Incorporating Cohabitation When Modeling Fertility of
Young Adults
Maciej MisztalUniversity of North Carolina at Chapel Hill
March 10, 2011
1
1 Introduction
Marriage is regarded as indispensable in economic models of fertility. Any estimation without
marriage or a proxy would undoubtedly lead to omitted variable bias. However, there is reason
to believe the nature of marriage is changing within industrialized countries as individuals
increasingly choose to have marriage-like relationships without the legal and formal commitment.
While fairly uncommon fifty years ago, cohabitation has risen to be predominant form of union
in America at the turn of the century (Schoen et al., 2007). Out of wedlock births rose to a
third of all births over the same period. These shifts in the establishment of families are seen
through virtually every ethnic and socioeconomic group Kennedy and Bumpass (2008). These
trends are interwoven with changes in fertility.
Incorporating cohabitation into fertility models may help explain these trends. There has
been an overall delay in age of first birth since the 1950’s, but it has occurred almost completely
among educated women (Schoen et al., 2007). If age of first birth has has remained relatively
constant, the the increase in cohabitation and related decrease in marriage would lead to the
increased the amount of children born out of wedlock. If cohabitation implies smaller commit-
ments and weaker bonds, these children may grow up in less stable households. Cohabitation
also has a significantly different impact on the workforce. While single women work consider-
ably more than married women, their labor force participation is in some datasets statistically
indistinguishable to that of cohabiting women (Gemici and Laufer, 2010). Further more, there
are changes in trends within cohabiting couples. The portion of women likely to be married to a
partner with whom they have been cohabiting for more than five years is falling (Schoen et al.,
2007). If these trends are persistent, they will be more apparent among younger generations.
Economists have been slower than other social scientists to focus on cohabitation and changing
transitions in youth. A large part of this reason is sparse, incomplete, or unreliable data. The
NLSY97 is one more recent dataset that attempts to more carefully define and document co-
habitation alongside marriage. Relationships are tracked on a monthly basis. Additionally, the
data allow for the careful examination of the schooling decisions of the youth and the precise
timing of their entry into the workforce.
2
This paper intends model the schooling, labor, marriage and cohabitation decision on fertil-
ity. The focus will be on young women between the ages of 14 to 28 in the NLSY97 cohort. The
fertility decision is restricted to mothers that choose to keep their children in their household.
The decision not to raise the child may suggest that the pregnancy was less planned and more
unexpected The outcomes we model are likely correlated through unobserved time-varying and
permanent characteristics. We will estimate using the discrete random effects method that will
help provide unbiased estimates of fertility decisions on future wages.
This paper intends to add to the literature in three ways. First, while the NLSY97 dataset
has been used widely in the study of adolescents, its rich data concerning further education and
full time labor decisions is only beginning to be released. Comparing estimates to the widely
used NLSY79 may be useful in addressing which aspects of the life cycle decisions have changed
over the last generation. Second, it is able to to more accurately track cohabitation in addition
to marriage when addressing the fertility decision. Finally, in contrast to studies using the
National Longitudinal Study of Adolescent Health (Add Health), this model is able to track
detailed year by year data on a wide variety of economically relevant factors. This study aims
to take advantage of these aspects of the data to gage the impact of changing employment and
living patterns on fertility.
2 Literature Review
The paper grazes topics that have been examined in depth in the literature. For models of
labor supply, consider Blundell and Macurdy (1999). Although our model only seeks to predict
outcomes, not preferences, much work has been done in the dynamic structural framework to
estimate sequential joint decision of young women. van der Klaauw (1996) creates a female
life cycle model that is able to structurally estimate women’s labor contributions along with
their (potential) marriage and divorce decisions with respect to education, race and potential
earning of each spouse. He finds the utility of marriage is increasing in the husband’s wage
but decreasing in the women’s wage. This suggests the theory of specialization to explain the
marriage premium. The marriage may be less useful and binding for both parties as their
3
contributions to the household become substitutes.
Keane and Wolpin (2006) provide perhaps the most complete structural model of women’s
life cycle decisions. They use a model that incorporates schooling, marriage, welfare acceptance,
labor, and fertility. They use the NLSY79 cohort through 1990. They confirm unobserved
heterogeneity is crucial to the understanding of younger women’s decisions in particular. They
find that causes of discrepancies in the outcomes of minority and white women are due to a
number of complex interrelated factors, unobserved to the econometrician. No one decision is
responsible for less productive outcomes in the others. They also conclude that the distribution
of unobserved “types” of women is imperative to understand the differences in outcome within
and between minorities as well as whites. This was similar to the result of their research
concerning youths dropping out of school (Eckstein and Wolpin, 1999), which also confirmed that
certain permanent unobservable types in every group where simply more predisposed towards
dropping out or staying in.
They take into account that the trends regarding cohabitation support the theory of special-
ization. Gemici and Laufer (2010) use NLSY97 to form a dynamic model of household creations
and dissolution, focusing on the matching aspect. Their model suggests cohabitation as a third
choice between marriage and remaining single. They then predict policies that would change the
cost of divorce. Their model suggests that marriage is optimal when the specialization within
the household is greater.
Musick (2007) uses data from the NSFG to study intended and non intended pregnancies
for married and cohabiting partners. She uses Applied Maximum likelihood to jointly model
intended pregnancies, unintended pregnancies, and the entrances into marriage and cohabiting
relationships, in an effort to control for endogeneity. After introducing controls for unmeasured
heterogeneity, she finds a significant reduction in the effect of cohabitation on birth, suggesting
that many of the women who give birth in cohabiting relationships are of the marrying “type”
but have a hard time finding a willing marriage partner.
This paper is related to a series of papers using the methods established by Lillard (1993).
It creates a set of simultaneous hazard functions that are interdependent. In the first paper
4
they are conception and marriage. The models further allow multiple instances and failures of
an event, in this case marriage or conception. The models are estimated using MLE and are
dependent on specific parametric assumptions. They are identified by the fact that the data have
events overlapping in different patterns. The consistent result of models using this technique is
the endogeneity bias involved in simple predictions of relationships on fertility decisions among
young women.
There is evidence in the literature that the poor schooling and wage outcomes associated
with teenage mothers are driven by omitted variable bias and unobserved heterogeneity. Geron-
imus and Korenman (1992) uses PSID, NLSYW, and NLSY79 data to examine teenage mothers
as compared to their sisters. They find evidence that the negative consequences of childbearing
are the result of family characteristics as well as unobserved individual characteristics. They
suggests that the prevalent cross-sectional literature unfairly attributes substandard income to
youth pregnancies. Hoffman et al. (1993) considers evidence of the cost of teenage pregnancy
using PSID information. The authors use the sisters of teen mothers to control for background.
They estimate the cost of teenage pregnancy is not much less significant than is usually assumed
and it is the disadvantaged background of most teen mothers that drives their poor outcomes.
Upchurch et al. (2002) suggest that motherhood outside of marriage may be relatively desir-
able for economically disadvantaged women who do not expect potential husbands to contribute
much, and other groups for whom marriage-appropriate men are scarce. They also want to in-
corporate multiple births into their model, including for those women who have been previously
married. The paper concerns itself only to nonmarital conception leading to live births. They
intend to quantify the effects of education marriage and marital dissolution and prior fertility on
nonmarital childbearing and compare models that control for heterogeneity but not endogeneity.
The results complement Brien et al. (1999). Non-black women who drop out of school
before they conceive are more likely to get pregnant than their enrolled counterparts, although
there is no long term effect. For all women, having been married decreases the probability of
nonmarital pregnancy after divorce. Those that are predisposed towards marriage are also pre-
disposed towards nonmarital pregnancy. Prior childbearing experience both within and outside
5
of marriage will also significantly affect further fertility decisions. Finally, race and having come
from a disadvantaged background have significant effects on the joint and sequential decisions
of women.
This paper will add to these strands of literature by combining both labor/schooling deci-
sions with the union/fertility decisions. . We control for the endogeneity by jointly estimating
the equations. We also control for time varying and permanent heterogeneity. The latter will
control for “types”. However, avoid addressing directly the decision to keep the child. The
decision is a theoretical quagmire and data are widely believed to be unreliable. Literature
of the supply side of adoption is sparse. Medoff (1993) runs a multinomial logit on aggregate
data for young women at the state level, using state policies as exogenous shocks. His results
suggest that even large changes in adoption and abortion policy would not significantly change
the number of children given up for adoption.
3 Data
The NLSY97 dataset follows a nationally representative sample of individuals across the country
born between 1980 and 1984. Data were first collected in 1997, when the respondents were
between the ages of 12 and 14. During the first year of the interview, extensive data were
collected on the household in which the respondents grew up, including an interview with the
caretakers. Data are now available for 12 rounds, into their mid-twenties. The data are designed
primarily with the labor decisions in mind, and keeps yearly updates of all their employment
. There is also information about schooling, including information about the high schools and
colleges as well as training programs.
Crucial to the purposes of this study is the information about relationship and fertility
decisions. Data are collected starting at the age of 16 and goes into detail about frequency and
length of relationships, as well as intentions of pregnancy and use of contraception. At sixteen,
any relevant event history prior to age sixteen retroactively obtained. Special attention is paid
to tracking the length and number of cohabiting relationships. To help ensure truthful answers
6
for topics such as sexual activity or drug use, certain parts of the interview are self administered
via laptop and headphones.
The data I am currently using are only geocoded for region and presence in a Metropolitan
Statistical Area. I will soon gain access to zip code and county data. This will allow me to at
minimum control for state policies and county policies, as well as schools. I do have general
statistics regarding the schools and peer groups in 1997.
I also have information on the amount of time and source of childcare, down to who drives
the children to school. Women without children in the home are regularly asked if they have
relatives or close friends who would be willing to take care of their hypothetical child. This
information is not collected in every period.
In addition to the Census data regarding the makeup of states and counties, there are
several policy variables that can be used for identification. During the formative years of the
participants, the Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA)
aimed to change the incentives related to teenage pregnancy. Around the same time, many
states with below average rates of university attendance implemented large scale merit-based
scholarships. Also, in 1997 legislation passed that encouraged states to extend Medicaid coverage
of low-income youth. In all these cases the policy shift is exogenous and impacts the NLSY
cohort. Because the implementation was done by state and varied over time, it will contribute
to identifying the model.
The sample consists of 4,385 females with potentially 12 years of data. Childborn refers to
years when the individual has a non-adopted child added to their household. Biologicalchildren
refers to those children.Cohabiting and Married are not mutually exclusive. Enrolled includes
periods when individuals attend college and higher education. The years schooling of parents
refers to those parents who raised them as of 1997, not necessarily their biological parents.Parents
religiosity is self reported as at least 500 on a 600 point scale. You will note year of menarche is
unreliable, and will not be used to control for fertility. The differing number of observations is
a result of varying ability to reconstruct absent data. At this point, data are kept on those who
have missed less than 3 consecutive periods, as most of the current data can be reconstructed
7
Table 1: Summary statistics
Variable Mean Std. Dev. Min. Max. Nchildborn 0.066 0.249 0 1 52620Cohabiting 0.245 0.43 0 1 46653Married 0.13 0.337 0 1 46653BiologicalChildren 0.408 0.804 0 8 46665Enrolled 0.505 0.5 0 1 46527Age 20.232 3.82 12 29 46715Urban 0.768 0.422 0 1 46455South 0.395 0.489 0 1 46508West 0.227 0.419 0 1 46508Northeast 0.163 0.37 0 1 46508Enrolled 0.505 0.5 0 1 46527HSDeg 0.512 0.5 0 1 46527CollegeDeg 0.093 0.291 0 1 46527AdvDeg 0.006 0.075 0 1 46527MomYearsSchooling 12.496 3.245 1 95 46884ParVReligious 0.365 0.481 0 1 34464Black 0.266 0.442 0 1 52620Hispanic 0.211 0.408 0 1 52620CitizenAtBirth 0.76 0.427 0 1 52620DadYearsSchooling 12.942 3.629 2 95 32424menarche 1994.469 2.079 1980 2001 51168
as they return. The appendix includes attrition rates
Income, assets, government assistance and other basic household information are prob-
lematic due to transition from guardians’ care to independence. Any advice on handling this
transition is welcome. Employment is being added. The data allow for several jobs in succession
and at once every year. Specific stop and start days are given for each job.
4 Theoretical Model
The empirical model is not structural, does not predict preferences, and simply models outcome.
We start with a general life cycle interpretation of the neoclassical model incorporating fertility
decisions, as set forth in Becker et al. (1960). Women will be followed from age 14. The woman
will have a budget constraint and a time constraint. She will strive to maximize current utility
subject to them. The woman derives utility from consumption goods Cit and leisure time Lit
8
The woman may also derive utility from her marital/cohabitation status mit if she is mar-
ried/cohabiting at time t. This utility may depend on the partner’s compatibility and charac-
teristics, and transitioning into and out of either state is presumed to be costly. The mother
derives utility from her children nit and has a one time additional positive shock when the
child is a newborn nbit. There is also utility from moving away from the parental household
qit.1 The Utility each period may also be influenced by time varying preference parameters over
work, schooling, and marriage, as well as time invariant preference parameter. Finally, it will
be influenced by age and demographic characteristics xit.
Uit = U(Cit, Lit,mit, nit, nbit, qit, θi, µit, xit) (1)
Time available to the woman is fixed in each period and can be distributed towards hours
worked hit, and schooling sit if one attends school(or training) in this period T s. What is left is
attributed to leisure and non market activities. If the model incorporates childcare, then nit will
interact with mit, qit and rit, availability of near by relatives, within the utility function. We
expect cohabiting partners to not provide as much childcare as husbands. Time for childcare is
implicitly incorporated into Lit.
T = hit + T s ∗ sit+ Lit(Nit,mit, rit, qit) (2)
Cit is determined within the budget constraint. For simplicity I assume the budget is
binding, although I do have data for assets and debts in the relevant periods. The budget is
made up of hourly wages multiplied by hours worked hit . Wages w(Sit, τit) are influenced
by Sit, the stock of Education, and τit,the stock variable of indicating tenure at a particular
job. qrit represents cost of living away from home. Unearned income can be contributed by an
individual’s partner Imit , one’s parents (Irit). There is also a cost to schooling per period, tit
and child care costs c(Nit, nit, ). I begin by assuming that income from partner and parents is
1Given the transition in the data, I may have to model this in the future. it is not currently in my empiricalmodel.
9
treated as exogenous. git represents government assistance2. The welfare eligibility rules would
be strict and depend on state and other factors reflected by actual welfare requirements. The
equation for eligibility is not yet modeled. Total income uncertain in every period, regardless
of labor choices. The error term may manifest itself as a one time unexpected cost or some
significant gift. It is represented by εit. It should be noted that the female discovers the value
of the random error after the decision to work in that period has been made. Consider living
with
Cit = hit ∗ w(Sit, τit) − qrit + git + Imit + (Ipit) − psit ∗ T s ∗ sit+ c(Nit, nit) + εit (3)
We allow cohabitationmit = 1 and marriage mit = 2 status to be a probabilistic event
which is dependent on the following:
Pr(mit = 0, 1, 2) = (Nit, Sit, wit, θi, qrit, xit) (4)
We will incorporate a stock variable Mit that increases with tenure of the partner. This
stock variable is presumed to be larger in the case of marriage. The factors that affect the
probability of marriage will also be relevant in case of divorce or separation. We would expect
the cost of a divorce with a husband to be larger than the cost of a separation from a cohabitor.
The timing of the model is such. In the beginning of the period the woman observes a
wage offer and a potential partner. In addition she knows the exogenous variables in the model
including non-earned income, preference parameters as well as individual and area characteris-
tics. The woman also observes the current state of stock variables including children, tenure,
and schooling as well as the resulting wage. She is also aware for the value of the her choice
2Also, not modeled empirically.
10
variables in the last period. So her information at the beginning of the period is
Zit = (Mit, IMi,t−1, ri,t−1, I
Pi,t−1, Gi,t−1, θi, µit, Xit, Bit, Sit, τit, Nit, wit, si,t−1, Hi,t−1, bit) (5)
She then decides how to allocate her time into work Hit, schooling sit, and leisure Lit and
whether to give add a child in the next period3 Afterwards, the present value of lifetime utility
associated with to income εit is realized, the woman decides whether or not to add another child
to the family Nit+1 = Nit+ 1.
Presumably, the woman would then maximize her present lifetime utility function.
5 Empirical Model
During the young adult period, there is great amount of endogeneity surrounding any decision.
If we are to focus on the period between the mid-teens and the mid-twenties, we may not be
able to account for the complex relationships between the decision to continue education, move
in with a partner, start a family or take care of one’s child.
The goal is to model the spacing of children for young women while taking into account
the endogenous decisions of schooling, working, and relationship status (married or cohabiting).
Furthermore, the decisions a woman makes in any period are dependent on progression of
decisions in early periods. I will follow the women from age 14, at which point we assume they
are in school and have no children4. The model will follow them through their mid-twenties,
assuming the individuals do not attrit. The timing of the model:
1.The female observes the history of all the decision variables from the previous periods as
well as her personal characteristics and all community-level conditions.
2. The female observes a wage draw wt.
3I am considering whether how/if to justify fertility as a decision decided after the joint decisions.4I have dropped observations not fitting this criteria.
11
3.The female makes a simultaneous decision over employment, marriage/cohabitation, and
schooling.
4. The woman then decides whether to accumulate another child for the next period5
The most tractable would be to simply include the lagged decisions from the previous period.
For the purposes of the following outline, we consider the possibility of having accumulated stock
variables that capture durations of decisions.
• Allow Eijt to summarize the entire history of hours worked for person i in county j in up
till period t. In the same way regard Mijt−1, Sijt−1, and Nijt−1 as containing the history
of marriage/cohabitation, schooling/training, and number of children,respectively.
• Xijt includes all exogenous characteristics, including ,race, gender, and so on. Spousal
income, money and childcare from relatives, government transfers are also considered
exogenous.
• Consider Pjit to be the vector of community characteristics for each of the above decisions,
including prices, tuition, male/female ratio, average rent, welfare requirements and so on.
• Errors can be decomposed into a time varying individual component κit ,a permanent
individual component ωt and an idiosyncratic component θit.
If we assume extreme value distributed idiosyncratic terms, we can estimate simultaneously the
log odds of a woman being employed. The employment decision is a multinomial logit estimating
working full-time (eijt = 2) or working part-time (eijt = 1) relative to not working (eijt = 0) :
5I am still deciding whether to keep this decision separate. Advice is welcome.
12
ln
[p(eijt = de)
p(eijt = 0)
]= α0 + α1Eijt−1 + α2Mijt−1+
α4Sijt−1 + α5Nit−1 + α6Pijt + α7Xijt + α8κit + α9ωi
de = 1, 2 (6)
The employment decision is jointly decided along with schooling and marriage cohabitation.
The binary logit for an additional year of schooling/training vs not having an additional year:
ln
[p(sijt = 1)
p(sijt = 0)
]= β0 + β1Eijt−1 + β2Mijt−1+
β4Sijt−1 + β5Nit−1 + β6Pijt + β7xijt + β8κit + β9ωi (7)
The marriage equation is a multinomial logit of being married (mijt = 2) or cohabiting
(mijt = 1) relative to being single (mijt = 0) :
ln
[p(mijt = dm)
p(mijt = 0)
]= δ0 + δ1Eijt−1 + δ2Mijt−1+
δ4Sijt−1 + γ5Nit−1 + δ6Pijt + δ7xijt + δ8κit + δ9ωi
dm = 1, 2 (8)
6 Preliminary Estimation
The logits in the appendix are naive and do not account for endogeneity. They are intended to
show strong correlation among the variables of interest and do not imply direct causal relation-
ships. As in the model, all “independent” state variables are lagged. Dummy variables for year
13
are suppressed.
14
References
Becker, Gary S., Duesenberry, James S. and Okun, Bernard. “An Economic Analysis of
Fertility”. In “Demographic and Economic Change in Developed Countries”, NBER Chapters,
pp. 225–256. National Bureau of Economic Research, Inc, 1960.
Blundell, Richard and Macurdy, Thomas. “Labor supply: A review of alternative ap-
proaches”. In O. Ashenfelter and D. Card, eds., “Handbook of Labor Economics”, vol. 3 of
Handbook of Labor Economics, chap. 27, pp. 1559–1695. Elsevier, April 1999.
Brien, Michael J., Lillard, Lee A. and Waite, Linda J. “Interrelated Family-Building
Behaviors: Cohabitation, Marriage, and Nonmarital Conception”. Demography, 1999. 36 (4),
pp. 535–551. ISSN 00703370.
Eckstein, Zvi and Wolpin, Kenneth I. “Why Youths Drop Out of High School: The Impact
of Preferences, Opportunities, and Abilities”. Econometrica, 1999. 67 (6), pp. pp. 1295–1339.
ISSN 00129682.
Gemici, Ahu and Laufer, Steve. “Marriage and Cohabitation”. draft, New York University,
Feb. 2010.
Geronimus, Arline T. and Korenman, Sanders. “The Socioeconomic Consequences of
Teen Childbearing Reconsidered”. The Quarterly Journal of Economics, 1992. 107 (4), pp.
1187–1214. ISSN 00335533.
Hoffman, Saul D., Foster, E. Michael and Jr., Frank F. Furstenberg. “Reevaluating
the Costs of Teenage Childbearing”. Demography, 1993. 30 (1), pp. 1–13. ISSN 00703370.
Keane, Michael P. and Wolpin, Kenneth I. “The Role of Labor and Marriage Markets,
Preference Heterogeneity and the Welfare System in the Life Cycle Decisions of Black, His-
panic and White Women”. PIER Working Paper Archive 06-004, Penn Institute for Economic
Research, Department of Economics, University of Pennsylvania, Jan. 2006.
Kennedy, S. and Bumpass, L. “Cohabitation and Childrens Living Arrangements: New
Estimates from the United States”. Demographic Research, 2008. 19 (47), pp. 1663–1692.
doi:10.4054/DemRes.2008.19.47.
15
Lillard, Lee A. “Simultaneous equations for hazards : Marriage duration and fertility timing”.
Journal of Econometrics, 1993. 56 (1-2), pp. 189 – 217. ISSN 0304-4076. doi:DOI:10.1016/
0304-4076(93)90106-F.
Medoff, Marshal H. “An Eepirical Analysis OF Adoption”. Economic Inquiry, 1993. 31 (1),
pp. 59–70. ISSN 1465-7295. doi:10.1111/j.1465-7295.1993.tb00866.x.
Musick, Kelly. “Cohabitation, nonmarital childbearing, and the marriage process”. Demo-
graphic Research, April 2007. 16 (9), pp. 249–286.
Schoen, Robert, Landale, Nancy S. and Daniels, Kimberly. “FAMILY TRANSITIONS
IN YOUNG ADULTHOOD.” Demography, 2007. 44 (4), pp. 807 – 820. ISSN 00703370.
Upchurch, Dawn M., Lillard, Lee A. and Panis, Constantijn W.A. “Nonmarital
Childbearing: Influences of Education, Marriage, and Fertility”. Demography, 2002. 39 (2),
pp. 311–329. ISSN 00703370.
van der Klaauw, Wilbert. “Female Labour Supply and Marital Status Decisions: A Life-
Cycle Model”. Review of Economic Studies, April 1996. 63 (2), pp. 199–235.
16
Table 2: The following is intended to reassure about the stability of cohabitationTotalCohabitationslagged
TotalCohabitations 0 1 2 3 4 5 6 7 Total
0 42,437.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 42,437.01 3,720.0 13,680.0 0.0 0.0 0.0 0.0 0.0 0.0 17,400.02 21.0 1,157.0 3,041.0 0.0 0.0 0.0 0.0 0.0 4,219.03 1.0 18.0 276.0 570.0 1.0 0.0 0.0 0.0 866.04 0.0 1.0 11.0 82.0 139.0 0.0 0.0 0.0 233.05 0.0 0.0 0.0 2.0 29.0 48.0 0.0 0.0 79.06 0.0 0.0 0.0 0.0 2.0 7.0 13.0 0.0 22.07 0.0 0.0 0.0 0.0 0.0 0.0 2.0 4.0 6.0Total 46,179.0 14,856.0 3,328.0 654.0 171.0 55.0 15.0 4.0 65,262.0
A Appendix
17
Tab
le3:
Fer
tility
Logit
(1)
(2)
(3)
Mar
rM
arr
Marr
chil
db
orn
chil
db
orn
chil
db
orn
EQ
UA
TIO
NV
AR
IAB
LE
Sco
efse
coef
seco
efse
chil
db
orn
L.C
ohab
itin
g1.4
07***
(0.0
927)
1.0
76***
(0.1
11)
L.M
arr
ied
1.5
09***
(0.1
11)
0.9
39***
(0.3
28)
L.C
oh
ab
itin
gA
nd
Marr
ied
-0.0
573
(0.3
46)
L.E
nro
lled
-1.2
31***
(0.1
05)
-1.1
14***
(0.1
03)
-1.1
18***
(0.1
05)
L.B
iolo
gic
alC
hild
ren
-0.9
98***
(0.1
29)
-0.8
16***
(0.1
28)
-0.9
51***
(0.1
29)
Age
1.5
56***
(0.2
43)
1.3
56***
(0.2
35)
1.4
39***
(0.2
40)
AgeS
qu
are
d-0
.0306***
(0.0
0534)
-0.0
269***
(0.0
0519)
-0.0
289***
(0.0
0529)
Urb
an
-0.2
67***
(0.1
01)
-0.2
70***
(0.0
962)
-0.2
43**
(0.0
989)
Sou
th-0
.0186
(0.1
38)
-0.0
0455
(0.1
26)
-0.0
175
(0.1
32)
Wes
t-0
.0823
(0.1
59)
-0.1
02
(0.1
46)
-0.0
950
(0.1
52)
Nort
hea
st-0
.140
(0.1
80)
-0.1
60
(0.1
65)
-0.0
928
(0.1
72)
HS
Deg
-0.6
46***
(0.1
43)
-0.5
47***
(0.1
37)
-0.5
92***
(0.1
41)
Colleg
eDeg
-1.4
31***
(0.2
15)
-1.2
75***
(0.2
09)
-1.3
77***
(0.2
14)
Ad
vD
eg-1
.070**
(0.5
07)
-0.8
54*
(0.4
80)
-1.0
25**
(0.4
99)
Mom
Yea
rsS
choolin
g-0
.106***
(0.0
255)
-0.0
883***
(0.0
231)
-0.0
951***
(0.0
242)
Dad
Yea
rsS
choolin
g-0
.104***
(0.0
232)
-0.0
885***
(0.0
209)
-0.0
949***
(0.0
219)
ParV
Rel
igio
us
-0.0
150
(0.1
13)
0.0
852
(0.1
02)
0.0
315
(0.1
07)
Bla
ck1.1
33***
(0.1
55)
1.0
84***
(0.1
41)
1.2
04***
(0.1
49)
His
pan
ic0.5
98***
(0.1
72)
0.5
62***
(0.1
56)
0.6
02***
(0.1
63)
Cit
izen
AtB
irth
0.6
83***
(0.2
03)
0.4
83***
(0.1
83)
0.5
97***
(0.1
93)
Ob
serv
ati
on
s16,4
25
16,4
25
16,4
25
Nu
mb
erof
PU
BID
1,6
94
1,6
94
1,6
94
Log
likel
ihood
-3282
-3261
-3232
Ch
i2638.0
734.5
726.1
Sta
nd
ard
erro
rsin
pare
nth
eses
***
p¡0
.01,
**
p¡0
.05,
*p
¡0.1
18
Tab
le4:
Fer
tili
tyL
ogit
wit
hL
ess
Vari
ab
les/
More
Ob
serv
ati
on
s(1
)(2
)(3
)(
Mar
r2M
arr
2M
arr
2ch
ild
born
chil
db
orn
chil
db
orn
EQ
UA
TIO
NV
AR
IAB
LE
Sco
efse
coef
seco
efse
chil
db
orn
L.C
ohab
itin
g1.
216*
**
(0.0
568)
0.9
71***
(0.0
670)
L.M
arr
ied
1.3
11***
(0.0
698)
0.9
69***
(0.1
81)
L.C
oh
ab
itin
gA
nd
Marr
ied
-0.2
13
(0.1
92)
L.E
nro
lled
-0.9
77***
(0.0
623)
-1.0
70***
(0.0
627)
-0.9
69***
(0.0
630)
L.B
iolo
gic
alC
hild
ren
-0.7
65***
(0.0
730)
-0.8
58***
(0.0
727)
-0.8
51***
(0.0
724)
Age
1.3
40***
(0.1
30)
1.5
10***
(0.1
33)
1.4
05***
(0.1
32)
AgeS
qu
are
d-0
.0272***
(0.0
0287)
-0.0
307***
(0.0
0292)
-0.0
288***
(0.0
0290)
Urb
an
-0.1
43**
(0.0
615)
-0.1
29**
(0.0
632)
-0.1
21*
(0.0
626)
Sou
th0.0
366
(0.0
791)
0.0
0733
(0.0
831)
0.0
117
(0.0
812)
Wes
t-0
.197**
(0.0
959)
-0.1
99**
(0.1
00)
-0.1
97**
(0.0
984)
Nort
hea
st-0
.162
(0.1
00)
-0.1
36
(0.1
05)
-0.1
16
(0.1
03)
HS
Deg
-0.5
29***
(0.0
736)
-0.5
75***
(0.0
755)
-0.5
52***
(0.0
749)
Colleg
eDeg
-1.2
36***
(0.1
22)
-1.3
35***
(0.1
25)
-1.3
10***
(0.1
25)
Ad
vD
eg-1
.038***
(0.3
27)
-1.1
38***
(0.3
37)
-1.1
65***
(0.3
36)
Mom
Yea
rsS
choolin
g-0
.114***
(0.0
132)
-0.1
25***
(0.0
139)
-0.1
18***
(0.0
135)
Bla
ck1.0
13***
(0.0
813)
0.9
95***
(0.0
852)
1.1
16***
(0.0
842)
His
pan
ic0.3
64***
(0.0
960)
0.3
46***
(0.1
01)
0.3
70***
(0.0
988)
Cit
izen
AtB
irth
0.0
482
(0.0
846)
0.0
904
(0.0
894)
0.0
697
(0.0
871)
Ob
serv
ati
on
s36,0
55
36,0
55
36,0
55
Nu
mb
erof
PU
BID
3,8
21
3,8
21
3,8
21
Log
likel
ihood
-8953
-9007
-8896
Ch
i21396
1253
1422
Sta
nd
ard
erro
rsin
pare
nth
eses
***
p¡0
.01,
**
p¡0
.05,
*p
¡0.1
19
Table 5: Base Regression for Enroll(1) (2)
EQUATION VARIABLES coef se
Enrolled L.Married -0.176* (0.0959)L.Cohabiting -0.570*** (0.0690)
L.childborn -0.257*** (0.0904)
L.BiologicalChildren -0.405*** (0.0533)
Age -2.583*** (0.120)
AgeSquared 0.0530*** (0.00270)
Urban -0.00681 (0.0636)
South -0.314*** (0.109)
West -0.372*** (0.125)
Northeast -0.105 (0.135)
HSDeg -1.971*** (0.0849)
CollegeDeg -5.947*** (0.146)
AdvDeg -28.33 (1,797)
MomYearsSchooling 0.289*** (0.0166)
Black -0.510*** (0.121)
Hispanic -0.468*** (0.147)
CitizenAtBirth -0.734*** (0.133)
Observations 36,163
Number of PUBID 3,821
Log likelihood -14121
Chi2 6288
Standard errors in parentheses
*** p¡0.01, ** p¡0.05, * p¡0.1
20
Table 6: Base Regression for MarriedMarried Married
EQUATION VARIABLES coef se coef se
Married L.Cohabiting 3.116*** (0.0851)L.childborn 0.0582 (0.111) 0.574*** (0.106)
L.Enrolled -0.704*** (0.0918) -1.156*** (0.0869)
L.BiologicalChildren 0.176** (0.0711) 0.0719 (0.0713)
Age 2.015*** (0.274) 3.013*** (0.280)
AgeSquared -0.0338*** (0.00573) -0.0520*** (0.00578)
Urban -0.157 (0.0986) -0.171* (0.0943)
South 0.536*** (0.168) 0.525*** (0.175)
West 0.00609 (0.196) 0.0938 (0.205)
Northeast -1.081*** (0.226) -1.109*** (0.232)
HSDeg 0.576*** (0.167) 0.352** (0.170)
CollegeDeg 1.392*** (0.215) 1.332*** (0.215)
AdvDeg 2.242*** (0.413) 2.577*** (0.413)
MomYearsSchooling -0.0996*** (0.0283) -0.163*** (0.0314)
Black -2.454*** (0.212) -3.351*** (0.215)
Hispanic 0.118 (0.221) -0.0350 (0.243)
CitizenAtBirth -0.0524 (0.208) 0.139 (0.224)
Observations 36,030 36,045
Number of PUBID 3,821 3,821
Log likelihood -6161 -6941
Chi2 2830 2564
Standard errors in parentheses
*** p¡0.01, ** p¡0.05, * p¡0.1
21
Table 7: Logit of Having a First Child Conditional on NoneVARIABLES firstchildborn
L.Cohabiting 1.343*** (0.0952)L.Married 1.259*** (0.315)L.CohabitingAndMarried -0.222 (0.333)L.Enrolled -1.181*** (0.0795)Age 1.642*** (0.182)AgeSquared -0.0360*** (0.00419)Urban -0.0380 (0.0824)South 0.0203 (0.102)West -0.187 (0.121)Northeast -0.0576 (0.126)HSDeg -0.486*** (0.102)CollegeDeg -1.367*** (0.166)AdvDeg -1.184*** (0.408)MomYearsSchooling -0.129*** (0.0162)Black 1.248*** (0.106)Hispanic 0.374*** (0.123)CitizenAtBirth 0.0548 (0.108)
Observations 28,028Number of PUBID 3,740Log likelihood -5334Chi2 824.4
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
22
Table 8: Logit of Having a Second Child Conditional on OneVARIABLES secondchildborn
L.Cohabiting 0.409*** (0.101)L.Married 0.178 (0.292)L.CohabitingAndMarried 0.250 (0.309)L.Enrolled -0.295*** (0.111)Age 0.377 (0.300)AgeSquared -0.00782 (0.00648)Urban -0.325*** (0.0941)South -0.0705 (0.105)West -0.146 (0.130)Northeast -0.201 (0.136)HSDeg -0.481*** (0.102)CollegeDeg -0.604*** (0.178)AdvDeg 0.245 (0.634)MomYearsSchooling -0.0307* (0.0165)Black 0.324*** (0.102)Hispanic 0.293** (0.122)CitizenAtBirth 0.0687 (0.110)
Observations 4,886Number of PUBID 1,576Log likelihood -2203Chi2 142.1
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Table 9: Number Who Do Not Respond by Year
non-responsesYear female male Total
1997 0 0 01998 282 316 5981999 346 430 7762000 421 483 9042001 491 611 11022002 486 602 10882003 558 672 12302004 615 867 14822005 713 933 16462006 629 796 14252007 702 864 15662008 662 832 1494Total 5905 7406 13311
23