Rational Choice, Culture of Poverty, and the Intergenerational Transmission of Welfare
Dependency
Author(s): Mwangi S. Kimenyi
Source: Southern Economic Journal, Vol. 57, No. 4, (Apr., 1991), pp. 947-960
Published by: Southern Economic Association
Stable URL: http://www.jstor.org/stable/1060325
Accessed: 26/06/2008 11:53
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=sea.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].
http://www.jstor.org
Rational Choice, Cultureof Poverty,and the
IntergenerationalTransmissionof Welfare
Dependency*
MWANGI S. KIMENYI
Universityof Mississippi
University,Mississippiand
Universityof Connecticut
Storrs, Connecticut
I. Introduction
It has now become evident that, contraryto the expectationsof the proponentsof the Great
Society, a significantnumberof recipients of transferbenefits have welfare careersthat last for
many years. Prolonged reliance on welfare has generateda wide range of issues related to the
process and patternsof welfare dependency and has raised questionsabout how best to reform
welfare policy. For example, it has been suggested that welfare use is "addictive,"creating a
"welfareculture" that is transmittedacross generations.Glicken [7, 31] states that most recipients of AFDC benefits are those people "whose parentswere recipients-people who have been
socialized into the welfaresystem as children."Duncan,Hill, and Hoffman[5, 469], on the other
hand, investigatethe patternsof welfare dependenceamong AFDC recipientsand conclude that
"[t]he stereotypeof heavy welfaredependencebeing routinelypassedfrommotherto child is ...
contradicted. . ."' Other studies [2; 24] have suggestedthat there are significantracial differences in the durationof time on welfarepointingto a race-baseddependencypattern.Rank [24],
for example, finds thatblack women remainon welfarefor significantlylongerperiodsthan white
women. Specifically, he finds the median numberof months on welfare to be 21.6 months for
white women and 45.2 for black women.2Otherstudieshave suggestedthatlong-timewelfareuse
*Researchsupportedby National Science FoundationGrantNo. NSF-8816029. I wish to thank Joane Thompson
and Craig Irwin of the TennesseeDepartmentof HumanServices for their assistancewith the data. While the data used
in this study are from a survey conductedunderthe auspicesof the governmentof the state of Tennessee, the views expressed are mine and should not be attributedto any agencyor any of its employees. I am gratefulto ChristopherColburn,
John Conlon, Bill Shughart,and Lewis Smith, for their comments on an earlierdraft of this paper. The final draft was
writtenwhile I was visiting the Institutefor Researchon Povertyat the Universityof Wisconsin-Madison.I thankCharles
Manski the Director of the Institutefor providingme with facilities to complete the project, and Duncan Chaplin and
Nancy Maritatofor their assistancewith the use of the Universityof Wisconsin'scomputersystem. I am very grateful to
an anonymousreferee whose commentshave been very helpful in preparingthe final draft. Any remainingerrorsare my
responsibility.
1. However,althoughthe authorsreachthis conclusion, they also find that "[t]he fractionof daughtersfrom highly
dependenthomes who themselves become highly dependent(20 percent)is much greaterthan the fractionof daughters
from nonrecipientfamilies who become highly dependent(only 3 percent)"[5, 469].
2. Rank [24] suggests that these differences could be a result of cultural and opportunitydifferences between
blacks and whites.
947
948
MwangiS. Kimenyi
is a rationalresponse to high state welfarebenefitsand yet othershave arguedthatdependencyis
a consequence of behavioralpatternsof the underclasssuch as early teenage pregnanciesthat are
consequencesof a "cultureof poverty."3
The studies that have attemptedto analyze welfareutilizationover time have generallyused
data from three main sources: the National LongitudinalSurvey (NLS) of Young Women, the
Panel Study of Income Dynamics (PSID), and local caseloadfiles.4Each of these data sets present
severalproblemswhen used to analyze welfaredependency.In both the PSID and NLS data sets,
welfarereceipt is observed only on an annualbasis and exits off and back on the welfarerolls that
occur within a year are not recorded. In studies using such data, any year in which an individual
reports welfare receipt is counted as a spell year. Consequently,an individualwho remains on
welfarethe whole year and one who receives benefitsfor only one monthor less in the same year
are consideredequally dependent. Studiesthatuse these types of datagrossly inflatethe extent of
welfaredependencyfor some recipients.
The NLS includes a sample of women aged 14-24 in 1968 and aged 25-35 in 1979. The
major advantageof the data from this survey is that it includes a large numberof recipients.
However,althoughthe sample is used to analyzeAFDC utilization,it is not representativeof the
AFDC populationbecause the NLS data are not confinedto AFDC paymentsbut includeall types
of public assistance. Furthermore,the NLS data set does not reportthe welfarestatusof women
in some of the years which leaves the researcherto make assumptionsabout welfare utilization
for those years.
The original PSID data consisted of some 5,000 poor familiesin 1968, with one thirdof the
sample being black and thereforethe sample is clearly not representativeof the nation's welfare
population. A more complex problem associated with the PSID data set is that it is difficult to
identify subfamilieswithin the sample particularlyfor females who neverlived independentlyand
yet are welfare recipients themselves. Thus the PSID data understatesthe extent of welfare use
and fails to reveal parent-to-childwelfareuse.
Caseload data sets are often accurateand containdetailedinformationaboutrecipientcharacteristics.However,these data show false exits when an individualchangesadministrativejurisdictions while still receiving welfarebenefits. Furthermore,most studiesusing these types of data
often look at welfare utilizationby sample recipientsbetween two time periods-say recipients
on welfarebetween to and tl. Such data ignorepriorhistoryof welfareuse and thereforeare poor
indicatorsof the length of time on welfare.5
The present study investigatesthe determinantsof the length of time individualsremainon
welfareusing a sample consisting of 991 AFDC recipientsfrom the state of Tennessee. The survey sample, which met standardsample selectioncriteria,was selected from a statewideuniverse
of 66,297 AFDC cases active as of January25, 1988. The data were obtainedthroughin-home
interviews conducted in March and April 1988. An importantadvantageof this data set is that
3. See for example, Gallaway and Vedder [6], and Murray[17]. For details regardingthe culture of poverty
hypothesis, see Lewis [15; 16].
4. Studies using these types of data include Duncan, Hill, and Hoffman[5] Murrayand Laren [18], O'Neil et al.
[19], O'Neil, Bassi, and Wolf [20], Rydell et al. [27], and Rein and Rainwater[25]. Thereare otherstudiesthathave used
data from income maintenanceexperiments,includingPlotnick[22] and Plant [21]. A relativelynew data set that may be
useful in the study of welfare dependency is the Survey of Income and ProgramParticipation(SIPP). The first panel of
SIPP contains data from October 1983. AlthoughSIPP containsa wide rangeof variables,each panel follows recipients
for only two and a half years. Kimenyi [13] providessome preliminaryestimatesof welfaredependencyusing the 1984
panel of SIPP.
5. This is the problemof left and rightcensoringdiscussedbelow (see note 7).
WELFARE DEPENDENCY
949
it contains informationthat takes into account previousepisodes of welfare use and ratherthan
counting any year of positive welfarereceipt as a spell year, such as in O'Neil, Bassi, and Wolf
[20], only the actual months of welfare receipt are counted. We thereforehave informationon
actual length of welfare careers precisely to the month. The data also include informationabout
recipients' employment history, marital status, education, and also informationthat can reveal
whetherthe recipients are themselves childrenof welfarerecipients, and other demographicand
economic variablesthat are necessaryto explain the process of welfaredependency.In addition,
the sample includes a wide rangeof recipientsbased on age, areaof residence, and so on, which
avoids biases introducedby selecting recipientsfrom specific categoriesor age groups such as in
Murrayand Laren [18].
After a randomsample was selected, all the recipientschosen respondedbecause the interviews were conducted simultaneouslywith case reviews giving an almost 100 percent response
rate and hence avoiding any biases thatwould be introducedby selective responses(for example,
longer term users could refuse to respondto the survey). Second, the problemsof jurisdictional
changes were eliminated because history of welfare use was recordedregardlessof where the
individualused such services.
While most studies on welfare dependencygenerallyfocus on the urbanpoor, particularly
in northernmetropolitanareas, the present study focuses on a sample from a typical southern
state. Some unique features of southernstates make the study of welfare dependency in those
states interesting. Some of these featuresinclude:high concentrationof low wage employment,
seasonalityof employment,high concentrationof povertyin some ruralareas, thinlabor markets,
and the low level of educationattainedby most residents.In addition,welfarepolicy in the South
has for long been influenced by powerful employersfor the purpose of increasinglabor supply
and maintaininglow wages. These factorsmay lead to patternsof welfaredependencythat differ
from those observed in other areas.
The focus of this paper is to empiricallyinvestigatethe factors that tend to lead to prolonged relianceon welfare-that is, those factorsthatreducethe probabilityof exit from welfare.
Section II outlines a model of welfare dependency and the methodology used to examine the
determinantsof durationof time on welfare. SectionIII presentsempiricalresults, and section IV
contains concludingremarks.
The empirical evidence presentedin this paper sheds some light on the process of welfare
dependency but at the same time suggests disturbingand challengingpublic policy questions
concerning welfare use and dependency. Althoughwe find economic variablesto be important
determinantsof welfare dependency,we also find evidence of intergenerationaltransmissionof
dependency. While blacks remain on welfare for longer periods than whites, this outcome does
not originatefrom racialdifferencesin attitudesaboutwelfarerecipiency,but is largely a result of
the lower probabilityof exit from welfareby way of marriage.
II. A Model of Welfare Dependency
We start with the basic premise that an individual only remains on welfare if utility on welfare Uw is greaterthan utility off welfare Uo. Let Uw = Uw(YT, ) where YTis the amountof
transferincome, and a is a measure of the disutility associated with "being on welfare." Let
Uo = UO(@YE,P,/3) where YE is potential income that would be earned from work, O is a
measureof the probabilityof obtainingemploymentand, P takes into accountthose factors that
950
Mwangi S. Kimenyi
increase the opportunitycost of working (such as the provisionof child care services). /3 is a
measureof disutility associated with work. The individualwill receive welfareat any particular
time t only if
Uw(Yr,a) - UO(OYE, ,/3) > 0.
(1)
Prolongedwelfare use implies that the conditionfor welfarereceipt holds throughtime periods,
to, tl, t2, t3, . . ., tn, or at least the condition will hold most of the time so that exits from the
welfare system will be brief. The participationfunctionin time t can then be expressed as:
f,
= Uw(YT, a) - UO(OYE, ,3)
> 0.
(2)
YT is determinedby the political process and is usually based on the numberof children in
the recipient household. Thus recipientscan only increaseaggregatetransfersby increasingthe
numberof children. On the other hand, the aggregatetransfersare reduced by an implicit tax
when an individual earns marketincome. The expected income from work YEwill depend not
only on economic factors such as the unemploymentrate, but also on the probabilityof obtaining
employmentwhich is influencedby factorssuch as education,location, and otherrecipientcharacteristicssuch as age, work experience, and so on. Race could also be an importantfactor if the
probabilityof obtainingemploymentdiffersacross racialgroups, otherthings equal.
The existing literatureon welfaredependencysuggests that attitudestowardwelfaremay be
shapedby the degree to which recipientshave been exposed to welfareusage. Recipientswhose
parentswere welfarerecipientsare more likely to have been "socialized"into the system so that
they accept welfareuse more readilythanthose from nonrecipientfamilies. Consequentlya proxy
for a could be whether or not a recipientcame from a welfarefamily. We have no proxy for /3
and it is assumedthat tastes for work are uniformacrossrecipientswith similarcharacteristics.
To analyze welfare dependencyusing the surveydata describedabove, some featuresassociated with the data need to be considered.To illustrate,considertwo welfarerecipientsX and Y,
both of whom had up to threedifferentperiodsof welfareutilization(xl, x2, x3, andyl, Y2,y3) as
shown in Figure 1. At the time of survey(tl), Y had exited the welfare system after completing
the third period and was thereforenot included in the survey while X was still on welfare and
was included in the sample. Severalproblemsare associated with durationdata of this nature.
First, the survey informationis only availablefor those on welfare at the time of survey. Such a
sample under-representsthose with shorterperiodsand over-representsthose with longer periods
who are likely to be on welfareat any particulartime-leading to the problem of length-biased
sampling.6Secondly, at the time of survey,X was in the middle of a period (x3). Consequently,
the length of time on welfare that is recordedis for incomplete episodes which means that for
those in the sample, the periods are right censored. Right censoring would tend to understate
the welfare careers of young recipientsand recent entrants.7Finally, an importantcharacteristic
6. A comparableexample is the problem associated with analyzing the durationof hospital stay by patients.
Althoughmost patientsare hospitalizedfor only brief periods(2 to 5 days), a small numberof patients(for example those
with chronic diseases or the elderly) are hospitalizedfor far longerperiodsof time (severalweeks or months). A survey
conductedat any period of time is likely to over sample those who have been in hospital for a long time because these
patientsare likely to be in the hospital. Consequently,such a sample would be biased if used to provide informationsay
of the averagehospital stay.
7. One other problem with most other survey data is that they do not include informationabout the length of
periods of welfareuse before interview(for example ignorexl and x2 in Figure 1). Such data are poor measuresof wel-
WELFAREDEPENDENCY
Recipients
x1
x2
X3
y3
y2
Y1
to
episodes
t1
Calendar
951
time
of welfare
(survey)
use
Figure 1. Durationon Welfare
of durationdata is the propertyof durationdependence.Durationdependence simply means that
the durationof a welfareperiod is dependenton the length of the period at any particulartime.
Positive durationdependenceimplies thatthe probabilityof exit from welfareincreasesthe longer
a recipient has been on welfare, while negative durationdependencemeans that the probability
of exit decreasesthe longera recipienthas been on welfare.
These type of problems are common in the analysis of durationdata such as the length of
time on welfare. Using standardregressionanalysiswouldthereforenot be appropriate.A suitable
method for handlingthese types of data involves the applicationof hazardfunctionsas discussed
by Kiefer [11]. Hazardfunctionsfocus on conditionalprobabilitiesof an individualleaving a particular state in time t,, given that the individualhas been in that state up to time tn-. We can
thereforeexplain the factors that lead to long term welfareuse by estimatinga model that would
provide informationregardingthe likelihood of an individualwith particularcharacteristicsremainingon welfare. Hazardfunctionsare particularlyattractiveto modeling welfaredependency
because it is possible to specify functionalforms thatcapturedurationdependence.8
The likelihood of an individual remaining in the same state is obtained by estimating a
survival function. To estimate the survival functionof welfareparticipation,we first define the
cumulativedistributionof time on welfareas
F(t,Rt) = Pr(T < t)
(3)
whereRt is a vector of recipientand programcharacteristicsthataffect the decision to participate
fare dependencybecause they understatedurationon welfare. Othersurveysthat gatherinformationbetween two periods
(say between to and tl in Figure 1) have the problem of both left and right censoring. We are avoiding the problem
of left censoring by considering all previouslycompleted spells (xl and x2). Our survey data are therefore reasonably
good given that right censoring cannot be avoided if the sample is obtainedfrom those on welfare. The problemof right
censoringcould be reducedif data were for only those who had alreadyexited from welfare.Unfortunately,once a case is
closed (which is the basic assumptionwhen benefits are interruptedby non receipt), informationabout those individuals
is difficultto obtain because of privacylaws. Nevertheless,the estimationprocedurediscussedbelow adequatelyhandles
rightcensored durationdata.
8. For an excellent review of literatureon hazard functions, see Kiefer [11]. For more detailed discussion on
models and methods of durationdata, see Kalbfleischand Prentice[10] and Lawless [14].
952
MwangiS. Kimenyi
in the welfare program, F representsthe results of a series of participationdecisions from f2o
throughf2t. The distributionfunctionrepresentsthe probabilitythat the durationon welfare (T)
for an individualwith R, vector of characteristicsis < t.
The density functionassociatedwith the above distributionis given by
f(t,Rt)
= 8F(t,Rt)/8t
(4)
and the survival function for participationdefined as the proportionstill on welfare at time t is
given by
S(t,Rt) =
- F(t,Rt).
(5)
Finally, we define the hazard function (the hazardrate of exits from welfare) conditional on
participatingup to time T = t as:
A(t,Rt) = lim Pr(t < T < t + 8tIT > (t,Rt))/6t
(6)
A(t, R,) = f (t, Rt)/S(t,R,)
A(t,Rt) = -d ln(S(t,Rt))/St.
(7)
(8)
By integratingthe hazardfunction, the survivalfunctionbecomes
S(t,R,) = EXP(-
ft
A(u,Ru)du.
(9)
We can then investigatehow differentvariablesaffectthe probabilityof continuedwelfarereceipt
(survival)by estimatingdifferentdistributionsof the hazardmodel.
III. Model and Empirical Results
To investigate the determinantsof welfare dependency,we specify a general hazardmodel of
durationon welfare as follows:
WDEP=f (EDUC, CUNEMPL,NCHILD,RURAL,TEENFCB,RAFDCC,
RACE,SEXRATIO,
EMARR,REMP,TOUEY)
where
WDEP =
EDUC =
CUNEMPL =
NCHILD
RURAL
TEENFCB
RAFDCC
=
=
=
=
RACE =
cumulativenumberof monthsthatrecipienthas receivedAFDC benefits;
numberof years thatrecipienthas been in school;
relative change in the unemploymentrate since time of entry to time of
survey;
numberof childrenin the recipientunit;
1 if recipientresides in a ruralarea, and = 0 otherwise;
1 if recipientwas a teenagerwhen she had her firstchild and = 0 otherwise;
1 if recipients parentswere themselves welfarerecipients, and = 0 otherwise;
1 if recipientis nonwhite, and = 0 if recipientis white;
WELFARE DEPENDENCY
953
SEXRATIO= ratioof females to males by race and age group;
EMARR= 0 if recipienthas neverbeen married,and = 1 otherwise;
REMP = 1 if recipient was employed shortly before the first welfare period (six
months or less), otherwise = 0;
TOUEY= total other unearnedincome.
The dependentvariableWDEPmeasureslengthof time thatrecipientshave received welfare
benefits. Although most of the recipients in the sample had been on welfarerolls for less than 4
years (55%), about 21% have welfarecareersof 8 or more years.9
Probably the most importantdeterminantsof how long an individualremains on welfare
are those factors that increase the potentialearningsin the labor market.Most importantamong
these include the recipients' level of education,the change in the unemploymentrate since joining welfare rolls, and the expected loss of income resultingfrom exiting the system. Recipients
with higher levels of educationare expected to exit from relianceon welfaremore easily because
they are more likely to obtain employmentand receive higherwage offers than those with lower
levels of education. On the other hand, increasesin the unemploymentrate are expected to lead
to longer welfare durations.CUNEMPLmeasuresthe relativechange in the unemploymentrate
since time of entry to time of survey,calculatedas follows:
CUNEMPL= [unemploymentrate at time of survey/
unemploymentrate at the time of entry].
If CUNEMPLis greaterthan 1, then the unemploymentrate has on average increased since the
time the recipient startedreceiving welfare benefits. Our argumentis simple: a recipient who
enters the welfare rolls because of a downturnin the economy is not likely to exit unless the
unemploymentrate falls below the level it was at the time of entry.Thus, those who enter when
unemploymentrate is very low are likely to remain on welfare for longer periods than those
who enter when unemploymentis high. We therefore expect a positive relationship between
CUNEMPLand the length of time on welfare.'0
The numberof childrenmay affect the length of time on welfarebecause of severalreasons.
First, more children mean that the recipientmust seek alternativechild care services, which increases the opportunitycost of earningmarketincomes. Second, havingmore childrennecessarily
implies that more time will be taken off from work (for example when childrenare sick) which
translatesinto longer and more frequentepisodes of welfareuse. Finally,the numberof children
in a recipienthouseholdcan be used to proxy the expectedloss of income of exiting from welfare.
Welfarebenefits such as AFDC paymentsare based on the numberof childrenin the household
and a recipient's resources. If a recipient motherexits from welfare, she earns marketincomes
that are not based on the numberof childrenin the household. Consequently,the more children
there are in a recipient household, the largerthe expected loss of income from exiting from wel9. Many studies define welfaredependencyas merelyparticipationin welfareprogramswhile others have defined
dependencybased on the amountof welfare income received relativeto earnedincome [1; 3; 4; 26]. The present study,
on the other hand, defines dependency as the cumulativelength of time that a recipientreceives welfarebenefits (in this
case AFDC).
10. The unemploymentdata are based on the state unemploymentrate and are obtainedfrom variousissues of the
U.S. StatisticalAbstract [28]. It would be more accurateto use locationalspecific data (e.g., county unemploymentrate)
but unfortunatelywe do not have these data for all the years. Furthermore,use of such data would requireus to assume
thatrecipientshave not changedjurisdictionssince time of entry.
954
Mwangi S. Kimenyi
fare." We would thereforeexpect the numberof childrento increasedurationon welfare, that is,
reducethe probabilityof exit from welfare.
Even within the same state, employmentopportunitiesmay differsignificantlybetween rural
and urbanareas so thattheremay be systematicdifferencesin the durationof welfareuse resulting
solely from locational factors. In addition, because of larger seasonal variationsin job opportunities in rural areas (for example, farm work duringthe winter) than in urbanareas, we may
expect to find longer welfare careersamong ruralrecipientsthan urbanrecipients. On the other
hand, recipientsliving in innercities where employmentopportunitiesare practicallynonexistent
are likely to depend on welfare for extendedlengths of time. How area of residence affects welfare durationswill depend on the employmentopportunitiesavailableand we have no a priori
predictionas to how the variableRURALis relatedto dependency.
Some studies have suggested that one of the characteristicsof long-termAFDC use is the
at
age which the recipients have their first child, with teen mothershaving significantlylonger
welfaredurations.12
First, early teenage pregnanciesnecessarilyhindershumancapitalformation
which translatesinto lower earningpotential, and consequentlyloweremploymentopportunities.
Second, early teenage pregnanciesare often associatedwith the "cultureof poverty"which suggests that teen mothers are more likely to come from families that exhibit culturalpatternsthat
more readilyaccept welfareuse. The variableRAFDCCcontrolsfor whetherthe recipientwas an
AFDC child. We expect recipientsfrom welfarefamiliesto remainon welfarefor longer periods
themselves. This is because individualsfrom welfare families may have a lower disutility from
welfareuse than others.
Racial differencesin the length of time on welfaremay resultfrom the labor marketexperiences or from differences in attitudestowardwelfarereceipt. If, for example, blacks face labor
marketdiscrimination,other things equal, they would be expected to use welfarefor longerperiods. On the other hand, if there are racial differencesin attitudestowardswelfare, for example if
more blacks have been socialized into the system to the extent that they have a positive attitude
towardwelfare, we are likely to observe longerwelfarecareersamongstthese recipients.
One of the most importantroutes of exit from welfare is by way of marriage.Darity and
Myers [3; 4] have suggested that racial differencesin welfaredependencymay be partly due to
the natureof marriagemarkets.Specifically,black females are faced with a very unfavorablemarriage marketbecause there are relativelyfar fewer noninstitutionalmales comparedto the number
of females. The shortage of marriageablemales reduces the potential of exit from welfare by
way of marriage.The sex ratio variable(female/male) is includedin the model to control for the
role of marriagemarketson welfare use. We expect the sex ratio to have a positive impact on
welfare dependency. Simply, the more females there are relativeto availablemen, the lower the
probabilityof exit from welfare.'3
11. Note that the amount of AFDC would be a good proxy for the expected loss of income if the amountvaried
from recipient to recipient-say for example for recipientsin differentpoliticaljurisdictionssuch as states. Because all
data are from the same state, we do not include the amountof AFDC since this is capturedby the numberof childrenin
the recipienthousehold.
12. See for example papersdiscussed in Hopkins [9].
13. Sex ratio data for the state of Tennesseefor 1988 were not availableand the data used in this study are for the
entire country.We calculated the female sex ratio by race for 5-year age groupsbeginningwith the 15-19 age group for
females. A basic assumptionis that for females in a particularage group(for example 15-19), the relevantage groupfor
males from which the females are likely to get marriagepartnersis the next older group(20-24). The sex ratio data used
in this study seem to be appropriatebecause nationaldata correspondclosely with data for Tennesseefor the years when
such data are available(1980 census). We have to note thatmarriagemarketsaredifficultto analyzebecause it is not clear
how the relevantmarketshould be defined-county, city, state, and so on.
WELFAREDEPENDENCY
955
Table I. Descriptive Statistics
Variable
Mean
WDEP
EDUC
CUNEMPL
59.19
10.41
1.02
Standard
Deviation
58.75
2.25
0.33
NCHILD
1.89
1.08
RURAL
TEENFCB
RAFDCC
RACE
0.45
0.66
0.29
0.52
0.49
0.47
0.45
0.49
SEXRATIO
1.17
0.19
EMARR
0.45
0.48
REMP
0.32
TOUEY
18.30
0.46
74.76
Although most AFDC recipientsare single mothers, some of those in the sample were married and are currentlysingle because of divorce, separation,or the death of a spouse. Ever married females are expected to have fewer and shorterperiods of welfare use than never married
single female householders. The presence of a spouse in the householdat sometime in the past
could have reduced the demand for welfarebecause of spouse's income, or could have made the
household ineligible for receipt of AFDC benefits.
Welfaredependency is expected to be influencedby employmenthistoryof the recipient as
shown by the labor marketstatusvariable(REMP).This variableis an importantindicatorof the
extent to which a recipientis attachedto the labormarket,and may also indicateattitudestoward
work. We would thereforeexpect those recipientswho held a job shortlybefore enteringthe welfare system to experience shorterspells of welfare use. Finally, althoughthis study focuses on
AFDC recipiency,it is necessaryto investigatehow othergovernmenttransfersaffect the duration
on welfare. TOUEYmeasuresthe aggregateof all transfersthatthe recipientgets. The largerthis
amountis, the more attractiveit is for the recipientto remainon welfare.'4
Estimation and Results
Except for unemploymentand sex ratio variableswhich are calculatedfrom data obtained from
the U.S. StatisticalAbstract[28], all the other variablesare from the survey of AFDC recipients
describedearlier.Table I shows the means and standarddeviationsof these variables.
We estimatedthree distributionsof the hazardmodel-the Weibullhazard,Loglogistic hazard, and the lognormal hazard. The dependentvariablewas specified as right censored for all
observations.The maximumlikelihoodresultsare shown in TablesII (Weibull),III (loglogistic),
and IV (lognormal).'5By and large, the results are consistentfor the three distributionsand are
largely consistent to our expectations. In the discussionthatfollows, we focus only on the results
of the Weibullhazard.
Focusing on TableII, severalgeneralizationscan be made. First, durationon welfareis influ14. Total unearnedincome includes social security income, supplementalincome, and unemploymentcompensation. Food stampamountsare excludedbecause typicallyall recipientsreceive food stampsand the amountsare dependent
on household size. The unearnedincome could also be viewed as a proxy of expected loss of exiting from welfare.
15. For a detailed discussion on the estimationprocedureof hazardmodels and the implicationsof differentdistributions, see documentationof the LIMDEPcomputerprogramby Green [8].
956
Mwangi S. Kimenyi
Table II. DurationModels for Time on Welfare
Weibull Hazard
1
Constant
EDUC
CUNEMPL
NCHILD
RURAL
TEENFCB
RAFDCC
RACE
3.410
-0.0359
(0.116)
0.976***
(0.079)
0. 120***
2.344
-0.0294
(0.0121)
0.9350***
(0.0862)
0.1041***
0.0738
0.0715
(0.0649)
0.1038*
(0.0596)
0.9847**
(0.4496)
0.1650**
(0.0678)
-0.04288
REMP
(0.0630)
-0.3005***
(0.0581)
Log-likelihood
2.309
(0.028)
EMARR
N
3
(0.028)
SEXRATIO
TOUEY
2
0.0002533
(0.000390)
895
-569.16
(0.0658)
0.1596***
(0.0620)
0.9904**
(0.4388)
-0.0210
(0.0800)
0.9794***
(0.200)
-0.2447***
(0.0594)
0.000078
(0.000390)
935
-550.54
-0.0300
(0.0119)
0.9371***
(0.0853)
0.1041***
(0.0285)
0.0651
(0.0614)
0.1562***
(0.060)
0.9835**
(0.438)
0.9894***
(0.166)
-0.2465***
(0.0585)
0.000079
(0.000390)
935
-550.57
Notes: N less than 991 because of missing data or miscoded responses. Standarderrorsin parentheses.Asterisks
denote significanceat the 1 percent (***), 5 percent(**), and 10 percent(*) levels.
enced significantly by economic conditions such as the unemployment rate and the labor market
attachment of recipients. Those who enter the welfare rolls when the unemployment rate is low
are likely to remain on welfare for an extended period of time. On the other hand, those who
had recent employment history before the start of a period of welfare use are likely to use welfare
for only short durations. The number of children increase duration on welfare as expected. This
is because the per capita loss from exiting the welfare system is higher for those with more children. In addition, provision of child care services increases the cost of getting off welfare. The
cultural adaptation variables-teenage births and history of welfare use in the family increases
the duration of time on welfare. Having been married before, area of residence,l6 and amount
of other government transfers do not significantly affect duration on welfare. An interesting but
16. The survey instrumentclassified recipient'sarea of residence into seven categories: 1. majorurbancenter, 2.
suburbsof a majorurbancenter, 3. medium size city, 4. suburbsof a mediumsize city, 5. small town, 6. suburbsof a
small town, and 7. ruralarea. In the presentstudy we classified 1-4 as urbanand 5-7 as rural.Changingthe classification
so thatonly categories 1 and 2 are consideredurbandid not changethe results.
WELFAREDEPENDENCY
957
Table III. DurationModels for Time on Welfare
Loglogistic Hazard
1
Constant
EDUC
CUNEMPL
NCHILD
RURAL
4.755
-0.0367
(0.029)
0.968***
(0.212)
0.122*
(0.069)
0.0321
(0.1678)
TEENFCB
RAFDCC
RACE
0.1132
(0.1555)
0.8942
(1.196)
0.1827
(0.1671)
SEXRATIO
EMARR
-0.06758
(0.1627)
REMP
-0.2813*
TOUEY
N
Log-likelihood
(0.1540)
-0.0000432
(0.000931)
895
-141.91
2
3.548
-0.0293
(0.0302)
0.8986***
(0.2256)
0.1257*
(0.070)
0.0229
(0.1698)
0.1714
(0.1600)
0.9800
(1.2055)
-0.0322
(0.2010)
0.94357**
(0.492)
-0.2228
(0.1560)
0.000268
(0.000930)
935
-139.42
3
3.600
-0.0302
(0.0295)
0.9019**
(0.2249)
0.1755*
(0.0702)
0.0134
(0.1583)
0.1665
(0.156)
0.9701
(1.203)
0.9814**
(0.405)
-0.2316
(0.1550)
0.000267
(0.000930)
935
-139.43
Note: See Table II.
rather surprising result is that education has no effect on the duration on welfare. This is probably
because for most low wage employment, number of years that a recipient has been in school may
be a poor signal for job performance.
Finally, the effects of race and sex ratio on welfare dependency are interesting. The coefficient on the sex ratio variable is positive and statistically significant. If we omit the sex ratio,
results show that being black tends to increase the likelihood of a recipient remaining on welfare.
When both sex ratio and race variables are included in the model, however, the coefficient for
race becomes negative (but insignificant) indicating that white recipients have a lower probability
of exiting from welfare if sex ratio is held constant across the races. This result gives credibility
to the Darity and Myers [3; 4] conclusions that it is not welfare use that leads to female headship, but rather it is female headship that leads to welfare dependence, and that female headship,
particularly among black females, is a result of a shortage of males.
Mwangi S. Kimenyi
958
Table IV. DurationModels for Time on Welfare
LognormalHazard
1
Constant
EDUC
CUNEMPL
NCHILD
RURAL
TEENFCB
RAFDCC
RACE
3.843
-0.0361
(0.168)
0.958***
(0.116)
0.119***
(0.039)
0.0498
(0.0948)
0.1104
(0.0874)
0.9781***
(0.0701)
0.1733*
(0.0970)
REMP
TOUEY
N
Log-likelihood
2.677
-0.0293
(0.0169)
0.9179***
(0.1228)
0.1090***
(0.039)
0.0432
(0.0944)
0.1668*
(0.0888)
0.9490***
(0.0690)
-0.05562
(0.0920)
-0.28609***
(0.0860)
0.0000843
(0.000541)
895
-202.67
3
2.723
-0.0301
(0.0166)
0.9209***
(0.1221)
0.1088***
(0.0398)
0.0346
(0.0880)
0.1624*
(0.086)
0.9597***
(0.0691)
-0.0287
(0.1128)
1.0899***
(0.281)
SEXRATIO
EMARR
2
1.0502***
(0.229)
-0.2314***
-0.2343***
(0.0862)
-0.000118
(0.000530)
-0.000117
(0.000530)
935
198.25
(0.0853)
935
-198.31
Note: See TableII.
IV. Conclusion: Dependency Theories and Public Policy
The results presented in this paper suggest that long-term welfare use is a result of economic
constraints, (such as the lack of employment opportunities), rational choice (which focuses on
considerations of the expected loss of exiting from welfare), and cultural adaptation. In addition,
we find marriage markets to be very important particularly in explaining the frequently observed
racial differences in duration on welfare.
The results have some important public policy implications in relation to welfare policy. First,
the results can be used to target specific policies to potential long term users. For example job
training related programs could focus on teen mothers who have no prior work history. Secondly,
it does appear that because of the child care costs, the choice not to work appears most rational.
Because extended welfare use is likely to be associated with negative duration dependence, it may
be necessary for the government to design child care programs so that welfare mothers can enter
the labor market and thereby avoid the multiplicative effects of prolonged welfare use. At the
WELFARE DEPENDENCY
959
same time, more strict enforcementof child supportpolicies could help some mothersenter the
labor marketwhich would translateinto reducedwelfaredependency.
It may appearas though the problemof unfavorablemarriagemarketfor black females may
be out of the scope of public policy. However,it is necessaryto realize that it is because of their
economic statusthatmany black males are involvedin activitiessuch as drugs,crime, alcoholism,
and so on, which make them unsuitableas marriagepartnerswhile others end up in institutions
such as prisonsand mentalhospitals, and yet othersarekilled at an earlyage, reducingthe pool of
males in the population.Unless those trendsare reversed,black females will continueto be faced
with unfavorablemarriagemarkets.This will translateinto increasesin femaleheadedhouseholds
and consequently increased welfare dependency.Thus it appearsthat welfare policies to reduce
dependencyparticularlyamongblack females, shouldalso focus on the employmentopportunities
of black males, drug treatmentprograms,and so on.
Finally, the fact that educationhas no effect on the durationon welfare suggests that education does not improve the employmentprospectsof recipients.'7This is disturbingbecause it
suggests that the high drop out ratesin the Southrepresentrationalbehavior.Welfarereformpolicies should thereforebe accompaniedby educationreformsuch thatremainingin school improves
employmentprospects of the poor.
17. The numberof years in school does not tell us whetherthe recipienthas a high school diplomaor not. Having
a high school diploma may actuallybe the relevantsignal used by employersin hiringdecisions. Kimenyi [12] finds that
having a high school diploma significantlyreduces the durationof food stamp receipt. Results based on the numberof
years a recipient attendedschool as the measureof educationshouldthereforebe interpretedwith caution.
References
1. Barr, N. A. and R. E. Hall, "The Probabilityof Dependenceon Public Assistance."Economica, May 1981,
109-23.
2. Coe, RichardD., "WelfareDependency:Fact or Myth." Challenge,September-October1982, 43-49.
3. Darity,William A. Jr., and SamuelL. Myers, Jr., "Changesin BlackFamilyStructure:Implicationsfor Welfare
Dependency."AmericanEconomicReview, May 1983, 59-64.
4.
and
, "Does WelfareDependency Cause Female Headship?The Case of the Black Family."
Journalof Marriageand the Family, November 1984, 765-79.
5. Duncan, Gregg J., MarthaS. Hill, and Saul Hoffman, "WelfareDependenceWithinand Across Generations."
Science, January1988, 467-71.
6. Gallaway,Lowell and RichardVedder,"PayingPeople to be Poor."NationalCenterfor Policy Analysis, Policy
ReportNo. 121, 1986.
7. Glicken, Morley D., "TransgenerationalWelfareDependency."Journal of ContemporaryStudies, Summer
1981, 31-41.
8. Green, William. LIMDEP:Version5. Mimeo., New York, 1988.
9. Hopkins, Kevin R. WelfareDependency:Behavior, Cultureand Public Policy. Alexandria, Virginia: Hudson
Institute, 1987.
10. Kalbfleisch, J. D. and Ross L. Prentice. The StatisticalAnalysisof FailureTimeData. New York:John Wiley
and Sons, 1980.
11. Kiefer, Nicholas M., "Economic DurationData and HazardFunctions."Journalof EconomicLiterature,June
1988, 646-79.
12. Kimenyi, Mwangi S. "WelfareDependence: A HazardFunctionAnalysis." WorkingPaper, Departmentof
Economics and Finance, The Universityof Mississippi, 1990a.
. "WelfareUse, WelfareDependency,and LaborMarketsin the Deep South."WorkingPaper, Depart13.
ment of Economics and Finance, The Universityof Mississippi, 1990b.
14. Lawless, J. F. StatisticalModelsand Methodsof LifetimeData. New York:JohnWiley and Sons, 1982.
15. Lewis, Oscar. Five Families:MexicanCase Studiesin the Cultureof Poverty.New York:Basic Books, 1959.
960
MwangiS. Kimenyi
16.
. A Studyof Slum Culture:Backgroundsto LaVida.New York:RandomHouse, 1966.
17. Murray,CharlesA. Losing Ground:AmericanSocial Policy, 1950-1980. New York:Basic Books, 1984.
18.
and Deborah Laren. "Accordingto Age: LongitudinalProfilesof AFDC Recipients and the Poor by
Age Group."Preparedfor the WorkingSeminaron the Familyand AmericanWelfarePolicy, September,1984.
19. O'Neil, JuneA., Douglas A. Wolf, LaurieJ. Bassi, andMichaelHanan.AnAnalysisof Timeon Welfare.Report
to the Departmentof Health and HumanServices, Washington,D.C.: UrbanInstitute, 1984.
20.
, Laurie J. Bassi, and Douglas A. Wolf, "The Durationof WelfareSpells." Reviewof Economicsand
Statistics,May 1987, 241-48.
21. Plant, Mark W., "An EmpiricalAnalysis of WelfareDependence."AmericanEconomicReview, September
1984, 673-84.
22. Plotnick, Robert D., "Turnoverin the AFDC Population:An Event History Analysis." Journal of Human
Resources,Winter 1983, 65-81.
23. Rank, Mark R., "Exiting From Welfare:A Life-TableAnalysis." Social Service Review, September 1985,
358-76.
24.
, "Racial Differencesin Lengthof WelfareUse." Social Forces, June 1988, 1080-101.
25. Rein, Martinand Lee Rainwater,"Patternsof WelfareUse." Social ServiceReview, December 1978, 511-34.
26. Robins, Philip K., "ChildSupport,WelfareDependency,andPoverty."AmericanEconomicReview,September
1986, 768-88.
27. Rydell, Peter C., Thelma Palmeria,GerardBlasis, and Dan Brown. WelfareCaseloadDynamics in New York
City. New York:The Rand Institute, 1974.
28. U.S. Departmentof Commerce, Bureauof the Census, StatisticalAbstractof the UnitedStates, Washington,
D.C.: U.S.G.P.O., (variousissues).