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