Premarital Cohabitation and Subsequent Marital Dissolution: A Matter of Self-Selection?" Lee A. Lillard; Michael J. Brien; Linda J. Waite Demography, Vol. 32, No. 3, Family and Household Demography (Aug., 1995), 437-457 Stable URL hitp:/flinks.jstororg/sii sici=0070-3370% 28 199508%2932%3A3%3C437%3APCASMD%3E2.0,CO%3B2-X Demography is currently published by Population Association of America. ‘Your use of the ISTOR archive indicates your acceptance of JSTOR’s Terms and Conditions of Use, available at hup:/www,jstororglabout/terms.hml. ISTOR’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 hupufwww.jstor.org/journals/paa htm]. Each copy of any part of @ JSTOR transmission must contain the same copyright notice that appears on the sereen or printed page of such transmission. STOR is an independent not-for-profit organization dedicated to creating and preserving a digital archive of scholarly journals. For more information regarding JSTOR, please contact support @jstor.org, hupslwwwjstor.org/ Fri Sop 17 07:15:18 2004Demography, Vol. 32, No. 3, August 1995 Premarital Cohabitation and Subsequent Marital Dissolution: A Matter of Self-Selection?” Lee A. Lil RAND 1700 Main Street Santa Monica, CA 90407 Michael J. Brien Department of Economies University of Virginia 114 Rouss Hall Charlottesville, VA 22903 Linda J. Waite Population Research Center NORC and the University of Chicago 1155 E. 60th Street Chicago, TL 60637 rd Married couples who began their relationship by cohabiting appear to face an increased risk of marital dissolution, which may be due to self-selection of more issolution-prone individuals into cohabitation before mariage, ‘This. paper uses newly developed econometric methods to explicitly address the endogeneity of cohabitation before martiage in the hazard of marital disruption by allowing the Uunobserved heterogeneity components to he corelated across the decisions to cohabit, and to end a marriage. These methods are applied to data from the National Longitudinal Study of the High School Class of 1972. We find significant heterogeneity in both cohabitation and marriage dstuption, and discover evidence of selfseletion into cohabitation, Cohabitation has become an increasingly common type of union in the United States ‘over the past several decades, with far-reaching consequences for later marriage, for childbearing, and for the stability of unions. In fact, couples appear to be forming unions at about the same rate now as several decades ago; cohabitation substitutes for marriage among. many couples at younger ages (Bumpass, Sweet, and Cherlin 1991). Individuals forming new unions after divorce are especially likely to cohabit—at least initially —rather than marry (Bumpass and Sweet 1989), ‘Most cohabitations are rather short-lived; couples move relatively quickly into either “This research was supported by Grant P-S0-12639 from the Cente fr Population Research, NICHD, We ‘wan to thank Constant W. A. Pais for beipfu isussons. The software for this aplication was developed by Lee Lili and Constantin Panis, We want 10 recognize Richard Baumann for assistance in developing the event history data Copyright © 1995 Populton Assocation of Amica 437438 Demography, Vol. 32, No. 3, August 1995 marriage or the end of the union. Forty percent of all cohabiting couples either marry or stop living together within @ year, and only one-third are still cohabiting after two years (Bumpass and Sweet 1989; Thornton 1988). Although most theories of marital choice predict that cohabitation would increase the stability of later marriages for those couples Who marry, evidence to date suggests the opposite: couples who cohabit before marriage seem to end their marriages at significantly higher rates than couples who never lived together before the wedding (Bennett, Blanc, and Bloom 1988; DeMaris and Rao 1992; ‘Teachman and Polonko 1990), In this paper we attempted 0 reconcile the theoretical with the empirical literature; we develop and test a series of hypotheses about the relationship between premarital cohabitation and the stability of later marriages. We focus primarily on the determinants of the choice to cohabit before a marriage, on the determinants of marital dissolution, and especially on the relationship between cohabitation before a particular marriage and the stability of that marriage. We build on related work by Waite and Lillard (1991) that examines the relationship between children and marital disruption. The analysis uses data on cohabitation, marriage, fertility, and marital dissolution from the National Longitudinal ‘Study of the High School Class of 1972 (NLS-72). ‘We model the decision to begin a first or later union as a matriage or as a cohabitation: we also simultaneously model the relationship between cohabitation before a marriage and later disruption of that marriage. This strategy explicitly recognizes that the decision 10 cohabit before marriage is a potentially endogenous variable in the divorce process, in that some couples may chose to cohabit because they think they would face high chances of disruption if they married; modeling divorce simply as a function of cohabitation can lead 10 biased estimates of the effects of the various determinants, We avoid this problem by estimating equations simultaneously for these related hazard processes. We estimate this mode! with methods, developed in Lillard (1993), that permit us to ‘model the hazard process of one component of family formation when it depends on either the past outcome ot the actual hazard of another process. We extend Lillard (1993) to include the simultaneous estimation of continuous choices—measured as the waiting time to an event—and discrete choices, in this case whether to cohabit or to marry at the Star of a union. The empirical analysis includes individual and time-varying covariates, controls for multiple forms of duration dependence, and—crucial for the simultaneous estimation — controls for unobserved. heterogeneity. We address explicitly the endogeneity of cohabitation by allowing the unobserved heterogeneity components to be correlated across the two decisions, BACKGROUND ‘The Decision to Cohabit At the start of a coresidential union, a couple devides either to live together—perhaps before marrying—or to marry directly. In almost half of recent marriages the couple had lived together first (Bumpass and Sweet 1989). Glick and Spanier (1980) use current living arrangements from the Current Population Survey to describe individuals living with one cother adult of the opposite sex, whom the authors identify as cohabitors. They find that never-married cohabitors tend to be relatively young, to be relatively highly educated, and to live in large metropolitan areas. They also find a cohabitation rate for blacks three times as great as for whites, Bumpass and Sweet (1989) also examine the relative risk of beginning a cohabitation for never-married individuals and report that higher rates of cohabitation are associated withPremarital Cohabitation 489 coming from a single-parent family. Religion also seems to affect the prosperity to cohabit (Landale and Forste 1991; Thorton, Axinn, and Hill 1992) Analyses of the transition from single to married may provide some insights into the choice of cohabitation versus marriage at the beginning of a coresidential union Characteristics of the individual, his or her family of origin, current situation or competing roles, and characteristics of the marriage market all affect the probability that a young man for woman single at the beginning of a period will have married by the end. Those from intact families and those with relatively highly educated parents are less likely than others 10 marry before age 20 or 21. Being a student reduces chances of marriage, presumably by providing competing activites or interests (Hogan 1978). Educational attainment, net of age, increases chances of marriage (Goldscheider and Waite 1986; Waite and Spitze 1981). Cohabitation and Marital Disruption Most theories of marital search and marital stability focus on the role of incomplete information in later disruption (Becker, Landes, and Michael 1977). According t0 these theories, individuals choosing a spouse do so after searching for some period and bargaining \with potential partners over the terms of a match. Those who require spouses with relatively rare characteristics must either search longer (or more intensively) of face a poorer match than those with more ordinary requirements. Those who conduct an abbreviated search tend to make poorer matches because they have less time to collect information about the market and about their spouse before marriage. Also, those whose future life is uncertain may find that they or their spouse expected a different life than that which materializes (Oppenheimer 1988) Tn every case, the key to marital stability is information—about the potential spouse and about the kind of life which marriage to that person would entail. Clearly, couples who live together before they decide to marry have both the time and the opportunity to collect 4 good deal of information about what marriage would be like. Those who don’t like the prospects presumably end the cohabitation; in any event, they would be relatively unlikely to proceed to marriage. Thus. premarital cohabitation may serve a screening function (Teachman, Thomas, and Paasch 1991). This selection process should mean that only the best-matched couples will turn their cohabitations into marriages and that these couples should have even more stable marriages than couples who never cohabited, because couples ‘who did not live together will know less about the quality of the match and the character of the martiage. In fact, however, evidence to this point suggests that couples who cohabit before marriage have substantially less stable unions than do couples who married without living together first (Axinn and Thornton 1992b; Bennett et al. 1988; DeMaris and Rao 1992; ‘Teachman and Polonko 1990; Thomson and Colella 1992), Bennett et al. (1988) argue that the relationship between cohabitation and later marital disruption results from selectivity: couples who cohabit do so at least in part because they have less commitment to the institution of marriage than do couples who marry directly. Because of this lack of commitment to marriage as an institution, couples who cohabited may be more willing to tend an unsatisfactory relationship—and perhaps less willing to work to make a relationship satisfactory than other married couples. Thomson and Colella (1992) find that couples who cohabited before marriage report lower-quality relationships and less commitment to the institution of marriage than couples who did not cohabit. Another type of selectivity may explain the higher observed marital instability of cohabitors: Booth and Johnson (1988) find some evidence that cohabitators bring characteristics (0 the relationship which make them “poor marriage material.” These40 Demography, Vol. 32, No. 3, August 1995 include personality problems, alcohol or drug abuse, financial irresponsibility, problems with the law. and unstable unemployment patterns. AS a result of these characteristics, cohabitors often may bring t0 a marriage a deviant lifestyle or a disregard for conventions, and these make them more likely to divorce. Cohabitation also appears to select individuals who are relatively approving of divorce as a solution to matital problems (Axinn and Thomton 1992b). Such attitudes then may result in higher levels of divorce and separation for these couples. Teachman and Polonko (1990) also argue that cohabiting couples often marry because of pressure from family and peers, which results in less stable marriages than if they marry from a desire to do so, Also, at least one study suggests that the experience of cohabitation causes young adults to become more accepting of divorce (Axinn and Thornton 1992b), possibly because couples with “successful” cohabitations— those ending in marriage —have evidence that reasonable alternatives to marriage exist (Thomson and Colella 1992). In support of this reasoning, Waite, Goldscheider, and Witsberger (1986) find that nonfamily living—including periods of cohabitation— increases individualism among young adults, ‘The impact of cohabitation on the stability of a marriage may depend on how long it takes couples to move from living together to marriage. Previous research reports that disruption of a marriage preceded by cohabitation becomes more likely as the length of the cohabitation increases (Bennett et al. 1988; Teachman and Polonko 1990). Thomson and Coletta (1992) find that couples who cohabited for relatively long periods before marriage differ more strongly from those who married directly in reported quality of the union, ‘commitment to the institution of marriage, and likelihood of divorce than do couples who cohabited briefly. Bennett et al. (1988) argue that couples who cohabit for long periods may develop individualistic modes of behavior which are incompatible with roles in marriage. HYPOTHESES Sel ity We test Bennett et al.’s (1988) assertion that couples who cohabited before their marriage are selected for their relative willingness to divorce. This willingness could result from a lack of commitment to the institution of marriage generally, or from a lack of ‘commitment to this particular spouse, In addition, cohabitors may tend to have characteristics that make them “poor marriage material” (Booth and Johnson 1988), Information Insofar as couples who cohabit do s0 10 acquire information about the quality of the ‘match, those who proceed to marriage with their cohabiting partner should be those whom the cohabitation reveals to be good matches. We hypothesize that, net of selectivity, those ‘who cohabit with their future spouse will be less Tikely 0 disrupt their marriage than those ‘who marry without cohabiting with the person they eventually marry, Duration of Cohabitation Given variation in the commitment 10 the institution of marriage, it may be the case that those couples who marry without cohabiting have—on average—the greatest ‘commitment, couples who marry only after a long period of cohabitation have much lower levels, and couples who cohabit for shorter periods are intermediate in their commitment.Premarital Cohabitation an ‘Then we would expect length of cohabitation to be related positively to chances of marital dissolution, as others have found. Yet when we take into account the adverse selection into cohabitation of those least committed to marriage and most likely to divorce, we may see no additional effect of length of cohabitation. If long periods of cohabitation, however, cause couples to develop behaviors that make success in martiage more difficult, as Bennett et al, (1988) suggest, then we should observe an effect of length of cohabitation even after we take selectivity into account, EMPIRICAL MODEL. In this seetion we develop an empirical mode! of the risk of marital dissolution which explicitly incorporates 1) a direct effect of having begun the relationship as a nonmarital cohabitation and 2) potential self-selection of more divorce-prone individuals into premarital ‘cohabitation. The unit of observation is a woman, including all of her marriages reported in the panel period covered by the NLS-72. Because we focus on marital disruption, we include only women who had ever married by the date of the last interview, when they were about age 32, and only the marriages that began before that date For each marriage m there is an equation—a hazard equation—relating the instantaneous probability of marital dissolution of time ¢, conditional on the marriage not having dissolved before that time, to explanatory variables including whether the couple cohabited before the marriage. This equation is denoted as follows: In Kft) = By + B,Age,,(t) +8sDurMar,,(t) + B,BInt,(0) + ByX%,(t) + ByMar2,, "+ 8,Col wo where the covariates are listed in Table 1 and are discussed further below. The direct effect of having cohabited with the partner before marriage m on the hazard of dissolution of the marriage is measured by 8, the coefficient associated with Col, In this paper we attempt to determine whether the apparent effect of having cohabited with a marriage partner before marriage on the hazard of dissolution of that mariage is real or is due to self-selection into cohabitation by women who have a relatively high risk of marital dissolution in all their marriages for other (unobserved) reasons. This is equivalent to asking whether the indicator variable for cohabitation, Coh,, is endogenous in the ‘equation for the hazard of dissolution, The following specification is designed to incorporate this possibility, to test its exogeneity, and to obtain a consistent estimate of the true effect. To do this we jointly model marriage dissolution and the cohabitation decision, For each marriage, there is also a second equation—a probit index function—determining the probability of cohabitation before that marriage. This function is Ig =o + OX + E+ e where the cohabitation decision is determined by Lith, >0 Coby =f jit 2 5 @ ‘The covariates included in Eq, (2) are also listed in Table 1; & is a woman-specifie component, while v, is assumed to be marriage-specitic. ‘The issue of ‘endogeneity or self-selection is determined by whether Coh, is independent of the woman-specific unmeasured stochastic portion of the hazard’ of dissolution, represented by € in Eq. (1). It is not independent; thus Coh,, is endogenous or self-selected, if there are unobserved factors which influence both the Woman's Fisk ofa2 Demography, Vol. 32, No. 3, August 1995 ‘marital dissolution, €, and her propensity to cohabit before marriage, £, so that they are correlated. If Cok, is endogenous, then estimation of the parameters of the dissolution process must account for the endogeneity; otherwise the estimates will not be consistent ‘The observation of the duration of multiple marriages for some women identifies the distribution of unobservable and facilitates the test for exogeneity. We consider both processes jointly, using full-information maximum likelihood. This framework allows us to test both for the direct effects of cohabitation and for the effects of self-selection on the risk of marriage dissolution, In the next sections we develop this framework in ‘greater detail ‘The Marriage Dissolution Process To model the marriage dissolution process empirically we use the continuous-time “failure-time” model of marriage duration shown above in Eq. (1). “Duration dependence” has two primary sources—DurMar,(1) and Agey(t). DurMar,(t) represents the effect of marriage duration, and may reflect the accumulation of information about the quality of the match as well as the accumulation of marriage-specifie capital.' Age, reflects the effects of maturation on the hazard of marital dissolution.? The individual's age also affects the initial level of the hazard at the time when the marriage begins (i.e., vvia age at marriage), but continues as well to have an effect during marriage. Within the marriage, Age,(£) and DurMar,(t) are both collinear with time, but separate duration effects are identified because individuals marry at different ages (i-e., there is variation in the starting points of the splines).® A third source of duration dependence is the duration since the birth of the last child within the marriage, Bint,(t).* These sources of duration dependence combine to determine the “baseline” hazard’ and survivor funetions, whieh respectively are Ni t)=exp{By + B,Age,(t) + B:DurMar,(t) + 8\Bint,(} 4) and skit) = exe ff expt, + 8,Agen(0)+8:DurMary ) + 8:Bhnr(v)as} © for mariage. All other covariates and the residual ¢, which reflects individual heterogeneity, shift the hazard proportionately, These. inchde vector of exogenous time-varying. (or constant) covariates, X;,(t), and the potentially endogenous variables: 1) an indicator for second or higherorder marriages, Mar2yy and.) an indiator for having. begun the marrage as a cohabitation, Cob. The exogenous covariates, which ae listed in Table I, include demographic and background characteristics, spouse charctristes, and variables that capture the legal environment of divorce. The. prior marriage indicator represents “prior occurrence" dependence: is effet Is esinated consistently ony ifthe models" account for heterogeneity. Woman-specific unobserved heterogeneity in the hazard of marital dissolution, eaptured by «, is constant cross a woman's, marriages and is assumed to be istbuted normally i., e~ M(Q:))* This heterogeneity tem induces eorelaion in durations across marriages of a woman in the aggregate because It is assumed to be constant for a given woman but ochasic across women. even thovgh a given woman's marriage Joations ae independent ite is known (es conditional on ° ‘On the basis of Egs. (4) and (S |. the “conditional” (on the heterogeneity term e)Premarital Cohabitation 43 hazard and survivor functions respectively are RL()= hy (OexplB.XU(t) + ByMar2,,+8,Coh, + €} © and ~exp{8,X4(08 Mar?,, +8,Coh,, +e) | e SidtoXnl 1.6) = Tye where / isthe numberof subintervals within which covariates X¥(r are constant, and t {We use these conditional hazard and survivor functions % obtain the likelihood of ‘observed data on mariage durations. The aggregate survivor function fora marviage may be derived by integrating the conditional servivor function over the range of the unobserved heterogeneity distribution; the aggregate hazard function is the derivative of the aggregate Survivor function with respect to duration ‘The Premarital Cohabitation Process ‘The decision to cohabit before marriage is specified by probit Eq. (2) and (3) for each marriage, with marriage-varying covariates and woman-specific heterogeneity, &, in the propensity to cohabit. All covariates X;, are measured at the beginning of the relationship, ‘whether cohabitation or marriage. AS noted in Table | and discussed further below, the set of covariates included in XS, does not coincide perfectly with those included in X', ‘The residual term, £, represents woman-specific unobserved heterogeneity in the propensity to cohabit with partners before marriage, and thus is constant across marriages. It may be correlated, however, with the heterogeneity term in the equation or the hazard of ‘marriage dissolution. ‘The heterogeneity components are assumed to be jointly normally distributed, and are represented by Grice | * ‘The critical test of self-selection or endogeneity is whether oye — 0. ‘The residual term, v,, is specific 10 marriage m and is independent from marriage to marriage (and independent of all other stochastic terms). Each of these residual terms is assumed to be distributed normally. That is, v,, ~ M(0,1) for each m. ‘We observe multiple marriages for each woman, each preceded by a cohabitation decision, The “conditional” (on the heterogeneity term €) probabilities of cohabitation are independent across the M marriages observed for a particular woman, so that their joint probability is the produet of the probabilities. We denote the conditional probability for ‘marriage m by ((2Coh,,— lay +X, +8), co) ‘where ® denotes the cumulative normal ‘bution, Estimation: The Joint Marginal Likelihood Function ‘The parameters of the procohabitation process are identified and estimable from the information on cohabitation alone. Conditional on the woman-specific heterogeneityor Demography, Vol. 32, No. 3, August 1995 ‘Table |. Equation Specifications and Variable Defi Process Cohab- Marriage Variable Label tation Dissolution ‘Time/Durition gett) Individuals age-spine x x DurMar(t) Marriage duration-spline x Bins) ‘Time since bint of last chi-spline x Cohabitation with Marriage Partner Con Prior cohabitation with spouse x Cohabitation Duration Duration of prior cohabitation x Children with Marviage Partner Pregnant Currently pregnant x x Current Cohababitation No. Children bor cohabiting x ‘Current Marriage No. Children born in marriage x ‘Children with Past Partners No. Children Total no, children prior relationships x No, Relationship No. children born in no telationship x Past Marriage No. childen bom in past marriages x Past Cohabitation NNo. children born in past cohabitation x Marriage to Past Partners Mar Second or third marriage x x Education ‘Some college Altended school beyond h.s., 90 BA x x BAS College graduate oF more x x Not enrolled [Not enrolled in school at time x Background ‘Black Race is black x x Catholic Religion is Catholic (vs. Protestant) x x Other religion Not Catholic or Protestant (vs, Protestant) X x Rural Small townrural area in 1972 (vs. large city) X x Medium city “Medium city in 1972 (vs large city) x x City missing City size unknown in 1972 x x Intact family Lived with both parents in 1972 x x In Cohabitation Equation Only Moth. ed. >12 “Mother has education greater than has x Fath, ed, > 12 Father has education greater than b.s x Rent Rental prices (logs) in sate x Home price Median home value in state x Sex ratio Marriageable men/woman (log) x In Dissolution Equation Only Handicap Evaluated to be handicapped x ‘Tough divorce law Tough divorce laws at time tin state x Easy divorce law Easy divorce laws at time tin state x Law cost Cost of legal services-lawyers x ‘Age difference Husband age minus wife age x NOTE; Omitted categories are indicated in parentheses for dummy variables with two oF more groups.Premarital Cohabitation 4s ‘component (and covariates), the cohabitation probabilities are independent, but they are correlated in the aggregate precisely because of the heterogeneity component. ‘The conditional likelihood of the M cohabitation decisions is the product of the individual probabilities; the marginal-likelihood function, integrating over the distribution of possible parameter values of heterogeneity, is given by 1 = fo ()TT occcon, ~ 1109 +0ix209+ 86, o yo NO which is an M-dimensional “one-factor” probit model, Because the heterogeneity ‘component is unknown for any woman, the marginal joint density includes the implicit ‘correlation in outcomes induced by the unknown parameter. ‘Similarly, conditional on the vector of unobserved heterogeneity components (€,E) and all covariates (including Cok,), no remaining correlation exists either between the hazard of dissolution and the propensity to cohabit or across a woman's marriages in either process. ‘The conditional probabilities of outcomes are thus independent, and the joint conditional likelihood is the product of the conditional probabilities of the outcomes. ‘Our estimates are based on the marginal (unconditional) joint probabilities of observed ‘outcomes, both whether to cohabit and the marriage duration (completed or censored), for all of the M marriages of each woman in the sample. Again, the unconditional joint probabilities are not independent, and the correlations are induced by the unknown heterogeneity components. The joint marginal (unconditional) likelihood for all of a ‘woman's observed outcomes is then the integral of the joint conditional likelihood times the marginal density of the two heterogeneity components. That i, es 2 [ov as | a. 0 wf foe Th ox X40 UHL 0" “s B(2Cofiy=1) (ag +a; Xi()+ €))MEle, ay Where fis the completed duration of mariage m if D,, ~ 1 and isa censored duration if Dy = 0. Estimation is based on maximization of the marginal-likelihood function.” This joint ‘marginal-likelihood function incorporates correlation between marriage and cohabitation as ‘well as the correlation across marriages (replications) for each process (dissolution or cohabitation). Identification A key feature of our model is its ability o distinguish 1) the effects of self-selection imo cohabitation before marriage from 2) the direct effects of having cohabited with a particular partner before marriage. We observe women with multiple marriages, and see ‘whether each marriage was preceded by a period of nonmatital cohabitation lasting a month ‘of more. We have @ two-equation model for each of a woman's martiages—one for the probability of premarital cohabitation and one for the hazard of dissolution of the resulting marriage. The individual effect in the cohabitation equation represents the woman’ average propensity to cohabit before marriage. The individual effect in the equation for the hazard of46 Demography, Vol. 32, No. 3, August 1995 ‘marriage dissolution represents the woman’s average “risk” of dissolution. The correlation between these two individual effects represents a woman's self-selection into cohabitation: 4 positive correlation indicates “adverse” selection of women with a high average risk of marital dissolution into cohabitation on average, and a negative correlation indicates “positive” selection of women with a low risk of marital dissolution into cohabitation, A test of zero correlation between the heterogeneity components is atest for the exogeneity of ccohabitation—that is, for self-selection into cohabitation. When the correlation is 0, the likelihood funetion in Eg. (11) factors into two parts, one for marriage dissolution and the ‘other for cohabitation, so that treating Coh,, as exogenous is appropriate. When the correlation differs significantly from 0, however, estimates based only on the marriage durations will be biased. ‘Afier conditioning on the two individual effects (¢,£), and after accounting for their correlation, the remaining variation in Col, from marriage to marriage “within person” represents the true direct effect of having cohabited with the current partner on the hazard of dissolution of the marriage to that partner. This true effect is purged of the effects of self-selection. It is the “structural” effect of cohabitation after accounting for its endogeneity. This approach represents a parametric fixed-effect estimator.* Because identification is attained through within-person replication, no exclusion restrictions (instruments) are required for identification; all covariates may enter both equations. One ‘may apply exclusion restrictions if they are desited, however. We present results with and ‘without the restrictions implied in Table 1 10 explore the robustness of our results to these overidentifying restrictions DESCRIPTION OF DATA Our analysis uses data on cohabitation, marriage, and marital disruption from the National Longitudinal Study of the High School Class of 1972 (NLS-72). The NLS-72 began in spring 1972 with a sample of 19,001 high school seniors. Follow-up surveys were administered in autumn 1973, 1974, 1976, and 1979 to more than 23.000 members of the ‘ame graduating class, including most of the seniors who were in the base-year sample. In spring 1986 a subsample of 12,841 of the original seniors were interviewed again. This follow-up was unigue in that it was an unequal-probability subsample which included with certainty all individuals in the original sample who had a marital disruption or who were single parents.” The female respondents of the 1986 follow-up constitute the basic sample for the analysis in this paper ‘The NLS-72 provides a wide array of measures of the student's family background and. demographic characterises. The survey also obtained information about activities at the time of the follow-up surveys and between the surveys. The large size of the sample, representing all young adults enrolled in their senior year of high school, combined with the richness of the longitudinal information, makes the NLS-72 well suited for the questions we pose here Because of the detailed information on cohabitation and on later marriage to that partner, the NLS-72 data set is especially appropriate for a study of the consequences of ‘cohabitation. Participants in the 1986 follow-up of the NLS-72 were asked for a retrospective history of cohabitation, marriage, marital distuption, and childbearing. This history includes the starting and ending dates of up to three marriages or marriage-like relationships lasting one month or more. together with some information about each partner. For each relationship, respondents were asked the date of the start of the relationship: whether they ever married the person, and the date of the marriage: whether any childrenPremarital Cohabitation aT were produced from this relationship: whether they were still living with the person at the 1986 interview; and how the relationship ended, if it did so. These detailed retrospective histories allow us to study the behavior of respondents as they move into and out of a series ‘of unions, We also know which marriages were preceded by cohabitation, and the length of both the marriage and the cohabitation. Further, the retrospective family histories can easily be matched to contemporaneous data on educational achievement and enrollment status, state of residence, and various other detailed background information ‘The NLS-72 has one shortcoming for our purposes. The sample includes only those ‘who reached their senior year of high schoo! and excludes those who dropped out of school before that point. Although a small proportion of this cohort failed to complete high school and although racial differences in completion rates are not large," cohabitation seems to be more common among dropouts than among those with more education. In addition, findings from these data may not apply to more recent cohorts, in which rates of cohabitation are higher than they were for this cohort. Marital Disruption In this analysis we combine divorces with separations under the general rubric of, marital disruption or dissolution, We include separations because separation often precedes Tegal divorce, sometimes by several years, and because separation without subsequent divorce is a frequent disruption pattern among blacks (Glick and Norton 1977), Table 2 shows the percentage of marriages disrupted by 1986, for first and later ‘marriages, and for those preceded by cohabitation and those not preceded by cohabitation ‘This tabulation shows slightly lower percentages disrupted for marriages preceded by cohabitation than for others, both for first marriages and for second and higher-order marriages. This result may be due to the longer potential durations of marriage for couples ‘who did not cohabit, Other studies find quite small gross effects, either positive (Bennett et al. 1988) of negative (Teachman and Polonko 1990), of cohabitation on marital stability during the early years of marriage." Explanatory Variables Table 1 presents names and definitions of variables used in the equations for both marital dissolution and cohabitation. The equation for marital dissolution contains some variables which clearly affect the risk that the marriage will break up, but do not affect the ‘Table 2. Marital Disruption and Duration by Cohabitation Status, Censored versus, Disrupted Marriages Duration (Years) eee % Disrupted Disrupted Censored __Marriages: First Marriages No pre-cohabitaton 317% 44 og 3.812 Pre-cohabitation 28.5% 37 67 1083 2+ Marriages: No pre-cohabitation 24.9% 28 49 433, Pre-cohabi 2.4% 31 4. 514.448 Demography, Vol. 32, 3, August 1995 probability that it is preceded by cohabitation. ‘These variables include marriage duration and the presence of children bom to the marriage, in addition to the measures of cohabitation. AS stated above in our discussion of identification, the model is completely specified without recourse 0 exclusion restrictions. Even so, we exclude several variables from each equation as additional sources of identification in keeping with our preferred specification: below we discuss the rationale for these exclusions. As we point out in the next section. the main results are robust across model specifications with excluded variables, and models without excluded variables. The model of marital disruption includes a number of other variables that previous research has found 0 influence the chances of dissolution, These are race (Michael and ‘Tuma 1985). religion (McCarthy 1979; Teachman and Polonko 1985), current age (Lillard ‘and Waite 1993), education (Glick and Norton 1977), and the duration of the marriage (Waite and Lillard 1991). South and Spitze (1986) found that the effects of these and similar variables on divorce and separation do not change over the course of the marriage. Earlier we discussed the characteristics that previous research found to increase the chances of cohabitation. These include age and education (Willis and Michael 1994), race and city size (Glick and Spanier 1980), religion (Landale and Forste 1991); Thornton etal 1992), coming from a single-parent family (Bumpass and Sweet 1989), and school enrollment (Landale and Forste 1991; Rindfuss and VandenHeuvel 1990). We also include presence of children from previous relationships and whether the woman has been married before (Bumpass and Sweet 1989), ‘The equation for cohabitation before marriage includes a single measure of the total rhumber of children born with previous partners, whereas the equation for marital dissolution includes the number of children bor in previous relationships separately for those conceived while the woman is 1) neither married nor cohabiting, 2) married, and 3) cohabiting. As we sce later, this distinction is unimportant We begin with variables that are allowed to affect the decision to cohabit at the start of 44 relationship but to affect marital stability only through their effect on cohabitation Education of the woman's mother and father affects the timing of the transition from single to married (Goldscheider and Waite 1986) and the chances of cohabitation before marriage (Bumpass and Sweet 1989; Thornton et al. 1992). Also, although the structure of the parental family —whether two-parent or other type—has substantial effects on the disruption of @ young person's own marriage, no research has found similar effects for parents" education. We see later that father’s education has a significant effect on the chances of ‘cohabitation: this finding suggests that itis an effective instrument for marital disruption We also include two measures of the benefits of shared housing and the costs of setting up a separate residence; we argue that the higher the costs of rental and owner-occupied housing. the more financially attractive the prospect of sharing a residence and the higher the chances that a dating couple will decide to cohabit, We index these costs with measures of rental prices and median home value in the state of residence. As we note later, both of these measures have sizable effects on the probability of cohabitation before marriage; again, the findings suggest the effectiveness of these measures as instruments, Finally, we include a measure of the sex ratio in the state of residence, which indexes the status of women and the competition that young men and young women face for partners (South and Trent 1988). Research on the effects of the relative availability of women shows aa decline in both the rates of marriage and the bargains women get in the marriage market, the greater the number of females per male (Grossbard-Shechtman 1993; South and Trent 1988). Thus, lower sex ratios should lead to more cohabitation before marriage. Recent evidence. however, suggests that they also may affect the chances that existing marriages will end (South and Lloyd 1995), ‘The equation for marital dissolution contains some variables which clearly affect thePremarital Cohabitation 49 risk that the marriage will break up. but do not affect the probability that itis preceded by cohabitation. These include marriage duration and the presence of children born to the marriage, as well as the measures of cohabitation. In addition, we include in the equation for marital disruption two indicators ofthe laws governing divorce in the state and the extent to Which these laws allow no-fault versus fault-based divorce (Jacob 1988), The disruption ‘equation also includes a measure of the legal costs of divorce, as reflected in the mean ‘earnings of lawyers in the state during the current period. Residence in a state in which divorce is difficult and/or costly decreases the chances of divorce for individual couples (Peters 1986), Measures of both of these variables are specific to the state of residence at the start of the year in question, and so capture changes over time both in state of residence and in divorce laws and lawyers’ earnings in each state.!2 RESULTS Cohabitation and Marital Disruption We begin with tests of our hypotheses about the effect of cohabitation before marriage fon marital stability. We estimate both a probit function indicating that a nonmarital cohabitation was chosen at the start of the relationship and a hazard model of marital separation, Tables 3, 4, and 5 present the estimates for these models; probit estimates for the cohabitation decision ‘are displayed in Table 3, estimates for the hazard of marital dissolution in Table 4, and estimates of the error structures for both models in Table 5. The first column of each table contains estimates based on a model that assures no unobserved heterogeneity. This is the model used in previous research; we present it here to allow ‘comparison with the results from that research. The second column, which has the same basie specification as the first model, controls for individual-specific heterogeneity in each equation but assumes independence. Finally. the models in the third and fourth columns control for the endogeneity of premarital cohabitation in the matital separation equation by allowing for a correlation in the heterogeneity components across the two processes. ‘The estimated correlation between these heterogeneity components appears in the third column of Table 5 Not all NLS-72 respondents have formed any union by age 32, their average age at the 1986 reinterview. Some have cohabited but have not married. Because this paper examines the impact of cohabitation on the stability of subsequent marriage, we include only those individuals who had ever married by about age 32. Thus, we exclude 16.8% of the white females and 38.1% of the black females, who had not married by the final interview. In addition, time spent in a cohabitation reduces the potential duration of marriage by the last, “observation, Thus, all else being equal, those persons who cohabited before marriage will have shorter durations of marriage than those who married directly. We must keep these caveats in mind when interpreting the results, Decision to Cohabit ‘Table 3 presents results for the model of the choice of cohabitation before marriage. We focus here only on Column 3, which presents results for the equation for cohabitation estimated jointly with the equation for marital disruption, First we see an increase with age, through age 24, in the chances that a coresidential tunion begins as a cohabitation, After that age, chances of choosing cohabitation over ‘marriage decline with increasing age. We recall that only those who form a coresidential tunion chose whether to live together frst or marry directly. Those who stay single simply450 Demography, Vol. 32, No. 3, August 1995 Table 3. Probit Index Function for Premarital Cohabitation o @ @ @ Iniercept Tons7e* —1sao1e> 1Sansee —DogTE= Individuals Age 0.0761" 0.1006" O.oss* ——o.1506% 0.0909" O.1181%** 120% —_.T093¢e =0.0382" 0.05008 0.0516" —0.0867=* Mariage 2+ 0.6737" O.8731%* 0.6988" 0.60169" Children with Past Partners Currently pregnant 05124 0.66608 — — 0.68946 —0,72508* Number of children 0.0097 0.0024 0.0036 —0.0075 Education Not enrolled 0.0879 0.1119 =0.1089 =0.0040 Some college 0.0081 0.0123 015s 0.0888 BA+ 0.0634 0.0851 10.0890 0.1587" Background Black 0.0503 0.0836 0.0632 0.0787 Cathotic 0.0952" 0.13208 0.1333 — 0.0714 Other religion 0.0822 0.0685 0.0718 0.0870 Mother's education > 12 yrs 0.0101 0.0068 0.0070 Father'seducation > 12 yrs 0.0880" ona 0.1051 Tact family ~025dge 0.3430" — —0,3498"* — —0,3550"= Rural 0.0043" — 0.1366" 0.1469" 0.226246 Medium city 0.1084" oid O1187 0.1142 City size missing 0.0936 = 0.1329 0.1386 0.1932 Rent o.szz* dot Tago Home price O27 g.agase= 9.439084 Sex ratio 0.23 2802 0.307 +p S10: p= 05: * p< 1; worked defer the decision. Very few of the NLS-72 respondents form any type of union before age 18. Some marry directly after high school. Then, apparently, cohabitation becomes more likely than marriage as they grow older, at least until the mid-twenties, We can speculate that school. job training, and the process of starting a career make young adults reluctant to marry; at the same time, age brings them increasing independence from their family of ‘origin and gives them more freedom to live with someone if they chose. Parents of young adults may prefer that they marry rather than cohabit, and parents of teenagers may have more means available to encourage compliance with their wishes (Axinn and Thornton 1994). After the median age of marriage—about the mid-twenties—young adults more often have completed the early transitions that are part of becoming an independent adult We see no effect of school enrollment, however, in the choice of cohabitation over ‘marriage. Students may defer commitment to any union because of the competing demands of their studies, but appear no less likely than nonstudents to marry if they form a coresidential union, Early research on cohabitation reported relatively high rates for blacks: we find no racial gap in the choice of cohabitation among those forming unions. Blacks, however, ane less likely than whites to have formed any union by the final survey (Willis and Michael 1994). As in previous research, we see an effect of religion: Catholics who form unions are significantly less likely than others to cohabit and more likely to marry. Landale and Forste (1991) found, in a sample of mainland Puerto Rican women, that Catholics were morePremarital Cohabitation 4st Table 4. Hazard of Marital Dissolution o. @ @ @ Invercept TTaorTt* 12.771 — 12,7997 — 13,1658 Individuals Agel2-18 years 15220" 15780" 1s773* 6613+" 19-24 years 0.13028 -0.13356"* = 0L1241Y8" 0.120606 25-28 years 0.0915 =0.1016""* — =0,09314%" 0.0919 29+ years 0.0480 =0.0839" = 0.0578" 0.0532 Marriage Duration 0-2 years 04343" O.si9see* ——o.sisore® 0.513680 35 years 0.0507 O31 0.12808 OLLIBR¥* 6+ years OTT o.1a4e* 1227 TTA CCohab with Marriage Partner Prior cohabitation o.31s9%* 0.37528" 0.0065, ~0.1872 Cohabitation duration 0.0008 =0.0106 0.0168 0.0227, Children with Mariage Pastner Currently pregnant 1.6002" = 1.6288" = 1.630240" — 1.627206" Current cohabitation 0243 0.2046 —0.2097 =0'3042 Current marviage —O.so0g*** — —0,5144e = 0.5203¢¢" — 015180" ‘Time since Birth of Last Child Inercept —0,6906"* — —0.59Rsre" 0.593488 0. s¥0Rer 0-1 years OS310* 0.7668" OLTS9a* ——_—O.THS3¢4 2+ years 0.0108 0.0139 0.0129 0.0129 Mariage to Past Partners Marriage 2+ og9ser"* 0.1189 0.0324 0.0325 (Children with Past Partners 'No relationship 0.0610 ont 0.0819 0.0887 Past cohabitation o.a7i8 0.2009 0.1045 0.0874 Past marriage 0.0902 0.1026 0.1033 0.1175 Education Some college 0.0643 0.0613 0.0659 0.0707 BA + 0.088 = 0.1872 0.1276 -0.1121 Background Black 0.1769" 0.2069 0.2182" 0.1999 Catholic 0.0619 0.1008 0.1138 =0.1296 Other religion 0.0496 0.0887 0.0777 0.0751 ‘Tough divorce laws 0.1294 0.1323 0.1322 Easy divorce laws O27 O1s226 —O.1584°6 Law costs ($100,000) 0.0035 0.0080 =0.0013 Intact Family 025368" 0.33346 0.3689" 0.38046" Handicapped 0.2869 0.3970 0.3937 0.3661 Rural 0.1573" ——0,19928 0.22018" 0.23576 Medium city 0.0611 10.0909 0.0908 0.0861 City size missing 0.0702 “01197 0.1432, 0.1476 Age difference 0.0095 0.0117 0.0122 0.0122 Fp <1; p= 05: 5% p< Ol: worailed likely to cohabit than those of other religions. As others have found, however, we see that persons from an intact family of origin are significantly less likely to cohabit than those raised in other types of families (Thornton et al. 1992) Several characteristics of the area in which people lived before forming a union affect the type—legal or informal—that they chose. Those from rural areas are more likely to choose marriage over cohabitation than people from urban areas. In addition, chances of482 Demography, Vol. 32, No. 3, August 1995 Table . Error Structure for Marital Disruption and Cohabitation Models oO @ 8) o Standard Deviation Marital Disruption 0.0000 1,093" 1.04579" 9.93900» Cohabitation 0.0000 0.8401" 0.8716" 1.02738 Correlations Marital Disrupt-Cohabitation 0.0000 0.35518" 0.4806" Log: Likelihood TBI S53 — 123314 © p< 08 p= OF ed. cohabiting rather than marrying rise with rental prices in the state of residence. We speculate that individuals may double up to save money, and thatthe housing requirements, for this type of arrangement are more modest than those for a married couple. Some young, couples may cohabit until they can afford a place suitable for young married people: this stage lasts longer, the higher the rents, The same pattern appears for home prices: the chances that a young couple will cohabit increase with the cost of houses in the state. Again, ‘we speculate that some couples may cohabit until they can afford to buy a house, at which point they marry. ‘We expected that the sex ratio in the area of residence would affect choice of union type, with more males per female increasing the choice of marriage over cohabitation (South and Trent 1988). Table 3, however, shows no effects of sex ratio on choice of cohabitation versus marriage. Effect of Premarital Cohabitation on Marital Disruption Among many other variables, an important feature of the specification for the hazard of ‘marital disruption is an indicator variable for whether the marriage was preceded by a period of nonmarital cohabitation. Column | of Table 4 is similar to previous work in this area in that it models a single process with no heterogeneity. It shows a positive and significant effect of cohabitation before marriage on the hazard of marital separation. This measure, as stated earlier, reflects only cohabitation with the partner eventually married. This result replicates findings from earlier studies (Bennett et al, 1988; Teachman and Polonko 1990) and suggests that premarital cohabitation destabilizes the marriage. It is consistent with theories suggesting that couples who choose to cohabit are relatively uncommitted to the Institution of marriage or have other traits that make them poor martiage material. The duration of the cohabitation does not appear to significantly affect the likelihood that the marriage will dissolve To examine whether unobserved differences across individuals are important, we also incorporate an individual-specific heterogeneity term into the models in Column 2. In the ‘equation for cohabitation, this term reflects the woman's propensity to cohabit, net of observable characteristics. In the equation for marital dissolution, this term reflects the ‘woman's propensity, for reasons that we do not observe. to disrupt any marriage that she centers. These results are presented in Column 2 of Tables 3, 4, and 5. Focusing again on the effect of premarital cohabitation in the equation for marital dissolution (Table 4), we see that controlling for individual-specific heterogeneity has no effect on the coefficient for prior cohabitation: it remains statistically significant and negative. Controlling for individual-specific heterogeneity generally affects covariates correlated with durationPremarital Cohabitation 453 dependence, such as age and marriage duration. Respondents with a relatively high propensity to separate from their marriage partners would tend to drop out sooner; without the controls for unobserved heterogeneity, this tendency could lead to biased estimates of the duration dependence. We see that this bias is unimportant here, We argued earlier that the decision to cohabit atthe start of a coresidential union may ‘depend on the partners commitment o marriage and on the potential stability of a marriage Couples may cohabit because a marriage would have a high risk of failure: thus the decision to cohabit is endogenous in the disruption equation. To control for this endogeneity we allow the heterogeneity components to be correlated across the 1wo processes —marital disruption and cohabitation, The estimates for this model appear in Column 3 of Tables 3, 4, and 5. Insofar as cohabitation before a marriage is endogeneous in the process that causes disruption, cohabitation is correlated with the error term for the disruption equation. AS ‘mentioned above, this introduces a bias in the coefficient for cohabitation, When we control or this endogeneity we correct the bias; this correction may change the size and significance of the coefficient, The sign of the correlation coefficient indicates the direction of the bias Comparison of the results in Columns 2 and 3 shows that in this case the bias was substantial. When we include the correlation between the heterogeneity components across the equations for marital dissolution and premarital cohabitation, the effect of cohabitation ig reduced from sizable and significant to negligible and insignificant. Thus models of cohabitation and marital disruption that take into account the potential selectivity into ‘cohabitation of those most likely to disrupt a marriage show no effect of cohabitation itself fon the risk that the marriage Will end, ‘These results suggest that the previous findings linking cohabitation before marriage with increased risks of disruption were due entirely to the selection of the most divorce-prone into cohabitation, Further, Table 5 shows that the estimated correlation between the propensity to cohabit and the propensity to distupt a marriage is positive and significant. This finding suggests that there are unobserved differences across individuals which make those who are most likely to cohabit before any marriage also most likely to end any marriage that they enter. ‘These results support the reasoning, presented by Bennett et al. (1988), Teachman and Polonko (1990), Thomson and Colella (1992), and others, that people cohabit because they are relatively uncommitted to the institution of marriage Yet once we control for the endogeneity of cohabitation in the process leading to marriage breakup, having cohabited has no effect on chances of disruption. If information about the future spouse is important for the success of the marriage, and if one obtains more information from cohabiting than from dating, then we would expect a positive effect of having cohabited on marital stability, once we control for selection into cohabitation, Our results, however, fail to support this reasoning. ‘Although we see effects of having cohabited with the spouse before marriage in models that do not allow for correlation in the heterogeneity component across models, none of our models shows an effect of the duration of the cohabitation, Thus our results for a straightforward model of marital disruption do not match those of some previous studies (Bennett et al. 1988; Teachman and Polonko 1990). Duration of cohabitation, however, may not be linear in its effects on disruption Couples who cohabit for very short periods before marriage may have definite plans 10 marry when they begin living together. These couples, committed both to this particular ‘match and to the institution of marriage, may differ from couples who cohabit for extended periods before marrying. We test this reasoning by estimating an alternative specification of | the first three models contained in Table 4. This specification includes a dummy variable for cohabitations of three months or less, a variable for cohabitations from 3 to 12 months, and 4 Variable for cohabitation longer than 12 months. The omitted category is no premarital484 Demography, Vol. 32, No. 3, August 1995 cohabitation. Table 6 presents coefficients for these variables from the complete models. ‘The results in Table 6 show negative effects of about the same size for very short, medium, and long-duration cohabitations on the hazard of disruption in Models | and 2; this finding suggests very litle or no difference between these groups in impact on marital stability. In Model 3, which controls for heterogeneity, we see only very small effects on the hazard of disruption for all three measures; again, this suggests that very short cohabitations have no advantage as to stability To gauge the sensitivity of our results to the exclusion of some variables from one ‘equation but not from another, we reestimated each model that we presented in Tables 3, 4, and 5, excluding from both equations all variables that were excluded from either equation, so that there are no exclusion restrictions. As discussed earlier, this model is identified only by “within-person” marriage-to-marriage variation in cohabitation before marriage, and is not identified by any exclusion restrictions. The results from the model that allows for a correlation across the two processes are presented in the fourth column of Tables 3, 4, and 5. The results ofthis exercise Show that the main results hold in this specification: we find Statistically significant effects of cohabitation on the hazard of marital disruption only when ‘we do not take into account the self-selection of the most divorce-prone women into premarital cohabitation Other Determinants of Marital Disruption We summarize very briefly our findings on the other determinants of marital disruption, because these match quite closely the findings of previous research. Our model of marital disruption includes measures of two distinct types of “duration.” each a clock, so to speak, that started ata different point. One of these is the individual's age: the other isthe duration of the marriage. Both have significant effects on the chances that the marriage will end. From age 18 on, the hazard of disruption decreases significantly with each one-year increase in the woman's age: the largest declines occur between ages 18 and 24, smaller declines from 24 to 28, and smaller. less significant declines for each year of age after 28 ‘The coefficients for marital duration show that the hazard of disruption is lowest at the wedding. rises quite substantially over the first several years of marriage, and then increases ‘much more modestly, Three variables in the model index shated parenthood: whether the woman is currently pregnant, the number of children born in this marriage, and the number of children born while the woman is cohabiting with the man she eventually marries. Tables 3.4, and 5 show strong negative effects of a current pregnancy on the chances of disruption, ‘and smaller but still very significant effects of children born to this marriage. Children born ‘Table 6. Hazard of Marital Dissolution: Categorical Measures of Duration of Cohabitation o Cohab. with Marriage Partner Prior Cohabitation x 0-3 Months 0.38009" 0.4242" 0.0582 Prior Cohabitation x 4-12 Months 0.2302 *« 0.2611 Prior Cohabitation x 13+ Mont 0.36899 0.37809" NOTE: This model includes the other covariates present in Table 4. p = 08: ‘otal.Premarital Cohabitation 455 with this husband but before the marriage, however, do not stabilize the marriage (Although the coefficient is negative, itis estimated imprecisely.) CONCLUSIONS In this paper we use information from a large sample of young adults to examine the relationship between cohabitation before marriage and the subsequent stability of the marriage. We correct for adverse selection, of those most. prone to divorce, into cohabitation, When we do not correct for selectivity, we find that having cohabited with this partner before marriage has a strong positive effect on the chances that the marriage will end. Correction for selectivity, however, completely eliminates the effect of prior ‘cohabitation on marital dissolution. One’ explanation of this result—as others have argued—is that those with the least commitment to the institution of marriage are not only ‘most likely to cohabit atthe start ofa relationship but also most likely to dissolve a marriage that occurs later ‘We reasoned that couples who live together before they marry have a chance to gather information about the quality of marriage to this partner. Presumably those who dislike the prospects do not proceed to marriage: the couples who eventually marry are selected for the high quality of their match. According to this reasoning, cohabitation before marriage should increase the stability of the marriage, once we remove the effects of adverse selectivity, because couples who marry without first living together lack intimate information about the quality of the match, ‘Our results failed to support this reasoning: we found no significant positive effect of ‘cohabitation on marital stability, once we removed the effects of selectivity. It may be that some couples use the information gained during cohabitation to avoid bad matches and legalize good ones. It also may be, as Axinn and Thornton (1992b) and Thomson and Colella (1992) believe, that the experience of cohabiting changes people's views of marriage, making them less strongly committed (0 the institution and more willing to divorce. This change in commitment to marriage as an institution may decreuse marital stability, even while improved information about the quality of the match increases it In addition, cohabitation may be so different from marriage in its structure and expectations (Brines and Joyner 1993) thatthe information gained during cohabitation is not especially useful in predicting the quality of marriage with this partner, Dating couples may be able to match on the most important characteristies without living together. Also. events that unfold after the wedding may be an important component of marital success and may be unrelated to having cohabited. Because of the increasing importance of cohabitation in the process of union formation together with mounting evidence that informal unions differ in fundamental ways from legal unions, an understanding of the causes and consequences of these choices is a key to understanding the family today. Also, as this research helps to point out, we must focus on tunderstanding the connections between different decisions—in this case, the decision to cohabit before marriage and the decision to disrupt the mariage. NOTES “This isa vector of piecewise-linear spine variables with nodes at 1Wo and five years This is also a vector of piecewise-linear spline variables. with nodes at ages 18. 24. and 28 * Because the NLS-72 represents a single cohort, the effets of calendar tine are confounded in age and duration of mariage. and are not identitied separately.456 Demography, Vol. 32, No. 3, August 1995 “This is also a vector of piecewise-linear spline variables with a node atone year it is defined 10 be O before the fist child. The duration effect itself begins at 0, but the hazard undergoes a shift due to the birth of a child, and shift due to being pregnant. These shift variables are included in XY) * We use the normal distibation because of is ease of computation inthe bivariate case. as noted below. Other functional forms, such asa bivariate finite mixture, willbe considered in later research "The term 1), = means thatthe endpoint to the th, or last intemal, 1, isthe curent time ‘at which the survivor funetion i being evaluated. "Tis function uses analytic first derivatives and scores (BHH) second derivatives. "Ic is parametric because of the joint normality assumption for the two individual effects (heterogeneity components). For a discussion of fixed-effect estimators in panel data, see Hsiao (1986), "Sample weights are used in the analysis to make the results representative of a population of high school seniors in the United States. " Bighty-six percent of black 16- and 17-year olds versus 89.74 of whites ofthe sume age were enrolled in school in October 1970 (unpublished tabulations, U.S. Bureau of the Census), "Table 2 also indicates that much higher percentages of second and third marriages were preceded by a nonmartal cohabitation (22% vs. 54%). The prevalence of cohabitation andthe patern fcross murviages are roughly consistent with data’ from the National Survey of Families and Households (NSFH) as presented by Bumpass and Sweet (1989). ‘Those authors find, for example, ‘hat for First marriages oceurting between 1975 and 1979, 30% ofthe espondents cohabited with their spouse before marrage. For second martiages in the same period, 40% cohabited with their spouse. See Bumpass and Steet (1989) fora comparison ofthe NSFH withthe NLS-T2 in the procedures used to collect cohabitation histories, and how the differences may account for some differences in ‘measuring the prevalence of cohabitation, "The equation for marital disruption also includes @ measure of whether the individual was evaluated in high school as having a handicap. We argue tha existence of a handicap increases the cost of suecessul marital search and may inerease chances of disruption in the absence of either mote intensive or more extensive search, AS we see later, however, this variable bas no effects on the hazard of disruption, REFERENCES: Axinn, W. G. and A. Thomton, 1992a, “The Influence of Parental Resources on the Timing of the “Transition to Marriage.” Social Seience Research 21:261-85, 1992b, “The Relationship between Cohabitation and Divorce: Selectivity or Causal Influence?” Demography 20:387-74 Becker. G. S.. E, Landes. and R. T, Michael, 1979, “An Economic Analysis of Marital Instability.” “Fournal of Political Economy 88:1141-83. Bennet. N. G., A. K, Blanc, and D. E, Bloom. 1988, “Commitment and the Modetn Union ‘Assessing the Link between Premarital Cohabitation and Subsequent Marital Stability.” American Sociological Review $3:127-38, Booth, and D. Johnson. 1988, "Premarital Cohabitation and Marital Success." Journal of Family Issues 9255-72 Brines, J. and K. Joyner. 1993. “The Ties That Bind: Principles of Stability in Married and ‘Cohabiting Unions.” Presented atthe annual meetings ofthe Population Association of America Cincinnati Bumpass. L. and J. Sweet, 1989. “National Estimates of Cohabitation.” Demography 26:615-25, Bumpass. L. LJ. A. Sweet, and A. Cherlin 1991. “The Role of Cohabitation in Declining Rates of Mariage.” Journal of Marriage and the Family 53:913-27, DeMaris, A. and K. V, Rio, 1992, “Prematital Cohabitation and Subsequent Marital Stability’ in the United States: A Reassessment." Journal of Marviage and the Family 54:178-90, Glick, P. C. and A. J. Norton. 1977. “Frequency, Duration, and Probability of Marriage and Divorce.” Journal of Marriage and the Family 33:307-17,Premarital Cohabitation 457 Glick, P. C. and G. B. Spanier, 1980, “Maried and Unmarried Cohabitation inthe United States.” “Journal of Marriage and the Family 42:19-20. Goldscheider, F. K. and L. J. Waite. 1986. "Sex Differences in Entry into Marriage.” American ournal of Sociology 92:91-109, Grosshard-Shiechtman, 8. 1993. On the Economics of Marriage: A Theory of Marriage, Labor, and Divorce. Boulder. Westview. Hogan, D. P, 1978, "The Effects of Demographic Factors, Family Background, and Early Job “Achievement on Age at Mariage.” Demography 15:161-75. Hsiao, C. 1986, Analysis of Panel Data. New York: Cambridge University Press. Jacob, H. 1988, The Silent Revolution: The Transformation of Divorce Law in the United States. Chicago: University of Chicago Press. Landale, NN. 8, and R. Forse. 1991. “Patterns of Entry into Cohabitation and Marriage Among Mainland Puerto Rican Women.” Demography 28:587-607, Lilla, L. A, 1993, "Simultaneous Equations for Hazards: Marriage Duration and Ferility Timing.” owrnal of Econometrics 56:189-217. Lillard, L. A. and L, J, Waite. 1993. “A Joint Model of Childbearing and Marital Disruption.” Demography 30:683-81 MeCarthy, J. 1979, "Religious Commitment, Affiliation, and Mariage Dissolution.” Pp. 179-97 The Religious Dimension: New Directions in Quantitative Research, edited by R, Wuthnow. New York: Academic Press. Michael, R. T. and N. B, Tuma. 1985. “Entry into Marriage and Parenthood by Young Men and ‘Women: The influence of Family Background.” Demography 22:515—44. Oppenheimer, V.K. 1988, “A Theory of Marriage Timing.” American Jounal of Sociology 94:563-91 Peters, H. E. 1986, "Marriage and Divorce: Informational Constraints and Private Contracting “American Economic Review 76:437-54 Rindfuss, RR, and A. VandenHeuvel. 1990. “Cohabitation; A. Precursor to Martiage or an [Altemative to Being Single?” Population and Development Review 16:T03-26. South, S.J. and K. M. Loyd. 1995. “Spousal Alternatives and Marital Dissolution.” American Sociological Review 60:21-35, South, S. 3. and G. D. Spitze, 1986, “Determinants of Divorce over the Marital Life Course.” American Sociological Review $1:583-90, South, S. J. and K. Trent, 1988, “Sex Ratios and Women’s Roles: A Cross-National Analysis.” ‘American Journal of Sociology 93:1096-1115. Teachman, J.D. and K, A. Polonko, 1988. “Timing of the Transition to Parenthood: A “Multidimensional Bir Interval Approach.” Journal of Marriage and the Family 47:867-79, 1990. “Cohabitation and Marital Stability in the United States.” Social Forces 69:207-20. ‘Teachman, J. D., J. Thomas, and K. Paasch, 1991, “Legal Status and the Stability of Coresidenial Unions.” Demography 28:571-86, ‘Thomson, E, and U. Colella, 1992. “Cohabitation and Marital Stabilitiy: Quality or Commitment?” Journal of Marriage and the Family 54:259-68. ‘Thomon, A. 1988. "Cohabitation and Mariage inthe 1980s.” Demography 25:497-S08. ‘Thorton, A.. W. G, Axinn, and D. H. Hill. 1992, “Reciprocal Effects of Religiosty. Cohabitation, ‘and Mariage.” American Journal of Sociology 98:628-S1 Waite, L, J. F. K. Goldscheider, and C. Wisberger. 1986. "Nonfanily Living and the Erosion of “Traditional Fanily Orientation among Young Adults.” American Sociological Review S1:S41-S4, Waite, L. J. and L. A, Lillard, 1991. “Children and Marital Disruption.” American Journal of Sociology 96:930-53. Waite, L. J. and G. D. Spitze. 1981 Ts:681-94, Willis, R.. and R. T, Michael, 1994, “Innovation in Family Formation: Evidence on Cohabitation in the U.S." pp. 119-145 inj, Ermisch and N, Ogawa (eds.), The Family, the Market andthe State fof Aging Societies. London: Oxford University Pres. “Young Women's Transition to Martiage."” Demograph