size: 16,617 observations.
Notes: 1. This table includes only individuals whose household income is less than three times the
poverty threshold income in the baseline wave (i.e., two waves before the food stamp
opening).
2. The overall probability of an opening for this subgroup is 0.7 percent.
3. Probability of opening | event: proportion of individuals experiencing the event who enter the
Food Stamp Program within one or two waves. Probability of event | opening: proportion of
individuals entering the Food Stamp Program who experienced the event one or two waves
previously.
26
Exhibit n.10
OCCURRENCE OF TRIGGER EVENTS FOR OPENINGS:
ONE ADULT WITH CHILDREN
Event
Percent of
subgroup
with event
Conditional probability of;
opening | event event | opening
Household income decreased significantly,
primarily because of:
Decrease of earnings to household
member
Loss of unemployment insurance
benefits to household member
Decrease of other unearned income to
household member
34.1
0.8
8.3
6.1
0.0
3.7
57.5
0.0
8.5
Departure of member with earnings 1.5 5.1 2.1
Departure of member with other income 0.0 — 0.0
Miscellaneous 1.4 0.0 0.0
New household member without income
Infant 2.3 22.4 14.1
Other 6.0 5.6 9.1
Startup of cash assistance, with none of the
above events
2.9 7.6 6.0
ALL EVENTS 53.0 5.5 80.2
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 1,558 observations.
Notes: 1. This table includes only individuals whose household income is less than three times the
poverty threshold income in the baseline wave (i.e., two waves before the food stamp
opening).
2. The overall probability of an opening for this subgroup is 3.6 percent.
3. Probability of opening | event: proportion of individuals experiencing the event who enter the
Food Stamp Program within one or two waves. Probability of event | opening: proportion of
individuals entering the Food Stamp Program who experienced the event one or two waves
previously.
27
Exhibit n.ll
OCCURRENCE OF TRIGGER EVENTS FOR OPENINGS:
MULTIPLE ADULTS WITH CHILDREN
Event
Percent of
subgroup
with event
Conditional probability of:
opening|event event (opening
Household income decreased significantly,
primarily because of:
Decrease of earnings to household 47.6
member
Loss of unemployment insurance 1.0
benefits to household member
Decrease of other unearned income to 4.8
household member
2.6
5.2
3.7
55.8
2.2
8.0
Departure of member with earnings 4.1 4.3 8.0
Departure of member with other income 0.6 9.5 2.3
Miscellaneous 0.4 2.4 0.4
New household member without income
Infant 5.3 4.0 9.7
Other 3.3 6.1 9.1
Startup of cash assistance, with none of the
above events
2.5 4.0 4.6
ALL EVENTS 63.6 3.0 86.3
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 23,177 observations.
Notes: 1. This table includes only individuals whose household income is less than three times the
poverty threshold income in the baseline wave (i.e., two waves before the food stamp
opening).
2. The overall probability of an opening for this subgroup is 2.2 percent.
3. Probability of opening | event: proportion of individuals experiencing the event who enter the
Food Stamp Program within one or two waves. Probability of event | opening: proportion of
individuals entering the Food Stamp Program who experienced the event one or two waves
previously.
28
Exhibit n.12
OCCURRENCE OF TRIGGER EVENTS FOR OPENINGS:
CHILDREN LIVING WITH ONE ADULT
Event
Percent of
subgroup
with event
Conditional probability of;
opening | event event | opening
Household income decreased significantly,
primarily because of:
Decrease of earnings to household
member
Loss of unemployment insurance
benefits to household member
Decrease of other unearned income to
household member
Departure of member with earnings
Departure of member with other income
Miscellaneous
New household member without income
Infant
15.9 6.5 42.0
1.1 3.3 0.7
9.2 3.7 6.2
1.1 19.8 4.0
0.1 27.9 0.5
1.0 0.0 0.0
3.3 30.2 18.0
Other 6.2 10.2 11.5
Startup of cash assistance, with none of the 2.2 16.0 6.3
above events
ALL EVENTS 54.9 7.1
•»
70.4
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 2,739 observations.
Notes: 1. This table includes only individuals whose household income is less than three times the
poverty threshold income in the baseline wave (i.e., two waves before the food stamp
opening).
2. The overall probability of an opening for this subgroup is 5.5 percent.
3. Probability of opening | event: proportion of individuals experiencing the event who enter the
Food Stamp Program within one or two waves. Probability of event | opening: proportion of
individuals entering the Food Stamp Program who experienced the event one or two waves
previously.
29
Exhibit H.13
OCCURRENCE OF TRIGGER EVENTS FOR OPENINGS:
CHILDREN LIVING WITH MULTIPLE ADULTS
Event
Percent of
subgroup
with event
Conditional probability of:
opening | event event | opening
Household income decreased significantly,
primarily because of:
Decrease of earnings to household member
Loss of unemployment insurance benefits
to household member
Decrease of other unearned income to
household member
Departure of member with earnings
Departure of member with other income
Miscellaneous
New household member without income
Infant
Other
Startup of cash assistance, with none of the
above events
ALL EVENTS
48.9
4.9
2.6
2.2
63.5
3.0
5.5
6.8
6.0
3.5
58.5
0.8 7.2 2.3
4.7 3.2 6.0
3.5 5.2 7.2
0.5 5.7 1.1
0.4 4.2 0.6
10.4
6.9
5.1
86.9
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 20,764 observations.
Notes: 1. This table includes only individuals whose household income is less than three times the
poverty threshold income in the baseline wave (i.e., two waves before the food stamp
opening).
2. The overall probability of an opening for this subgroup is 2.6 percent.
3. Probability of opening | event: proportion of individuals experiencing the event who enter the
Food Stamp Program within one or two waves. Probability of event | opening: proportion of
individuals entering the Food Stamp Program who experienced the event one or two waves
previously.
30
From a comparison of Exhibit n.8 with Exhibit II. 10, it can be seen that this difference springs
not from a greater probability of a potential trigger event occurring, but rather from the fact that
the presence of children nearly triples the probability of an opening conditional on the potential
trigger event having transpired.
It is important to bear in mind that the subgroups are defined as of the baseline wave.
Thus, Exhibit n.10 shows entrance to the Food Stamp Program related to trigger events that
occurred to households that already consisted of a single adult and children. The creation of
such households through the breakup of a two-parent family, which may be an important trigger
event for some individuals, will not be seen in this table. Instead, this would appear as a
departure of an earner among multiple adult households with children. The trigger events of
importance for the single parents are rather the addition of new infants or other household
members without income, and the startup of cash assistance.
The patterns for multiple adults with children, in contrast, are quite similar to those for
the population as a whole (Exhibit n.ll), except that decreases in earnings to a household
member are relatively more frequent. Furthermore, the patterns for children living in multiple-adult
households resemble closely those of the adults in these households (Exhibit n. 13). The
same is not quite true, however, for children living with one adult. For these individuals, the
probability of a food stamp opening conditional on a trigger event occurring is high, over 7
percent (Exhibit H. 12).
Differences in patterns between the adults and children could come about in two ways.
First, if some of these households split up in the months following the baseline wave, the events
happening to the adults and the children of these households will not necessarily be the same.
Second, if the patterns are different for households with few children and many children, the
proportions of adults (i.e., families) experiencing the various events will not be the same as the
proportions of children who do so. It will be recalled from Exhibit n.2 that the probability of
a food stamp opening for single adults living with children was only 3.6 percent, while the
probability for children living with single adults was 5.5 percent. This could be explained by
households with more children having higher opening rates. A substantially greater proportion
of the openings for the adults than for the children are associated with a loss of earnings to a
31
household member (S7.S percent in Exhibit n.10, versus 42.0 percent in Exhibit n. 12), and a
lesser proportion with the departure of a household member with income. All of these estimates
for single-parent households with children, however, are based on rather small samples.
Summary
Approximately 2 percent of individuals not receiving food stamps in a given four-month
period, but with household income less than three times the poverty threshold, will commence
a spell of food stamp receipt in the subsequent four-month period. This percentage is markedly
higher in households in which none of the able-bodied adults have high school diplomas (S
percent), and is lower in those in which all of the adults are aged or disabled (less than 1
percent). In addition, members of one-adult households with children are relatively more likely
to begin to receive food stamps than other individuals (4 to 6 percent) while able-bodied
childless adults are less likely (around 1 percent).
Variations are also seen in the distribution of events leading to the receipt of food stamps.
For the population as a whole, loss of earnings to an ongoing household member is clearly the
most important factor, occurring in over half of all food stamp openings. This event is much
less common and less likely to be associated with a food stamp opening for individuals in house-holds
without earnings at baseline, or consisting entirely of elderly and disabled adults. For
members of households without earnings, loss of unearned income is especially likely to be
associated with an opening. The acquisition of a new baby or other household member without
income is a particularly significant trigger event for members of one-adult households with
children. Start-up of cash assistance in the absence of other changes in circumstances occurs
in conjunction with about 5 percent of openings-especially concentrated among noneamers,
households outside the labor force, and, to a lesser extent, the aged and disabled. Exhibit II. 14
summarizes the previous results on what percentage of openings for each subgroup is associated
with each of the major trigger events.
The probability that a trigger event will be followed by an opening varies markedly by
subgroup as well. For the population as a whole, 3 percent of those experiencing any trigger
event begin to receive food stamps. This percentage is substantially higher for members of
32
Exhibit 11.14
DISTRIBUTION OF TRIGGER EVENTS FOR OPENINGS: ALL SUBGROUPS
Decreased Departure of member:
Decreased unearned New other
earnings to income to with member
household household with unearned New with no New cash All
Subgroup member member earnings income infant income assistance events
Earners 64.3% 6.2% 7.9% 1.2% 11.1% 8.2% 4.2% 89.0%
Nonearners 10.7 23.1 1.0 2.4 6.4 8.7 8.7 54.4
High school graduates 58.0 8.3 6.6 1.3 9.4 8.1 4.6 83.3
High school uropouts 49.4 12.0 7.1 1.8 12.5 9.6 5.1 84.4
Able-bodied, childless 62.4 10.4 4.9 2.0 10.2 11.3 2.4 87.2
Aged and disabled, 22.4 14.2 2.7 1.0 1.6 4.5 7.1 49.8
childless
One adult living 57.5 8.5 2.1 0.0 14.1 9.1 6.0 80.2
with children
Multiple adults living 55.8 10.2 8.0 2.3 9.7 9.1 4.6 86.3
with children
Children living with 42.0 6.9 4.0 0.5 18.0 11.5 6.3 70.4
one adult
Children living with 58.5 8.3 7.2 1.1 10.4 6.9 5.1 86.9
multiple adults *
ALL INDIVIDUALS 53.1 9.7 6.4 1.5 10.1 8.3 5.1 81.8
Source: 1984 SIPP Pai 1 (June 1983 to June 1986).
Note: The percentages in this table represent the proportion of all food stamp openings that are associated with each event.
33
households without earnings (4 percent), households headed by high school dropouts (7 percent),
and members of one-adult households with children (6 to 7 percent), suggesting that these types
of households are likely to be on the economic margin. The probability of opening when a
trigger event has occurred is quite low for members of households consisting entirely of aged
and disabled adults (1 percent), suggesting that these households have achieved a certain
stability. Even for the subgroups with the greatest probability of a food stamp opening after a
trigger event, however, only a small percentage begin to receive food stamps.
Finally, we note that there are some (overlapping) subgroups for which the trigger events
analyzed here have little explanatory power; in particular, households without earnings, and the
aged and disabled. The kinds of events that lead these households to enter the Food Stamp
Program may be outside the scope of these data. Among these unmeasured events may be
increased medical needs, increased shelter needs (e.g., due to an eviction or rent increase),
outreach by community groups or by the food stamp agency itself, depletion of assets, and
disasters such as fire or theft. For some households, the immediate trigger may be the
simultaneous occurrence of several such events, no one of which would have had sufficient force
to bring about an application. Thus, trigger event analysis cannot be expected to explain all food
stamp openings, although it can shed light on the relative importance of certain occurrences.
34
AFTER THREE
DURATION OF RECEIPT
This chapter addresses the question of how long new foe.' stamp recipients tend to remain
on the program. Findings on lengths of completed spells are pre^nted first for individuals
covered by the program, and then for the longitudinal households of which they are members.
Length of Completed Spells for Individuals
Exhibit m. 1 presents the frequency distribution of lengths of completed spells for all
individuals who enter the Food Stamp Program.1 The mean and other summary statistics are
shown in Exhibit m.2.2 The key features are:
• The median length of receipt for new recipients is 6 months. That is, half
of all food stamp spells end in six months or less.
• The average spell length is considerably greater: 22 months.
• Over forty percent of all new food stamp recipient spells are 4 or fewer
months long. About a third are over 12 months long, and about 20
percent last more than 2 years.
Higher closure rates appear in the distribution at 4, 8, 12, and 16 months. These are an
artifact of the SIPP data, corresponding to concentrations of individuals who reported coverage
for exactly one or more full waves. This phenomenon is known as the "seam effect"--the
tendency of reported transitions to pile up at the seams between interview periods rather than
to be spread evenly across all months. The rise at 12 months, however, is probably not entirely
an artifact. Many spells of food stamp receipt last exactly 12 months because that marks
'See Appendix C for a description of the hazard rate methodology used to derive this
distribution.
2As discussed in Appendix C, the estimate of mean duration was calculated based on the
observed closure rate for all spells, including left-censored ones. It is thus based on a fuller
sample than the estimate of the median and other statistics of the distribution of completed spell
lengths.
35
Exhibit m.l
DISTRIBUTION OF LENGTH OF COMPLETED SPELLS:
ALL INDIVIDUALS
Months
Probability
of Closure
Cumulative
Probability of
Closure
12
3
4
56
78
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25+
12.7%
9.6
5.1
13.8
4.1
5.7
3.1
6.4
1.3
2.5
1.2
2.8
0.9
0.7
1.1
2.6
0.8
0.5
2.3
0.3
0.2
0.0
2.7
0.0
19.7
12.7%
22.2
27.3
41.1
45.2
50.9
54.0
60.4
61.7
64.2
65.3
68.1
69.0
69.7
70.9
73.5
74.3
74.8
77.1
77.4
77.5
77.5
80.3
80.3
100.0
Source: 1984 SEPP Panel (June 1983 to June 1986).
Notes: 1. Estimates are based on survival analysis of all non-left-censored spells beginning
in or after the fifth month of the observation period.
2. Median: 6 months.
3. Unweighted sample size: 2,623 spells.
36
Exhibit ffl.2
LENGTH OF FOOD STAMP SPFJ LS FOR SUBGROUPS OF INDIVIDUALS
Percent Percent Percent
receiving receiving receiving
Unweighted food stamps food stamps food stamps Mean
sample size Median <: 4 months £ 12 months > 24 months (months)
Earners 1,556 5 47.8% 76.8% 12.1% 13.8
Nonearners 1,067 10 31.3 55.6 31.3 30.1
High school graduates 1,688 6 43.8 69.8 18.3 17.0
High school dropouts 772 7 37.1 67.1 21.4 27.2
Able-bodied, childless 218 5 48.1 78.2 12.6 13.5
Aged and disabled 205 8 42.2 62.8 24.2 29.9
One adult with children 165 9 27.4 55.3 34.3 36.8
Multiple adults with children 839 5 47.2 72.9 12.7 15.5
Children with one adult 340 12 24.0 50.7 38.7 39.2
Children with more than one 785 6 40.4 70.9 15.5 19.5
adult
ALL INDIVIDUALS 2,623 6 41.1 68.1 19.7 21.6
Source: 1984 SIPP Panel (June 1983 to June 1986).
Note: 1. Estimates (except for mean) are based on all non-left-censored spells beginning in or after the fifth month of the
observation period.
2. Estimates of the mean are based on the closure rate in all spells in or after the fifth month of the observation period.
See Appendix C for details of computation.
31
the end of a certification period.1,2 The increase at 6 months is also consistent with the
widespread use of 6-month certification periods.
Exhibit m.2 summarizes the distribution of length of spells for the subgroups of
individuals. (The details of the distributions are presented in Appendix E). All subgroups are
defined as of the first month of receipt of food stamp benefits. The last line of the table
presents the corresponding summary statistics for the recipient population as a whole, taken from
Exhibit m.l.
Recipients whose households contain earners at the time the spell begins clearly have
much shorter spells on average than recipients whose households do not contain earners. The
median completed spell lengths for these two groups are 5 and 10 months, respectively, while
the corresponding means are 14 and 30 months. The remaining statistics tell the same story:
earners are substantially more likely than noneamers to exit within four months (48 versus 31
percent), and substantially less likely to receive food stamps for over two years (12 versus 31
percent).3 The overall difference between the two distributions is statistically significant at the
1 percent level.4
'These results are based exclusively on the core SIPP data, described in Appendix A. For
a discussion of the analogous information in the Welfare History Topical Module and its
unsuitability for the current research, see Appendix D.
^urstein and Visher (1989) derived quite similar statistics based on a nationally
representative administrative data set that covered the period from October 1980 to December
1983. Their unit of analysis was the food stamp case rather than the individual. Hence cases
with more members were weighted relatively less heavily than in the current analysis. They
found a slightly greater median spell length of 7 months, and a somewhat lower percentage of
spells lasting 4 or fewer months (36 rather than 41 percent). However, 33 percent of spells
were found to last over 12 months (versus 32 percent in the current study), and 20 percent were
found to last over two years (identical).
'Burstein and Visher found a median spell length of 6 months for cases with earnings.
Forty-two percent exited within 4 months, and 12 percent received food stamps for over 2 years.
As for the food stamp population as a whole, the administrative data showed a somewhat greater
concentration of spell lengths between 5 and 12 months relative to spell lengths between 1 and
4 months than did the survey data analyzed here; but otherwise the distributions look quite
similar.
*The log rank test on the survivor functions (described in Appendix C) yields a chi-squared
statistic of 107.7 for the null hypothesis that the two sets of food stamp spells come from the
same distribution.
38
Some variation is also seen when recipients are classified by educational status. It should
be recalled that this partition excludes members of households in which all the adults are aged
or disabled. For individuals in households which contain a high school graduate, the median
duration is 6 months and the mean is 17 months-somewhat shorter than for the recipient
population as a whole. For individuals in households where the adults do not have high school
diplomas, the median duration is a little longer (7 months), while the mean is substantially
longer (over two years).1
Finally, the demographic subgroups show a wide variety of patterns. The groups with
the shortest spells are the able-bodied adults-both those who live only with other able-bodied
adults, and those who live with other adults and children. Members of these subgroups have
nearly a SO percent chance of leaving the Food Stamp Program within four months of entry, and
only a 13 percent chance of remaining on the program for over two years. Mean duration for
these individuals is 14 to 16 months.
Children living with multiple adults, however, tend to have somewhat longer spells on
average than these adults. This suggests that larger households have longer spells. (The
difference in means would come about because a large household would have the same number
of adults as a smaller household, but would have more children.) In addition, it may be that
some of the adults split off from the households, leaving the children behind still as food stamp
recipients. Even so, these children have substantially shorter stays than their counterparts in
one-adult households-20 versus 39 months on average. In fact, children in one-adult households
have the longest spells of any of the demographic subgroups, with barely half leaving the
program within a year of entry. The adults in these households have slightly shorter spells2,
with a mean length of 37 months.3
'The log rank chi-squared for this comparison is 4.1, significant at the 5 percent level.
'Although the difference in median spell length between adults and children in these
households appears large (12 versus 9 months), the overall distributions of spell lengths do not
differ significantly (chi-squared = 1.00). That is, because of the small sample size for these
two subgroups, the summary statistics cannot be estimated very precisely.
3The distribution for one-adult households with children may be compared with the
distribution for the roughly similar subgroup of AFDC recipients in Burstein and Visher. The
latter had a median spell length of 14 months, with only 17 percent of spells ending with four
months and 34 percent lasting over two years. The administrative data for AFDC cases thus
39
The remaining group-the aged and disabled-has a mean duration of 30 months. One
quarter of this group remains on the Food Stamp Program continuously for at least two years.1
The full distributions of spell length were compared for four pairs of demographic
subgroups: able-bodied versus aged and disabled, children living with one adult versus children
living with multiple adults, single parents versus able-bodied childless adults, and single parents
versus adult members of multiple-adult households with children. In all four instances, the pairs
of survivor function's weic statistically significantly different at the 1 percent level.2
Length of Completed Spells for Households
There has been much controversy about the proper definition (if any) of a longitudinal
household.3 In the SIPP data, households are classified each month according to whether they
contain a family-i.e., two or more individuals related by blood or marriage-and whether they
are headed by an unmarried man, an unmarried woman, or a married couple. Both the identity
and marital status of the head are recorded as reported by the interviewer. The five household
types are thus:
• married-couple household
• other family household, female head
confirms that this household type tends to receive food stamps longer than other types, but shows
a greater concentration of longer spell lengths among those spells that last up to about two years
than is found in the survey data analyzed here.
'Burstein and Visher define the elderly as households containing an individual aged 65 or
older. For this subgroup, the administrative data show a median spell length of 19 months,
much longer than the eight-month median found for the aged and disabled in the SIPP data.
Only IS percent exited in four months, and 41 percent had spells that lasted more than two
years. The corresponding statistics from the SIPP are 42 and 24 percent. Thus the
administrative data show substantially longer spells for the elderly than do the survey data. It
could be argued that part of the difference could be due to the differences in subgroup definition,
and perhaps in the time frame (1980-1983 versus 1983-1986). As reported in Chapters Two and
Four, however, food stamp openings and closings for the elderly and disabled are only poorly
correlated with the occurrence of measured trigger events. This suggests another hypothesis,
that response error may be particularly great for this subgroup. If so, the average spell length
in the SIPP may be underestimated.
The log rank chi-squared statistics were 11.2, 32.4, 21.1, and 17.7, respectively.
'See, for example, McMillen and Herriot (1985), and Duncan and Hill (1985).
40
• other family household, male head
• nonfamily household, female head
• nonfamily household, male head.
In the SEPP data, a longitudinal household is said to continue from one month to the next
if it remains the same household type, if it retains the same reference person or householder,
and if it retains the same householder's spouse (if any). In other words, the key person(s) of
the household must be unchanged. Any of the following events will therefore lead to a disconti-nuity:
death or departure of householder, death or departure of householder's spouse, marriage
of householder, death or departure of either member of a two-member family household, birth
of a child to a woman living alone, or acquisition of a family member to a person living alone.
In the sample of original interviewees, one out of six experienced a change in household
reference person or spouse over the 32 months of observation.
The logic behind the SIPP household definition is that after a major change in
composition, the household is so altered that it cannot legitimately be called the same household
as before. An implication of this, however, is that the clock of food stamp receipt is reset to
zero for a group of individuals whenever the household type changes, but not otherwise. As a
consequence, the distribution of spell lengths for households may be misleadingly low, if many
groups of individuals continue to receive food stamps despite changes in household type.
Conversely, it could be misleadingly high, if many individuals leave and enter households that
receive food stamps. Suppose, for example, a married couple household that was receiving food
stamps for a year splits into two households, and both individuals continue to receive food
stamps for another year. Then the household level data will show three spells of receipt of one
year each, although at the individual level there were two individuals receiving food stamps for
two years each.
Situations like these suggest that analyzing spell lengths for individuals will provide more
useful information about how long people receive food stamps then analyzing spell lengths for
households. Most earlier research on the Food Stamp Program, however, has focused on the
household as the unit of analysis. For purposes of continuity and comparability, we have
therefore replicated the individual-level analyses presented above, using the Bureau of the
41
Exhibit m.3
DISTRIBUTION OF LENGTH OF COMPLETED SPELLS:
ALL HOUSEHOLDS
Months
Probability
of Closure
Cumulative
Probability of
Closure
I23
45
6
78
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25+
14.1%
9.3
6.9
10.5
4.4
6.7
3.2
4.2
1.7
2.1
1.4
2.6
1.3
1.6
1.7
1.5
1.4
1.2
1.7
C.4
0.0
0.0
2.4
0.0
20.0
14.1%
23.4
30.3
40.8
45.2
51.9
55.1
59.3
61.0
63.1
64.4
67.0
68.3
69.9
71.6
73.1
74.4
75.6
77.3
77.7
77.7
77.7
80.0
80.0
100.0
Source: 1984 SIPP Panel (June 1983 to June 1986).
Notes: 1. Estimates are based on survival analysis of all non-left-censored spells
beginning in or after the fifth month of the observation period.
2. Median: 6 months.
3. Unweighted sample size: 963 spells.
42
Census definition of the household that is employed in SEPP. Comparison of the individual- and
household-level distributions provides evidence as to how significant the distinction really is.
Exhibit ID. 3 shows the length of completed spells of food stamp receipt for longitudinal
households. Despite the ambiguity in the definition of a longitudinal household and the potential
for bias in estimated spell lengths, the distribution is practically identical to that for individuals.
The median spell length is identical at 6 months; the mean spell length of 21.3 months differs
only slightly from the mean for individuals; and the proportions of spells ending within 4, 12.
and 24 months are all very similar to the corresponding statistics in Exhibit III. 1. It appears that
the putative downward bias associated with household dissolution is either rendered unimportant
by the coincidence of food stamp transitions with major household changes, or else
counterbalanced by an upward bias from new entries and split-offs. The great similarity between
the two distributions is shown graphically in Exhibit HI.4.
Comparison of Exhibits m.2 and HI.5 indicates that within subgroups as well, the
distribution of length of completed spell is very similar for individuals and for households.1
The household-level data appear to yield somewhat longer spells for the aged and disabled.
Subgroups for which the household data indicate shorter spells are those in which the adults are
not high school graduates, and those containing children. Even these differences, however, are
relatively small.
While it is possible in principle that these differences represent the net effects of several
important counterbalancing forces, this turns out not to be the case. As demonstrated in
Appendix G, the events associated with individuals continuing to receive food stamps, while the
households to which they belonged no longer do so or have ceased to exist, are quite rare,
occurring to only 1 percent of recipients per month. The greatest concentration of these events
is seen among able-bodied, childless adults, with a monthly rate of 1.7 percent. Similarly, the
events associated with individuals ceasing to receive food stamps, while their households (or
former households) continue to do so, are also quite rare, occurring to only 0.6 percent of the
food stamp population per month. Again, the greatest concentration is among able-bodied,
childless adults, with a monthly rate of 1.1 percent. Thus, not only are the net effects of these
two kinds of events small but the separate effects are small as well. We conclude that
'The details of the distributions for subgroups of households appear in Appendix F.
43
Percent %
45
40
35
30
25
20
15
10
5
0 I
Exhibit III.4
DISTRIBUTION OF LENGTHS OF SPELLS FOR
HOUSEHOLDS AND INDIVIDUALS
I i
Households
Individuals
^~1 ■ rc*~i 1
1-4 5-8 9-12 13-16 17-20 21-24 25+
Number of Months
MM
CM
Exhibit III.5
LENGTH OF FOOD STAMP SPELLS FOR SUBGROUPS OF HOUSEHOLDS
Percent Percent Percent
receiving receiving receiving
Unweighted food stamps food stamps food stamps Mean
sample size Median £ 4 months £ 12 months > 24 months (months)
Earners 481 4 50.7% 76.7% 9.8% 12.3
Nonearncrs 482 10 30.5 57.1 30.4 29.0
High school graduates 544 6 44.2 72.0 14.5 16.2
High school dropouts 284 6 41.2 64.8 23.8 22.9
Able-bodied, childless 158 5 47.1 78.1 13.9 13.7
Aged and disabled 158 11 29.4 54.9 31.8 32.5
One adult with children 212 10 27.2 54.1 36.5 29.9
Multiple adults with children 414 5 48.2 72.9 9.4 15.6
ALL HOUSEHOLDS 963 6 40.8 67.0 20.0 21.3
Source: 1984 SIPP Panel (June 1983 to June 1986).
Note: 1. Estimates (except for mean) are based on all non-left-censored spells beginning in or after the fifth month of the
observation period.
2. Estimates of the mean are based on the closure rate in all spells in or after the fifth month of the observation period.
See Appendix C for details of computation.
45
distributions of spell lengths based on household level data, though potentially biased in theory,
are not visibly b; ".sod in practice.
Summary
Half of all new food stamp recipient spells reported in the SIPP are no more than 6
months long, and two-thirds end within a year. The average spell length is 22 months.
There are substantial variations from this pattern for certain subgroups, however.
Individuals in households that contain earners at the start of the spell, that consist entirely of
able-bodied adults, or that, if they contain children, include more than one adult, tend to receive
food stamps for considerably less time. This suggests that policies that are designed to hasten
the exit of such recipients from the food stamp rolls may be redundant. Longer spells are seen
among households that lack earners, those in which the only adults are aged and disabled, those
in which the only able-bodied adults are high-school dropouts, and especially those which consist
of a single adult with one or more dependent children. The implication is that policies that
addressed the barriers to employment of the larsr two subgroups (e.g., need for remedial
education and child care) could have the potential for shortening food stamp spells.
Although the definition of a longitudinal household is somewhat ambiguous, the spell
length distributions for individuals and households were found to be quite similar, both for the
various subgroups and for the population as a whole.
46
CHAPTER FOUR
CIRCUMSTANCES SURROUNDING EXITS FROM THE FOOD STAMP PROGRAM
Food stamp case closures can be thought of as consisting of four types: voluntary,
circumstantial, administrative, and jurisdictional. A voluntary closure is one that is explicitly
requested by an eligible recipient. A circumstantial closure represents a change in the
recipient's needs or resources that renders the case ineligible for food stamps. An
administrative closure occurs when a circumstantially eligible recipient fails to meet a
requirement such as work registration, monthiy reporting, or appearance at a certification
interview. Finally, a jurisdictional closure indicates a change in geographical jurisdiction, due
to the recipient transferring to another locality.
It is virtually impossible for any data base to identify all four of these types of closures.
Administrative records tend to be incomplete with regard to reasons for closure. At best, they
will indicate circumstantial closures only in those instances in which the agency has explicitly
determined ineligibility, e.g., via a recertification or a mon.nly report. Clients who lose
circumstantial eligibility may refrain from appearing for their next recertification or from filing
their next monthly report, however, in which case the agency records will show an
administrative closure instead. Alternatively, newly ineligible clients may call and request a
closure. Because the agency has not verified the change in circumstances, these will be recorded
as voluntary closures.
Survey data such as the SIPP, in contrast, can shed light on changes in circumstances
surrounding case closures. Unless a survey is explicitly designed to focus on reasons for non-receipt,
however, it will not include information on administrative requirements. Hence
administrative and voluntary closures cannot be distinguished.
In this chapter, we examine the relationship between changes in circumstances and exits
from the Food Stamp Program, using the same trigger event approach as is found in Chapter
Two. Voluntary closures are probably rare in the Food Stamp Program, because the costs of
participation are highest at the outset, while the benefits of participation are approximately
47
constant over time.1 Jurisdictional closures will not look like exits in these data, if the
household is followed to its new location. Administrative closures that last for only one month
have been filled in, i.e., the data indicate that no closure has occurred. Hence, we would expect
to find trigger events associated with the great majority of closures. This is indeed the case for
the recipient population in general, although the trigger event framework is less fruitful for the
aged and disabled. For all subgroups except this one, an increase in earnings to household
members is the most common trigger event for an exit. Variations are seen across all the
subgroups, however, in the relative importance of other events.
We conclude this chapter with an examination of recidivism to the program. Nearly 40
percent of recipients are found to reenter the Food Stamp Program within a year of leaving it.
Definition of Trigger Events
The primary trigger events that could potentially lead to a person no longer receiving
food stamps are:
• increased household income-due either to a member of the household
gaining income, or to someone with income joining the household;
• reduced need-the departure from the household of a person who has no
income; and
• departure from the SEPP sample-through death, institutionalization,
emigration, or induction into the Armed Forces.
As in the analysis of food stamp openings in Chapter Two, the household is defined as the set
of people currently living with the individual whose food stamp coverage is being considered.
Thus a new earner could enter an individual's household in two ways: the earner could move
in with the individual, or alternatively the individual could move to a different household which
contains the earner.
A closure is said to occur if an individual who was covered for one or more months
during a four-month wave is not covered for any months during the succeeding wave. Only
'The benefits of participation could decline if a recipient's entitlement decreased; but this
would represent a circumstantial change. Note that we do not attempt to measure eligibility
explicitly, but treat all increases in resources and decreases in needs as circumstantial changes.
48
individuals who were covered in the preceding wave are at risk for a closure. An individual
may contribute multiple observations to the analysis sample-as many as five, if food stamps
were received in each of Waves 3 through 7.1
As in Chapter Two, the trigger events have been defined rather broadly, in an attempt
to capture as much of the associated activity as possible. For example, an individual may depart
from the sample (through death, institutionalization, etc.) either in the wave of closure or in the
preceding wave. Thus, we may observe that a deceased sample member last received food
stamps in Wave 6. The closure is associated with Wave 7 (the first wave in which food stamps
were not received). The death itself may have occurred in either Wave 6 or Wave 7, and be
counted as the trigger event in either event.2
Similarly, a person who last received food stamps in Wave 6 may have experienced an
increase in total household income. This increase may be seen as higher income in Wave 7 than
in Wave 6 if, for example, the earnings first show up in month 1 of Wave I.3 Alternatively,
the increase may be seen as higher income in Wave 6 than in Wave 5, if a person got a job
during Wave 6 but still received food stamps for all or part of that wave. In either case, the
increase in income is counted as a potential trigger event.
'Wave 2 is not used as a preceding wave because case characteristics must be examined in
the next earlier wave in order to construct the trigger events, and Wave 1 data on household
composition are not comparable with those of later waves.
2If the death occurred in the wave before the closure, the death is not counted as a potential
trigger event for closure in the earlier wave. Thus by construction, events such as death cause
an exit with a probability of 1.
technically speaking, we would expect some overlap in months with new earnings and
months with food stamps before a person exited the Program because stamps are generally issued
at the beginning of the month. When these events are reported, however, it is likely that the
respondent would mentally classify months of the reference period as being either "food stamp"
or "earnings" months. Hence overlap would not necessarily be reported. Furthermore, even
if earnings were first obtained in the second or third month of the reference period, it would not
be too surprising if the respondent mentally backfilled them throughout the period. We thus
allow an earnings increase which appears to be simultaneous with an exit to count as a trigger
event for the exit.
49
i»J
In identifying the type of income increase, we first determine whether an increav*
occurred in the current or preceding wave. If increases occurred in both waves, we pick the
larger of the two. We then determine which component made the greatest contribution to the
increase in household income between the two consecutive waves: a new earner, an increase
in earnings to a current household member, a new member with unearned income, or an increase
in unearned income to a current member. A particular event such as a new job is thus a
potential trigger event for a food stamp closure in both the same wave and the following wave.
The question of how large a change in household income must be, in order to count as a trigger
event rather than a mere fluctuation, is explored in Appendix B. A cutoff of $400,
corresponding to an increase in household income of $100 per month, was selected. If the total
increase in household income exceeds $400, but no individual component does so, then the
change in income is classified as "miscellaneous."
Finally, decreases in the number of individuals without income are also examined both
between pairs of consecutive waves. The event considered is the absence of an individual from
the household in the later wave who was present without income in the earlier wave of the pair-regardless
of whether the total number of individuals who are not contributing income to the
household has gone up or down. The departure of such a person is a potential trigger event for
a closure in either the same wave or the following one.
The recipient subgroups are defined based on characteristics in the next to last wave
before the potential closure, called the baseline wave. For example, if an individual received
food stamps in Wave 4 and we are investigating whether a closure occurred in Wave 5, we
classify the individual according to characteristics in Wave 3. This ensures that the subgroups
are defined prior to the occurrence of the putative trigger events. (Recall that a change in
household income between Waves 3 and 4 may trigger a closure in Wave 5.) In particular, the
demographic and educational categories are determined as of the first month of the baseline
wave, while the presence of earnings in the household is determined by looking at the baseline
wave in its entirety.
50
Overall Probability of Exit
Exhibit IV. 1 provides an oveiview of the probability of closure for the food stamp
population as a whole and for the various subgroups. The bottom line of the exhibit indicates
that 15 percent of individuals covered by food stamps in a given wave were not covered in the
following wave. This corresponds to a monthly exit rate of about 4 percent-although the exits
tend to be piled up at the seams between the waves. There is, however, substantial variation
in this rate by subgroup. Individuals in households with earnings have a 22 percent chance of
exiting during a wave, while individuals in households without earnings have only an 8 percent
chance. Thus earners as a subgroup comprise less than half of the recipient population, but they
account for nearly three-quarters of the exits.1
Education makes almost as great a difference as presence of earnings in predicting exits.
Individuals in households containing an able-bodied, non-elderly high school graduate have a 19
percent chance of exiting, while individuals in households in which none of the able-bodied, non-elderly
adults have a high school diploma, have an exit probability of only 11 percent.
Among demographic subgroups, the greatest exit probabilities are seen among individuals
in households consisting only of able-bodied, non-elderly adults (23 percent) and among adults
living with other adults and children (20 percent). The lowest rates are seen for individuals
'This proportion of food stamp recipients living in households that contain an earner is
surprisingly high. Studies of the food stamp population based on Quality Control System data
show that only 20 percent of food stamp households have earnings.
Most of the difference is simply definitional. It is to be recalled that the unit of
observation in Exhibit IV. 1 is the individual, rather than the household; and that the presence
of earnings is determined based on a four month period rather than a single month. When we
examine the presence of earnings in the last month only of each wave, and weight each
individual by the inverse of household size (so as to count each household equally), the estimated
proportion of the food stamp caseload with earnings in the SIPP drops to 32.4 percent. The
remaining discrepancy of some 12 percentage points relative to the administrative data must be
attributed to (a) differential reporting of earnings between the survey and administrative data;
(b) misreporting of food stamp status in the SIPP; and (c) the fact that some household members'
earnings are not countable from the point of view of the Food Stamp Program.
51
Exhibit IV.l
OVERALL PROBABILITY OF EXITING FROM THE FOOD STAMP PROGRAM
BETWEEN TWO CONSECUTIVE FOUR-MONTH PERIODS
Percent closing
Percent of in next four Percent
recipients months of closings
Earners 47.9% 22.1% 70.8%
Noneamers 52.1 8.4 29.2
High school graduates 50.3 19.2 64.6
High school dropouts 38.4 10.9 28.0
Abie-bodied, childless 5.7 23.3 9.0
Aged and disabled, childless 11.3 11.7 9.0
One adult living with children 11.3 9.4 7.2
Multiple adults living with 24.0 20.3 33.0
children
Children living with one adult 22.2 9.2 13.9
Children living with multiple 25.6 16.1 27.9
adults
ALL RECIPIENTS 100.0 14.9 100.0
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 12,268 observations.
Notes: 1. The percentages shown pertain to Waves 3 through 8 combined.
2. For definitions of population subgroups, see Exhibit U. 1. High school graduate
and dropout subgroups do not sum to 100 percent of the population because
individuals in households containing only elderly or disabled adults are excluded.
52
living in single-adult households with children (9 percent). The two remaining subgroups are
not too far from the population average: the elderly and disabled (12 percent probability of exit)
and children living with multiple adults (16 percent probability of exit).
Occurrence of Trigger Events: All Recipients
Exhibit IV.2 shows the occurrence of the previously defined trigger events for all
recipients. As shown in the last line of the exhibit, 51 percent of recipients experience one or
more of these events. Their exit probability is then 24 percent, substantially higher than the rate
for the recipient population as a whole (15 percent). From another perspective, over 80 percent
of those that exit the Food Stamp Program experienced a trigger event.
Turning to the individual events, we see that in any four-month period, 0.7 percent of
recipients leave their households due to death, institutionalization, or other similar events.' Ali
of these individuals exit the Food Stamp Program, by definition. They account for 4.4 percent
of all closures.
Increases in household income are much more common. As noted above, a cutoff of
$400 between waves was used. Forty-seven percent of recipients experience an increase in
household income of at least this amount. In nearly two-thirds of these cases, the increase is
due solely or primarily to an ongoing household member obtaining or increasing earnings.
Nearly all of the remainder of increases in household earnings are attributable to increases in
unearned income received by ongoing household members. A small percentage of increases are
due to new household members bringing in earned or unearned income.
The second column of this exhibit shows an interesting pattern. Increases in household
income that are due to changes in earnings are one and one-half times to twice as likely to be
associated with a food stamp closure ihan those that are due to changes in unearned income-regardless
of whether the income is from an ongoing or a new household member. In fact,
'It is not completely clear what the "other" subcategory represents in this regard. These are
individuals who were assigned positive longitudinal weights by the Bureau of the Census,
indicating that they did not attrit from the sample, but rather left the SIPP sample frame of
households. An explanation that has been suggested is that some of these individuals were
assigned positive longitudinal weights in error--e.g., children who turned 15 in the course of the
panel and who were not followed when they moved to new households. (David McMillen,
private conversation.)
53
Exhibit IV.2
OCCURRENCE OF TRIGGER EVENTS FOR CLOSURES:
ALL RECIPIENTS
Event
Percent of all
recipients with
event
Conditional probability of:
exit I event event I exit
Left the sample
Died
Was institutionalized
Entered armed forces
Emigrated
Other
0.7%
0.2
0.1
0.0
0.0
0.3
100%
100
100
100
100
100
4.4%
1.4
0.8
0.1
0.1
2.0
Household income increased significantly,
primarily due to:
New member with earnings 2.6 28.3 4.9
New member with unearned income 0.7 19.0 0.9
Ongoing member obtaining or
increasing earnings
29.8 28.6 57.0
Ongoing member obtaining or
increasing unearned income
13.1 12.7 11.2
Other 0.7 11.1 0.6
Departure of or from persons without
income
8.8 21.8 12.9
ALL EVENTS 51.2 23.7 81.3
Source: 1984 SIPP Panel (June 1983 to June 1986).
Unweighted sample size: 12,268 observations.
Notes: 1. The overall p