An Empirical Test of Crisis, Social Selection, and Role Explanations of the Relationship Between Marital Disruption and Psychological Distress: A Pooled Time-Series Analysis of Four-Wave

Journal of Marriage and Family 64 (February 2002): 211–224 211

DAVID R. JOHNSON University of Nebraska—Lincoln

JIAN WU BellSouth*

l

An Empirical Test of Crisis, Social Selection, and Role

Explanations of the Relationship Between Marital

Disruption and Psychological Distress: A Pooled

Time-Series Analysis of Four-Wave Panel Data

Although a higher level of psychological distresshas been found in many studies of divorced com-pared with married individuals, explanations forthis difference remain elusive. Three basic theo-retical explanations have been proposed. Socialrole theory maintains that the role of being di-vorced is inherently more stressful than that ofbeing married; crisis theory attributes the higherstress to role transitions and transient stressors ofthe disruption process, and social selection theoryclaims that the higher stress levels among the di-vorced result from the selection of people withpoor mental health into divorce. Some empiricalsupport is available for each of these approaches,but all three have not been tested simultaneouslyin a longitudinal study. This research empiricallyevaluates the efficacy of these theories in a pooledtime-series analysis of a four-wave panel of mar-ried persons followed over 12 years. The pooled-time series random effects model was used to es-

Department of Sociology, University of Nebraska—Lin-coln, Lincoln, NE 68588-0324 ([email protected]).

*BellSouth, 2180 Lake Boulevard, Atlanta, GA 30319.

Key Words: divorce adjustment, panel study, psychologicaldistress.

timate the effects of social roles, crisis, and socialselection. The results provide evidence that thehigher stress levels of the divorced primarily re-flect the effect of social role with selection andcrisis effects making small contributions only.

A number of previous studies clearly show thatthe divorced report higher levels of psychologicaldistress than do the married (Booth & Amato,1991; Coombs, 1991; Gove, Hughes, & Briggs,1983; Mastekaasa, 1994; Ross, 1995; Waite,1995). Empirical studies have not been as defini-tive about the factors that account for this rela-tionship. Each of the three explanations that havebeen proposed have some empirical support. Asocial selection explanation maintains that personswith high psychological distress and mental dis-orders are disproportionately selected into divorceand less likely to remarry, yielding higher distressscores among the currently divorced (Aseltine &Kessler, 1993). According to crisis theory, the dis-ruption process and resultant role transitions tem-porarily elevate distress (Booth & Amato). Roletheory attributes the greater psychological distressreported by the divorced to the more difficult lifecircumstances they experience (Ross).

212 Journal of Marriage and Family

An adequate simultaneous test of these com-peting explanations in a single study has been lim-ited by small sample sizes, cross-sectional de-signs, limited duration of panel studies, lack ofmeasures of distress pre- and postdisruption, andanalysis design limitations. The purpose of thisstudy is to overcome some of these limitations bytesting these explanations with pooled-time seriesmodels in a large nationally representative four-wave panel sample extending over a 12-year pe-riod. Measures used include psychological distressboth pre- and postmarital disruptions.

SOCIAL SELECTION

One explanation for the higher psychological dis-tress of the divorced compared with the marriedis that married persons with preexisting mentalhealth problems are often inadequate marital part-ners and thus are more likely to be selected intodivorce than persons without these problems. Thishas been referred to as the social selection hy-pothesis (Avison, 1999). This theoretical perspec-tive asserts that at least part of the relationshipbetween distress and divorce is not causal and re-flects the influence of stable individual personalityand social characteristics on the odds of divorcingand on psychological distress. A number of re-search studies have examined the selection hy-pothesis (Aseltine & Kessler, 1993; Bloom, Niles,& Tatcher, 1985; Davies, Avison, & McApline,1997; Kitson & Sussman, 1982; Mastekaasa,1997; Robins & Regier, 1991), but the magnitudeof the selection effect and the extent of its impacton the observed difference in distress levels be-tween married and divorced persons are less cer-tain. Studies involving small samples followedover time have not found significant differencesin psychological disorders at the beginning of thetime period between persons who do or do noteventually divorce (Doherty, 1983; Menaghan,1985; Menaghan & Lieberman, 1986). Thesesame studies did find a large difference in psy-chological disorders at the end of the time periodbetween the two groups, suggesting that even ifselection is at work, it does not account for muchof the observed difference. Other studies withlarger samples followed over time (Booth & Ama-to, 1991; Wertlieb, Budman, Demby, & Randall,1984) found significantly higher psychologicalstress levels at the beginning of the time periodfor persons divorcing in the future compared withthose who were not, although controlling for these

did not eliminate significant differences in distressfollowing divorce.

The problem with interpreting these findings assupport for the selection explanation is that theactual breakup of the relationship is often the lastpoint in a long-term process of dissolution (Kitson& Morgan, 1990), causing a predivorce rise in dis-tress that may extend back several years. Whetherthe dissolution process is a cause or effect of pre-divorce distress levels would be difficult to sepa-rate in the short-term panel studies most commonin the literature. Panels that measure distress 5 to10 years predivorce would do a better job of sep-arating these effects because of the greater tem-poral distance from the event. High levels of psy-chological stress found many years predivorce aremore likely to reflect enduring personality factorsor more serious mental disorders and not the re-sults of stress from a dissolving relationship. Thepanel data used here contain measures of psycho-logical distress obtained up to 11 years before adivorce occurred. Finding that persons who di-vorce, compared with those who do not, have sig-nificantly higher stress levels many years before adivorce would provide some support for the se-lection argument. Of course, some marriages maybe habitually conflict ridden and unsatisfying,leading to high distress levels well before the startof a dissolution process. In these case, high levelsof distress many years before the marriage maysignal a chronic relationship problem independentof individual predisposing factors (Johnson &Booth, 1998).

It is important to differentiate between the typeof social selection process that accounts for whydivorced persons have higher psychological stresslevels than the married and the more general issueof personality and social factors that predict se-lection into divorce (Gottman, 1994). High levelsof psychological distress occurring well before adivorce may reflect underlying stable personalitydisorders, but many personal characteristics maymake people poor marital risks without elevatingtheir levels of psychological distress.

CRISIS THEORY

Crisis theory (Booth & Amato, 1991; Tschann,Johnston, & Wallerstein, 1989) views marital dis-ruption as a life crisis that temporarily changesmental health. Marital disruption is considered aserious life challenge that stresses those involvedin the process (Mastekaasa, 1992). A crisis in-volves a life events stressor that is a discrete, ob-

213Marital Disruption and Psychological Distress

servable event, the course of which involves botha beginning and an end (Wheaton, 1999). Lifechange theory (Holmes & Rahe, 1967), whichpredicts that role transitions are stressful, espe-cially those related to divorce or permanent sep-aration, also falls under the rubric of crisis theory.The role transition process is viewed as the sourceof stress; once the transition has passed, psycho-logical distress should return to pretransition lev-els (Booth & Amato; Wheaton).

If the stress of role transitions and the disso-lution process leads to higher distress scoresamong divorced persons, we would expect psy-chological distress to increase until the actual di-vorce or permanent separation (Kitson & Morgan,1990) and fall as the transition passes. Booth andAmato (1991) found evidence in panel data thatpsychological stress rose before the crisis and thenreturned to levels comparable to those of marriedindividuals two years after the marital disruption.This supports the predictions of the stress model.Studies only examining the relationship betweentime since disruption and psychological symptoms(Amato & Partridge, 1987; Aseltine & Kessler,1993; Brown, Felton, Whiteman, & Manela, 1980;Chiriboga, Roberts, & Stein, 1978; Plummer &Koch-Hattem, 1986), however, found no signifi-cant decline in distress as the distance from theevent increased. These findings are not consistentwith the crisis model to the extent that the distressis a consequence of transition process.

ROLE THEORY

Role theory focuses on the relatively constant andenduring, or chronic, stresses and strains of certainroles (Pearlin, 1999). Living as a divorced indi-vidual often involves social isolation, lack of so-cial support, economic hardship, and added child-care responsibilities (for parents), or stressesassociated with child responsibilities (custody ar-rangements, etc.; Booth & Amato, 1991; Mc-Lanahan, 1983; Ross, 1995; Waite, 1995). Notonly are there likely to be more chronic stressorsin the everyday life of the divorced comparedwith the married individual, but they may be lesslikely to effectively cope with these stressors be-cause of less available social support (Avison,1999).

According to role theory, the greater distresslevels of divorced persons are a permanent featureof that state. Because no adjustment process is im-plied, length of time since the divorce should notbe related to level of psychological distress. Re-

marriage or cohabiting by divorced personsshould reduce distress levels if this explanation iscorrect (Ross, 1995). Entering a new conjugal re-lationship would reduce the chronic stressors ofsingle parenthood, increase the odds of greatereconomic stability, and increase the levels ofavailable emotional and social support. Studiesthat found higher psychological stress levelsamong the divorced (Ross, 1995), and no declinein distress level over time from post-divorce(Amato & Partridge, 1987; McLanahan, 1983;Mastekaasa, 1994) provide evidence for the valid-ity of the social role perspective.

DIFFERENTIAL EFFECTS

It has been proposed that the psychological con-sequences of divorce or permanent separation willdepend on both pre- and postdivorce characteris-tics and conditions (Aseltine & Kessler, 1993;Booth & Amato, 1991). A particularly robust find-ing is the improved mental health status of per-sons leaving conflict-filled and low-quality mar-riages (Aseltine & Kessler; Booth & Amato;Gove, 1972; Menaghan & Lieberman, 1986;Tschann et al., 1989; Wheaton, 1990).

The resources and responsibilities availablepostdivorce also should condition the extent of thestress induced by this role. Previous studies haveidentified available income, education, number ofclose friends, residence, presence of children, andrelated factors that influence the course of post-divorce stress (Aseltine & Kessler, 1993; Berman& Turk, 1981; Booth & Amato; Doherty, Su, &Needle, 1989; Goetting, 1981; Hetherington, Cox,& Cox, 1978; Menaghan & Lieberman, 1986;Tschann et al., 1989). Many of these factors weretested in cross-sectional studies. Further tests ofsome of these effects are still needed in the con-text of a more complete longitudinal model of fac-tors producing psychological distress among di-vorced persons.

Findings from previous studies lead us to ex-pect that the pattern of psychological distress as-sociated with marital dissolution may differ bygender (Horowitz, Raskin White, Howell-White,1996). Divorced women compared with men aremore likely to have declines in standard of livingand are more likely to have custody of the chil-dren. Both of these may increase the chronic stressof the divorced role for women. Studies havefound that single mothers (including divorced) re-port higher psychological distress than marriedmothers (Avison, 1999). On the other hand, sev-


eral studies have shown that in regard to healthand psychological well-being men generally ben-efit more from a marital relationship than dowomen (Waite, 1995).

MODEL TESTED

In this article we develop and test a model thatcan simultaneously evaluate the extent to whichthese theoretical explanations account for thehigher levels of psychological distress observedfor divorced compared to married persons. Themodel is designed for multiple-wave panel data inwhich observations of psychological distress, cur-rent marital role, date of the occurrence of maritalevents (divorce, remarriage, cohabitation, etc.),and other background variables are measured ateach wave.

The data to be analyzed by this model are or-ganized as a pooled-time series. In such a data set,each wave of observations for each individual inthe sample is represented by a separate record(Johnson, 1995) so the total sample size is thenumber of individuals times the number of waves.This data structure violates the assumption of in-dependent observations because the same respon-dent contributes more than one record to the dataset. In this situation, ordinary least squares re-gression analysis is inappropriate. Two differentestimators—fixed effect and random effects–havebeen developed for data organized in this manner(Allison, 1994). The random effects model is ageneralized least squares solution in which themodel parameters are solved through a weightedcombination of within- and between-individualcovariances (Hsiao, 1986). This method allowsvariables to be included in the equation that donot vary over time for individuals (e.g., gender,race) and is normally more efficient than the fixedeffects estimator (Allison).

The fixed effects estimator for pooled time-se-ries models is based entirely on variation overtime within individuals in the sample. Variablesthat are invariant over time must be excluded fromthe model. In addition, because information aboutvariation between individuals is not used to esti-mate parameters, the method is less efficient andstandard errors are higher.

The fixed effects estimator, which is limited tovariables that vary over time within individuals,can produce estimates that are net of all observedor unobserved differences between individualsthat are time-invariant. The estimates are not bi-ased by cohort or selection effects or the effects

of individual differences that are time-invariant(e.g., background variables, genetic tendencies,stable personality predispositions).

For this analysis, the random effects model isthe estimator of choice because one of the threeexplanations we want to investigate involves a be-tween-individual effect—the effect of social se-lection. Because of the stronger inferences possi-ble with the fixed effects solution, we alsoestimate models by fixed effects with the time in-variant variables excluded. Doing this allows usto evaluate the possibility of specification errorsin the random effects solution. Both models re-quire at least two time periods for each individualin the sample but allow the number of time peri-ods included to vary among individuals. This ad-vantage is important in the analysis of multiple-wave survey panels in which attrition is common(Johnson, 1995).

The basic random-effects model for a pooleddata set with i individuals and t time periods is asfollows:

P 5 u 1 b S 1 b M 1 b D 1 b Wit 1 it 2 i 2 it 3 it

1 b C 1 b T 1 b F 1 b X4 it 5 it 6 it k ki

1 b Z 1 e ,j jit it

where u is a constant term, eit an error term, andthe bs are regression coefficients. The variablesare as follows: Pit psychological distress; Sit is ameasure of social selection (1 5 individuals whoever divorced or permanently separated, 0 5 allothers); Mi a measure of whether the respondenthad been in at least one previous marriage beforethe first wave (1 5 previously married, 0 5 allother); Dit current divorce status of individual i attime t (1 5 divorced / permanently separated, 05 all other); Wit current widowed status (1 5widowed, 0 5 all other); Cit current cohabitingstatus (1 5 cohabiting, 0 5 all other); Tit a mea-sure of the time to a marital disruption; Fit a mea-sure of the time since a marital disruption; Xki ktime-invariant control variables; and Zjit j time-varying control variables.

The Tit and Fit terms that measure proximity tomarital disruption are used to estimate crisis ef-fects. We hypothesize that if the crisis model iscorrect, we will find a significant increase in dis-tress as the divorce approaches and a significantdecrease following the disruption. In the modelstested, these were included in reverse coded formso that the higher the score, the closer the disrup-tion event. This coding also facilitates interpreta-


tion of the parameters associated with the socialselection effect (Sit). If the social selection modelis correct, we hypothesize that Sit will have a sig-nificant positive effect on psychological distressnet of other factors. The parameter Sit, coded as adummy variable to indicate whether the personever divorced or was permanently separated dur-ing the course of the study period, can be inter-preted as the effect on distress well before or aftera past or future disruption—not one close to theevent. Current marital role is measured using threedummy variables (Dit, Wit, and Cit), with the ref-erence (omitted) group being the currently mar-ried. If the social role explanation is correct, wehypothesize that the difference between the cur-rently married and the currently divorced will bestatistically significant not of other factors includ-ed in the model. Random and fixed effects esti-mates for this model were computed with thextreg procedure in STATA (StataCorp, 1997).

Models were also developed that included in-teractions between a number of the variables totest for differential effects on distress. Differentialeffects by gender for a most of the variables inthe model were tested by including sets of inter-action terms. Marital happiness in the wave im-mediately preceding a disruption was introducedinto the model to evaluate if the effects of a dis-ruption on psychological distress differed for per-sons departing a relatively unhappy versus a rel-atively happy marriage.

METHOD

The Sample

This study uses four waves of panel data spanning12 years. The first wave was collected over thetelephone in 1980, making use of a random-digitdialing national probability sample. Completed in-terviews were obtained for 2,033 persons aged19–55 in intact marriages. The refusal rate in 1980was 18%, and the overall response rate comparedfavorably with those for other telephone surveys(Groves & Kahn, 1979). Follow-up interviewswith all respondents who could be located andagreed to participate were conducted in 1983,1988, and 1992. The percentage of the originalsample reinterviewed in each of the subsequentwaves was 78% in 1983, 66% in 1988, and 58%in 1992. Attrition was greater for younger respon-dents, men, renters, southern residents, AfricanAmericans, and those residing in metropolitan ar-eas, but differences were small (Booth, Amato,

Johnson, & Edwards, 1993). Logistic regressionmodels with attrition over the four waves as thedependent variable found no significant effect ofpsychological distress measured in 1980 on panelattrition. Attrition also was not significantly relat-ed to several measures of marital quality (e.g.,marital happiness, divorce proneness) in 1980.

The analysis was based on a data set thatpooled the results from the four waves of thestudy. Records for respondents were included ifthey had answered the survey in two or more ofthe waves. The total sample size analyzed herewas 5,676 records from 1,593 different individu-als represented in the sample, with the averagerespondent contributing 3.56 records to the pooleddata set. During the 12 years of the study, therewere 243 divorces and permanent separations and84 remarriages. In the pooled data set, 804 of therecords were contributed by those who ever di-vorced or permanently separated during the 12-year period. The number of divorces and remar-riages may appear small for this size sample giventhat around one half of all marriages are expectedto end in divorce, with the majority of them re-marrying (Cherlin, 1981). Because the 1980 sam-ple was representative of persons in intact mar-riages, many were first interviewed after survivingthe earlier years of marriage (which have muchhigher dissolution rates). Therefore, we would ex-pect much lower rates of divorce than would havebeen found if the 1980 sample had been restrictedto newlyweds. A survival analysis of divorceprobability for the 385 persons in the sample whowere married 3 or fewer years in 1980 revealedan expected failure rate of 35% after 12 years ofmarriage, a figure that approaches the rate of mar-ital failures reported for this cohort (e.g., Cherlin).

Measures

A scale created from five items included in allfour waves was used to measure psychologicaldistress. The first item was an indicator of mentaldistress derived from responses to the question:‘‘Have you felt extremely unhappy, nervous, ir-ritable, or depressed?’’ Those responding no werecoded 0, yes in the past 3 years were coded 1, yesrecently were coded 2. Those who responded ‘‘re-cently’’ were asked in a follow-up questionwhether this led them to cut down on activitiesfor several days or more. Those responding yeswere coded 3. The second item was the respon-dent’s self-rating of their health status: ‘‘In gen-eral, would you say your own health is (1) excel-


lent, (2) good, (3) fair, or (4) poor?’’ A third itemtapped global happiness: ‘‘Taking all things to-gether, how would you say you are these days?Would you say you are (1) very happy, (2) prettyhappy, or (3) not too happy?’’ The last two itemswere perceptions related to life satisfaction:‘‘Would you strongly agree, agree, disagree, orstrongly agree that things are as interesting as theywere?’’ and ‘‘Would you strongly agree, agree,disagree, or strongly agree that you have gottenwhat expected out of life?’’ These five items werecoded so high scores indicated more stress, zscored (based on means and standard deviationsfrom the 1983 wave), and combined in a scale thatwas the average z score of the items responded tocoefficient alpha reliability for the scale was .65;none of the items could be excluded without re-ducing alpha.

A more complete scale of psychological dis-tress was available in some of the waves, but weused the five-item version because it was the onlyone available in all four waves. Waves 2, 3, and4 contained 24 items measuring psychologicalstress/demoralization, including 8 items fromLangner’s 22-item index of psychological disorder(Langner, 1962). To assess the validity of the five-item measure, we compared it with more completescales in the 1983 wave (which had the largestsample size with responses to all 24 items). Wecorrelated the five-item scale with a summatedscale of the eight Langner items and a 17-itemscale created from factor and scale analyses of all24 stress items. The five-item stress scale is cor-related at .58 with the eight-item Langner scale (a5 .69) and .87 with the more reliable (a 5 .86)17-item stress scale, giving us confidence that ourfindings can be generalized to more completemeasures of psychological symptoms.

Measures of social selection, social role, andcrisis were created as discussed in the basicpooled-time series model presented above. Socialselection was measured by two indicators. Thefirst was a dummy variable coded 1 if during the12 years of the panel study (1980–1992) respon-dents ever divorced or permanently separated andcoded 0 otherwise. The second dummy variableindicated if the respondent had been in anothermarriage before 1980 (1 5 yes; 0 5 no). Bothmeasures are time invariant.

The measures of social role are time varyingand were created from the respondent’s currentmarital status at the time of the interview. Threedummy variables measured whether the respon-dent was currently divorced or separated, wid-

owed, or cohabiting. The omitted (reference)group was the married.

To measure the effect of crisis, two variableswere created to ascertain the proximity of the in-terview to and from a marital disruption. Respon-dents who were divorced or permanently separat-ed were asked to report time since this disruptionhad occurred (in months). From this measure andthe decimal year of the interview, we calculatedthe exact decimal date of the disruption. For re-spondents divorcing more than once during thepanel period, the earliest divorce was selected andinterview waves following the second divorcewere excluded from the data set. The small num-ber of persons (24) with missing information onthe time of the disruption were assigned the meannumber of months from the disruption reported byrespondents with complete data in the same wave.

Use of decimal years in the coding allowedprecise estimates of duration. Once the exact dec-imal date of the disruption was created, the dis-tance from this event was calculated for each re-spondent for each wave by comparing the decimalyear in which the interview took place with thedecimal year of the disruption. Decimal years toand from a disruption were coded as separate var-iables. For example, if a divorce occurred in June1987 (decimal year 87.5) and the respondent wasinterviewed in the first wave in August 1980 (dec-imal year 80.67) and in Wave 2 in May 1992 (dec-imal year 92.42), then the coding for years to thedisruption would be 6.83 decimal years in Wave1 and the coding for years from the disruptionwould be 4.92 decimal years in Wave 4. Personsnot experiencing a disruption during the 12 yearsof the panel were coded 0 on both duration mea-sures.

The actual measures of decimal years to andfrom the disruption used in the regression modelswere transformed to aid in the interpretation of theregression parameter estimates. The number ofyears to and from the disruption was subtractedfrom 5 so persons closest to the disruption had thehighest scores. All persons 5 or more years fromthe disruption were assigned a score of 0. For ex-ample, for the years to a disruption measure, aperson interviewed 1 year before their marriagedisrupted would be coded 4 (5–1) for that wave,whereas a person 7 years from disruption wouldbe coded 0 (5–7 5 22, with all negative numbersrecoded to zero). This procedure allows us to in-terpret the regression coefficient associated withthe social selection variable as the effect on dis-tress of persons who are still five or more years


TABLE 1. DESCRIPTIVE STATISTICS FOR VARIABLES IN THE POOLED DATASET

Variables M SD Minimum Maximum

Psychological stressSocial selectionRole (divorced/separated 5 1)Widowed (yes 5 1)Cohabiting (yes 5 1)

2.0635.1350.0463.0106.0076

.6551

.3417

.2102

.1023

.0867

21.742.000.000.000.000

2.9841.0001.0001.0001.000

Decimal years from divorceDecimal years to divorceDivorced before 1980Respondent’s ageGender of respondent (F3 5 1)

.2294

.3484

.093740.7012

.6098

1.16001.4960.2915

10.0776.4878

.000

.000

.00017.000

.000

11.34711.5551.000

68.0001.000

Years of educationNo. of children in householdPresence of children (yes 5 1)Total family income (in $1000)

13.83651.2736.6422

36.3150

2.66671.2092.4794

16.4742

.000

.000

.0002.500

30.0007.0001.000

65.000

Note: N 5 5,676.

from the marital disruption. We chose 5 years asthe point beyond which the measure would becoded as 0 based on an exploratory analysis ofthe scatterplots of years to and from disruption onpsychological distress. Lowess (Fox, 1991) non-parametric smoothed regression lines were fittedto the scatterplot of psychological distress byyears to and from disruption. These curvesshowed that the regression line further than 4 to4 years predisruption was flat. Closer to the dis-ruption, the lowess analysis showed a linear in-crease in psychological distress. The curve of psy-chological distress by time from disruption wasbasically flat in both the lowess and regressionmodels. The same coding was used for both in theinterest of consistency, however. Alternative mod-els for the effect of distance from disruption usingdifferent thresholds also were tested and are con-sidered in the discussion section.

Demographic and background measures wereincluded in the analysis as statistical controls.These were age of the respondent (in years), gen-der, years of schooling, number of children under18 residing in the household, total household in-come at the time of the interview, and an indicatorof whether the household income information wasmissing (less than 2% was missing). Marital hap-piness reported in the wave immediately preced-ing marital disruption was also used in the anal-ysis of differential effects. The measure of maritalhappiness was an 11-item summated scale ofitems measuring happiness with various aspects ofthe marriage. It has been used in a number ofprevious studies (c.f., Johnson, White, Edwards,& Booth, 1986). Descriptive information for the

pooled-time series data set on each of these vari-ables is presented in Table 1.

RESULTS

The first regression model simultaneously esti-mates, with the random effects estimator, the in-fluence of selection, crisis, and social role whilecontrolling for basic background variables. It ispresented as Random Effects Model 1 in Table 2.We first examine the evidence for the crisis ex-planation. The years from disruption were in thepredicted direction (higher stress levels closer tothe disruption), but the coefficient was close tozero and not statistically significant. Years to adisruption was significant (p , .01). Psychologi-cal distress scores predicted by the model for per-sons immediately before the disruption (whowould have a score of approximately 5 on thismeasure) were around one third a standard devi-ation above persons 5 years from a disruption(who score 0 on this measure). This finding sug-gests that going through marital dissolution beforethe actual disruption is stressful and becomesmore so as individuals move toward actual break-up of the relationship. Because scores did not de-cline following disruption, the crisis explanationappears inadequate.

The social selection variable in this equationmeasures the effect on psychological stress forpersons who experienced a disruption but were 5or more years away from it. The positive sign ofthe regression coefficient indicated slightly higherdistress scores for persons 5 or more years pre- orpostdisruption and was significant at the .05 level.


TABLE 2. REGRESSION ESTIMATES FOR POOLED TIME-SERIES MODELS WITH PSYCHOLOGICAL DISTRESS AS THE

DEPENDENT VARIABLE

Variable

RandomEffects

Model 1b

FixedEffects

Model 2b

RandomEffects

Model 3b

RandomEffects

Model 4b

RandomEffects

Model 5b

Time from disruptionTime to disruptionSocial selectionPrevious divorceDivorced/permanently separated

.0068

.0635**

.0981*

.1253**

.3022**

.0077

.0699**

.2821**

.0076

.0620**

.1044*

.1367**

.2870**

.0063

.0628**

.1044*

.2145**

.2912**

2.0301.0991**.2545**.2118**.3328**

WidowedCohabitingAgeGenderYears of schooling

.135922.923**

.0034**2.02202.0309**

.10762.2986**

.0091**

.12262.2962**

.0058**2.02032.0267**

.11802.4604**

.057**2.02192.0267**

.12282.4535**

.0053**2.02792.0268**

Number of children in householdTotal family incomeMissing income informationChild present 3 previous divorceChild present 3 cohabiting

2.0170*2.0021**2.0874

.0197*2.0021**2.08872.1289*

.3356*

.0174*2.0021**209002.1296*

.3537*Lag happy 3 time from disruptionLag happy 3 time to disruptionLag happy 3 divorced or permanently separatedConstantR2

.1992

.044**2.4535

.232**.0970.051**

.0984

.052**

.0948**2.0736*2.3168**

.1208

.063**

Note: Number of records 5 5,676; number of individuals 5 1,593.*p , .05. **p , .01.

This provides some support for the social selec-tion explanation. Nonetheless, the size of the ef-fect is small (only about one third that of the di-vorce effect), suggesting that much of thedifference is not accounted for in social selection.

The model estimates do provide substantialsupport for the social role explanation. The effectof divorce or separation was strong and signifi-cant. Persons divorced at the time of the interviewhad a psychological distress score around onethird of a standard deviation higher than did mar-ried persons (the omitted group). Widows hadhigher distress scores as well, but the effect wasnot significant. A third indicator of social role iswhether a divorced person is cohabiting. Cohab-iting status had a strong effect on distress reduc-tion, effectively eliminating the negative conse-quences of divorce. Because all cohabitors in thissample also are divorced, the net effect of cohab-iting is to reduce the stressful effect of the divorcerole. To compare the stress of living in a cohab-iting relationship with the stress of living in a mar-ried relationship, we added the divorced and co-habiting coefficients, which yielded an effect sizeclose to zero (.3022 2 .2923 5 1.0099). Di-vorced cohabitors had about the same stress levelsas married persons.

Another significant effect noted in Model 1was that persons previously married in 1980 hadsignificantly higher stress scores than those intheir first marriages in 1980. This could reflecteither a social selection or role effect. We explorethese alternative explanations in a later model.

Model 2 in Table 2 solves for the coefficientswith the fixed effect estimator. This model con-trols for the additive effects of all measured andunmeasured differences between respondents, in-cluding selection and cohort effects because thecoefficients are estimated using information onvariation within individuals. Because measuresthat do not vary within individuals must be ex-cluded, we drop social selection, previous divorce,and gender from the model. We also drop yearsof schooling because, although it does vary slight-ly within individuals, most of these changes ap-pear to be reporting errors. The fixed effects es-timates confirm the findings from the randommodels analysis, with the significant coefficientssimilar in magnitude and in the same direction.

Model 3 in Table 2 adds two variables—num-ber of children in the household and total familyincome—that might play an intervening role inthe effects of the divorce role on psychologicalstress. Because both the number of children pre-


FIGURE 1. EXPECTED CHANGES IN PSYCHOLOGICAL

DISTRESS ESTIMATED FROM COEFFICIENTS IN

MODELS 3 AND 5

sent in the household and income often changesfollowing a divorce, we would expect the effectsof the social role variables to be attenuated whenintroduced as controls. Although both number ofchildren and income are significantly related tostress levels, including them in the equation has aminimal effect on any of the role or crisis coef-ficients. We cannot account for the greater stressof the divorced role by changes in income or num-ber of children in the household.

The implications of the effects found in themodel can best be visualized by plotting thechanges in stress expected in the model as indi-viduals age, dissolve their marriages, and remarry.In Figure 1 we plot the curves predicted by therandom effects from model 1 for three hypothet-ical persons from ages 30–46. These trajectoriesare indicated by the solid lines in the figure. Oneperson is in his or her first marriage and remainsmarried over the age range (labeled continuouslymarried in the figure). The second person also re-mains married throughout the period but had beenmarried previously (labeled remarried). The thirdperson divorces for the first time at age 38 andremarries at age 42 (labeled divorced and remar-ried). All other variables in the equation are heldconstant at their means during the period. The tra-jectories for predicted distress scores for the mar-ried and previously married are basically straight.There is little change in distress levels with age,and the previously married had slightly higherstress levels than the married. The person whodivorces at age 38 begins at about the same levelas the previously married person, due to the smallbut significant social selection effect observed.Five years before the disruption a sharp increasein distress level occurs. At the disruption, distress

increases slightly and remains relatively high andat a nearly constant level. Although the figureshows a gradual decline following divorce and be-fore remarriage, this decline was not statisticallysignificant in the model. When they remarry, theirpsychological distress dropped to almost exactlythe same levels predicted 5 or more years beforea disruption. Because their stress levels also areapproximately the same as those found for pre-viously married persons when the study began,pairing these findings together supports the oper-ation of a small but consistent selection effect.

The next set of models introduce interactioneffects into the basic model to test for differentialeffect on the stress process. A number of inter-actions were tested but most were not statisticallysignificant. Because previous studies have founddifferences between men and women in the effectsof marital disruption on stress, we examined anumber of interactions with gender. For example,the presence of children might make disruptionsmore stressful for women than men because wom-en usually retain custody of children. We alsomight expect disruptions including children to bemore stressful.

To fully evaluate the effects of gender on allthe coefficients in Models 1 and 3, we comparedthe fit of a model, which allowed separate coef-ficients for men and women for all variables inthe model with the fit of a model in which menand women had the same coefficients. There wasno significant difference in the R2 of these models,and only two of the coefficients had significantlydifferent gender effects. These were the effects ofeducation and the effects of income on distress.The effect of education was significantly larger forwomen, and the effect of income only held formen. We concluded that the effects of the divorceprocess on distress estimated in these models didnot differ by gender.

Interactions between presence of children andother variables in the model (including the firstseven independent variables listed in Table 2,Time to Disruption to Cohabiting) led to only twostatistically significant effects, both involvingnumber of children. The significant interaction ef-fects are found in Model 4 in Table 2. Presenceof children interacted significantly with the effectsof previous divorce and cohabiting on distress.Number of children was dichotomized into pres-ence (1) or absence (0) of children for the pur-poses of creating the interaction terms. When thisinteraction term is included in the equation, theadditive term for previous divorce measures the


effect for people without children. Results showthat those previously divorced without children re-port higher stress levels than do those previouslydivorced with children, although the distress lev-els for both were higher than for persons who hadnot been previously divorced. Previous divorcestatus had negative consequences, but the effectwas stronger for those without children. Thismight reflects a selection effect—the category ofpersons with multiple marriages but no childrenmay be more likely to include persons with seri-ous preexisting mental disorders who are poormarriage risk.

According to the second significant interactionterm, cohabiting following a divorce had a stron-ger effect on reduced reported distress (2.4604)among those without than it did for those withchildren present (2.4904—.3356 5 2.1148). Al-though cohabiting with children present may in-crease social support and reduce stresses for theparent, the strains to parent-child relationshipsbrought about by the introduction of another per-son to the household may mitigate some of thesepositive effects on the parent’s psychological well-being.

In Equation 5 of Table 2, we test for the effectnoted in previous studies: Persons leaving stress-ful or unhappy marriages often exhibit improvedpsychological well-being. To test for this effectamong those who divorced or permanently sepa-rated during the observed period, we divided thescores on the 11-item scale for marital happinessobtained in the interview immediately precedingthe disruption into two groups (1 5 relativelyhappy, 0 5 relatively unhappy) by cutting thescale at the median score. We then multiplied thisdichotomized lagged marital happiness indicatorby decimal years to and from marital disruptionand by the divorce social role indicator and addedthem to the equation. The results are presented inModel 5 of Table 2. A model was also tested thatdid not dichotomize lagged marital happiness andyielded similar findings. All three of the interac-tion terms were statistically significant. The im-plications of these interaction effects for predicteddistress scores for persons relatively happy andrelatively unhappy with their marriage in the pre-disruption wave are plotted as the dashed lines inFigure 1. We plotted expected scores for personswith relatively high (‘‘Happy before divorce’’)and relatively low marital happiness predisruption(‘‘Unhappy before divorce’’). The patterns of psy-chological distress predicted by the equation differdramatically. For persons in troubled marriages

(low happiness), the predicted level of stress in-creases steeply to the disruption then declinessharply following it. Distress level increased again(although this effect is not statistically significant)and then dropped following remarriage. For per-sons in a relatively happy relationship before thedisruption, only a slight increase occurred up tothe disruption, but the disruption itself produceda sharp gain in stress scores (although they do notreach the level predicted for a person leaving amore troubled relationship). Following the disrup-tion, the stress levels declined with increased timefrom the disruption, indicating some support for acrisis model in this specific group. Remarriage didlittle to reduce distress in this group.

DISCUSSION

Our findings clearly support a social role expla-nation for the higher stress levels of the divorced,provide only limited support for a stress model(and only for persons who were relatively happybefore they divorced), and found evidence of asmall social selection effect. Our findings appearto be somewhat at odds with those reported byBooth and Amato (1991), who analyzed the firstthree waves of data used here. It is possible thatsome of the assumptions about the models wetested, such as a 5-year threshold level for theeffects of stresses caused by the dissolution pro-cess, might be producing erroneous results. Also,the relatively limited measure of psychologicalstress might be producing biased results. Becauseof these concerns, we evaluated the robustness ofour findings in several different ways.

Two differences between findings by Boothand Amato (1991) and this study need to be rec-onciled. First, Booth and Amato concluded thatthe increased distress before divorce and declineobserved afterward is consistent with a stressmodel. The increase before divorce is consistentin both studies, except that we map this more pre-cisely by including a variable measuring time indecimal years to divorce, whereas they only usethe wave intervals (3 or more years) to time theevents. We only found a decline in distress fol-lowing disruptions of relatively happy marriag-es—the effect was not significant for the totalsample. This difference could reflect that we in-cluded a control for the effect of reforming unionsafter divorce, either marriages or cohabitingunions, and they did not. When Model 1 was re-estimated removing the social role variables (di-vorced, cohabited, and widowed) from the equa-


tion, the effect of years from disruption becamestatistically significant (p , .01) for the entiresample.

Our finding that persons leaving a troubledmarriage immediately tend to reduce their stresslevels is consistent with Booth and Amato’s(1991) finding, but we believe the results supporta social role rather than a crisis model. The psy-chological health of divorced persons tends to im-prove only when they move back into marriage orcohabiting relationships, that is, undergo anotherrole transition. There is no overall evidence in ourfindings that remaining divorced becomes lessstressful with time, a prediction of the stress mod-el. Improved psychological health while still di-vorced was only found for those who were rela-tively happy with their marriage predisruption.

Aseltine and Kessler (1993) argued that con-trolling for remarriage, which we do here, mayintroduce a selection bias to the extent that peoplewith psychological disorders have more difficultyremarrying, leading to higher distressed scores forthose remaining divorced. If the selection expla-nation were correct, we would not expect movingfrom divorced to remarried status to show declinesin stress scores, because this view would arguethat less distressed persons remarry. In the basicpooled-time series model, the effect of remarriageis estimated by moving from divorced to marriedstatus, which is represented by the same coeffi-cient (b for divorced or separated) as moving formmarried to divorce status (but having the oppositesign). We test if these were equal effects by in-cluding a term in the equation for remarriage(coded 1 if the married status represents a remar-riage, 0 otherwise). This effect was not statisti-cally significant. Because it measured the differ-ence between moving from married to divorcedand divorced to married, we conclude that the ef-fect of moving from divorced status to marriageis as strong as moving from married to divorcestatus. These findings cast doubt on a selectionexplanation for the lower distress levels observedfor the remarried.

If the social selection effect is primarily a func-tion of a greater divorce risk among persons withrelatively serious personality disorders who thusexit marriage quite quickly, it is possible that thissample may underestimate this effect. Persons inrelatively short-term marriages are underrepre-sented in the 1980 sample of the currently mar-ried. Although we believe this effect is likely tobe quite small, additional research with differentsamples is needed to confirm this. The findings

provide support for the explanation that the causaldirection is stronger from divorce to psychologicaldistress rather than from psychological distress todivorce but do not rule out the likelihood that en-during personality characteristics may both in-crease the risk of divorce and the susceptibility ofpersons to developing psychological distress post-divorce (Gottman, 1994). Studies including per-sonality measures predivorce would be needed toevaluate such an effect.

The analyses reported here set 5 years as a rea-sonable threshold, or outer limit, for the beginningeffects of psychological stress on a marital dis-solution process. The choice of 5 years was basedon extensive exploratory analysis using lowess re-gression smoothing techniques. To evaluate thesensitivity of this choice of thresholds, however,we repeated the analyses with shorter and longerones. The 5-year thresholds produced the best fit-ting model (measured by R2). The results did notdiffer in a substantively important way by thresh-old level used, except that with a 10-year thresh-old the social selection effect was in the same di-rection but no longer significant. We gather thatour basic conclusions of a strong social role effectand weak selection and crisis effects are robustwith respect to the choice of thresholds.

Although the high correlation between the five-item psychological stress scale we used and themore complete 17-item scale suggest that our find-ings can be generalized to other distress measures,we performed several analyses to evaluate the ro-bustness of the findings by changing the distressmeasure. In Waves 2, 3, and 4, we were able toaugment the scale we used with three of the morephysiological items from the Langner scale thatwere available for both married and divorced re-spondents in these waves. Excluding the firstwave from the analysis and using the augmentedscale produced results supporting the same con-clusions reached with the four waves and the five-item scale. The five-item measure includes a mea-sure of the recency of psychological distress andof overall happiness—components sometimesseparated in other studies (Booth & Amato, 1991).Because the measure for recency of stress asksabout distressful periods within the last 3 years,as well as recently, it is possible that the failureto find declines in distress with increased durationfrom the disruption may represent reporting onearlier events. To test for this possible bias, werepeated the analysis, dichotomizing this measureinto persons who did and did not experiencedstress recently. A random effects probit model for


pooled-time series data was estimated (Greene,1995) with this as the dichotomous dependent var-iable. Although the standard errors were higher inthis model and some of the effects that were sig-nificant in the analysis of the five-item scale wereno longer so, the key finding of a significant risepredisruption, significant role effects, and no de-cline postdisruption were replicated. Separateanalyses with the happiness item as the dependentvariable also were consistent with these findings.These tests show that our basic findings about therelative effects of the role, crisis, and social se-lection explanations were consistent across mod-ifications in the dependent variable.

A somewhat unexpected finding was that thecrisis model seemed to apply primarily to personswho were relatively high on marital happiness inthe wave immediately predisruption. In this groupthere was little evidence of a predisruption rise indistress but a substantial negative effect of the dis-ruption itself, which gradually decreased as thecrisis moved further into the past. Personality fac-tors may play a role. Because marital and personalhappiness are highly correlated, perhaps personswith a more positive outlook on life are able toadapt more quickly to the negative effects of acrisis. It also might signify that persons departinga marriage in which they were reasonably happyeither were not the initiators of the divorce or leftfor reasons other than long-term trouble. Lackinga long history of living in a troubled relationshipalso may accelerate their adaptation to stressfulevents. More research is necessary to evaluatethese possible explanations.

The pooled-time series analysis of a 12-yearstudy of persons in intact marriages in 1980, ex-amined in four waves of data, provides strongsupport for a social role explanation for higherpsychological distress levels among divorced per-sons. Most of these higher levels, compared withthose among the married, reflect the stress of liv-ing in the divorced role: Only a small part appearsto be due to selection of persons with high psy-chological distress into divorce and response tothe crisis. The psychological distress scores in thewaves in which persons were either divorced orpermanently separated were significantly higherthan those for persons who were married. Theseparated and divorced also were significantlymore distressed than the remarried or cohabitingpersons in the sample. These all support role the-ory, which predicts that living as a divorced orseparated person is generally more stressful. Al-though the purpose of this research was not to

identify the factors in divorced roles that accountfor higher stress levels, the effect clearly was notdue to changes in economic conditions or a resultof the presence of children. Other evidence sup-porting the social role explanation comes from an-alyzing the differential effect on distress by howhappy individuals were in their marriage predis-ruption. Those in the most troublesome marriages(least happy) actually improved their psychologi-cal health postdisruption, suggesting that the ef-fect of leaving a stressful role was more importantthan the crisis of the disruption.

The crisis model predicted an increase in dis-tress levels premarital disruption and a drop fol-lowing it. Overall, we only found a rise in psy-chological distress up to the divorce—no fallpostdisruption. Our findings did not support theview that role transitions and other changes oc-curring at the time of the divorce produce tem-porary stresses that dissipate after a period of ad-justment to new life conditions. The risepredisruption could be equally well explained bysocial role theory. As a marriage begins to dis-solve, living in one headed toward disruption be-comes more stressful. Living in an unhappy andconflict-filled marriage has been shown to behighly stressful (Ross, 1995), a finding we con-firmed here. Controlling for remarriage and co-habitation, there was no evidence of an overallsignificant decline in distress scores postdisrup-tion. The only support for the crisis model wasfound among respondents who left marriages inwhich they were reasonably happy in the waveimmediately preceding divorce. Among those un-happy with their marriages predivorce, psycholog-ical health appeared to decline rather than improvecommensurate with increase in time postdisrup-tion.

There was some support for a social selectioneffect, although this effect only accounted for asmall amount of the total effect of divorce status.Because we could only measure social selectionindirectly, based on the level of distress in themarriage more than 5 years before the divorce oc-curred, further research is needed before we canconclude that selection factors only have a smalleffect. Other studies that measure personality andsocial factors before the marriage occurred wouldprovide a better test. What we do know from thisstudy is that elevated distress levels of the di-vorced primarily occurred in the few years pre-ceding the divorce and were not the result of amore chronic distress condition.

Although women are more likely to have cus-


tody of the children and report more decline instandard of living following divorce than do men,there was no evidence in this study that the stressprocess with divorce differed by gender. Men andwomen may respond to stress in different ways(Horwitz et al., 1996), so research studies withmeasures of multiple stress outcomes would beneeded to further explore the gender effect.

Because we found that divorced persons’ psy-chological health improved when they formed anew relationship (either cohabiting or married),the social support and attachment benefits of anintimate living relationship (Ross, 1995) com-pared with divorced status may be the key ex-planatory factor in the higher distress levelsamong the divorced.

NOTE

This is a revised version of a paper presented at theInternational Conference on Social Stress Research inParis, France, in 1996. We would like to thank AlanBooth and Paul Amato for their comments on an earlierdraft. This study was supported in part by National In-stitute on Aging Grant R01 AG4146.

REFERENCES

Allison, P. D. (1994). Using panel data to estimate theeffects of events. Sociological Methods & Research,23, 174–199.

Amato, P., & Partridge, S. (1987). Widows and divor-cees with dependent children: Material, personal,family, and social well-being. Family Relations, 36,316–320.

Aseltine, R. H., Jr., & Kessler, R. C. (1993). Maritaldisruption and depression in a community sample.Journal of Health and Social Behavior, 34, 237–251.

Avison, W. R. (1999). Family structure and processes.In A. V. Horwitz & T. L. Scheid (Eds.), A handbookfor the study of mental health: Social contexts, the-ories, and systems (pp. 228–240). New York: Cam-bridge University Press.

Berman, W. H., & Turk, D. C. (1981). Adaptation todivorce: problems and coping strategies. Journal ofBehavioral Medicine, 4, 179–189.

Bloom, B. L., Niles, R. L., & Tatcher, A. M. (1985).Sources of marital dissatisfaction among newly sep-arated persons. Journal of Family Issues, 6, 359–373.

Booth, A., & Amato, P. (1991). Divorce and psycholog-ical stress. Journal of Health and Social Behavior,32, 396–407.

Booth, A., Amato, P. R., Johnson, D. R., & Edwards, J.N. (1993). Marital instability over the life course:Methodology report for fourth wave. Lincoln, NE:Bureau of Sociological Research.

Brown, P., Felton, B. J., Whiteman, V., & Manela, R.(1980). Attachment and distress following maritalseparation. Journal of Divorce, 3, 303–317.

Cherlin, A. J. (1981). Marriage, divorce, remarriage.Cambridge, MA: Harvard University Press.

Chiriboga, D., Roberts, J., & Stein, J. (1978). Psycho-

logical well-being during marital separation. Journalof Divorce, 2, 21–35.

Coombs, R. H. (1991). Marital status and personal well-being: A literature review. Family Relations, 40, 97–102.

Davies, L., Avison, W. R., & McAlpine, D. D. (1997).Significant life experiences and depression amongsingle and married mothers. Journal of Marriage andthe Family, 59, 294–308.

Doherty, W. J. (1983). Impact of divorce on locus ofcontrol orientation in adult women: A longitudinalstudy. Journal of Personality and Social Psychology,44, 834–840.

Doherty, W., Su, S., & Needle, R. (1989). Marital dis-ruption and psychological well-being. Journal ofFamily Issues, 10, 72–85.

Fox, J. (1991). Regression diagnostics. Newbury Park,CA: Sage.

Furstenberg, F. F., Jr., & Spanier, G. B. (1984). The riskof dissolution in remarriage: An examination of Cher-lin’s hypothesis of incomplete institutionalization.Family Relations, 33, 433–441.

Goetting, A. (1981). Divorce outcome research: Issuesand perspectives. Journal of Family Issues, 2, 350–378.

Gottman, J. (1994). What predicts divorce? The rela-tionship between marital process and martial out-comes. Hillsdale, NJ: Erlbaum.

Gove, W. R. (1972). The relationship between sex roles,marital status, and mental illness. Social Forces, 51,34–44.

Gove, W., Hughes, M., & Briggs, K. (1983). Does mar-riage have positive effects on the psychological well-being of the individual? Journal of Health and SocialBehavior, 24, 122–131.

Gove, W. R., Hughes, M., & Style, C. B. (1983). Doesmarriage have positive effects on psychological well-being of the individual? Journal of Health and SocialBehavior, 24, 122–131.

Greene, W. H. (1995). LIMDEP: Version 7.0: User’smanual and reference guide. Bellport, NY: Econo-metric Software.

Groves, R., & Kahn, R. (1979). Surveys by telephone:A national comparison with personal interviews. NewYork: Academic Press.

Hetherington, E. M., Cox, M. J., & Cox, R. (1978). Theaftermath of divorce. In J.H. Stevens, Jr., & M. Mat-thews (Eds.), Mother-child, father-child relation-ships. (pp. 149–176). Washington, DC: National As-sociation for the Education of Young Children.

Holmes, T. H., & Rahe. R. H. (1967). The social read-justment rating scale. Journal of Psychosomatic Re-search, 11, 213–218.

Horowitz, A. V., Raskin White, H., & Howell-White, S.(1996). The use of multiple outcomes in stress re-search: A case study of gender differences in re-sponse to marital dissolution. Journal of Health andSocial Behavior, 37, 278–291.

Hsiao, C. (1986). Analysis of panel data. Cambridge,U.K.: Cambridge University Press.

Johnson, D. R. (1995). Alternative methods for thequantitative analysis of panel data in family research:Pooled time-series models. Journal of Marriage andthe Family, 57, 1065–1077.

Johnson, D. R., White, L., Edwards, J., & Booth, A.(1986). Dimensions of marital quality: Towards meth-


odological and conceptual refinement. Journal ofFamily Issues, 7, 31–50.

Johnson, D., & Booth, A. (1998). Is marital quality aproduct of the dyadic environment or individual fac-tors? Panel evidence from individuals in successivemarriages. Social Forces, 76, 883–904.

Kitson, G. C., & Morgan, L. A. (1990). The multipleconsequences of divorce: A decade review. Journalof Marriage and the Family, 52, 913–924.

Kitson, G. C., & Sussman, M. B. (1962). Marital com-plaints, demographic characteristics, and symptomsof mental distress in divorce. Journal of Marriageand the Family, 44, 87–101.

Langner, T. S. (1962). A twenty-two item screeningscore of psychiatric symptoms indicating impairment.Journal of Health and Human Behavior, 3, 269–276.

Mastekaasa, A. (1992). Marriage and psychologicalwell-being: Some evidence on selection into mar-riage. Journal of Marriage and the Family, 54, 901–911.

Mastekaasa, A. (1994). The subjective well-being of thepreviously married: The importance of unmarried co-habitation and time since widowhood or divorce. So-cial Forces, 73, 665–692.

Mastekaasa, A. (1997). Marital dissolution as a stressor:Some evidence on psychological, physical, and be-havioral changes during the preseparation period.Journal of Divorce and Remarriage, 26, 155–183.

McDowell, I., & Claire, N. (1987). Measuring health:A guide to rating scales and questionnaires. NewYork: Oxford University Press.

McLanahan, S. S. (1983). Family structure and stress:A longitudinal comparison of two-parent and female-headed families. Journal of Marriage and the Family,45, 347–357.

Menaghan, E. G. (1985). Depressive affect and subse-quent divorce. Journal of Family Issues, 5, 295–306.

Menaghan, E. G., & Lieberman, S. (1986). Depressive

affect and subsequent divorce. Journal of Family Is-sues, 6, 162–172.

Pearlin, L. I. (1999). Stress and mental health: A con-ceptual overview. In A. V. Horwitz & T. L. Scheid(Eds.), A handbook for the study of mental health:Social contexts, theories, and systems (pp. 161–175).New York: Cambridge University Press.

Plummer, L. P., & Koch-Hattem, A. (1986). Familystress and adjustment to divorce. Family Relations,35, 523–529.

Robins, L. N., & D. A. Regier (Eds.). (1991). Psychi-atric disorders in America: The epidemiologicalcatchment area study. New York: Free Press.

Ross, C. E. (1995). Reconceptualizing marital status asa continuum of social attachment. Journal of Mar-riage and the Family, 57, 129–140.

StatCorp. (1997) Stata statistical software: Release 5.0.College Station, TX: Stata.

Tschann, J. M., Johnston, J. R., & Wallerstein, J. (1989).Resources, stressors, and attachment as predictors ofadult adjustment after divorce: A longitudinal study.Journal of Marriage and the Family, 51, 1033–1046.

Turner, R. J., & F. Marino (1994). Social support andsocial structure: A descriptive epidemiology. Journalof Health and Social Behavior, 35, 193–212.

Waite, L. (1995). Does marriage matter? Demography,32, 483–507.

Wertlieb, D., Budman, S., Demby, A., & Randall, M.(1984). Marital separation and health: Stress and in-tervention. Journal of Human Stress, 10, 18–26.

Wheaton, B. (1990). Life transitions, role histories, andmental health. American Sociological Review, 55,209–223.

Wheaton, B. (1999). The nature of stressors. In A. V.Horwitz & T. L. Scheid (Eds.). A handbook for thestudy of mental health: Social contexts, theories, andsystems (pp. 176–197). New York: Cambridge Uni-versity Press.

Documents

An Empirical Test of Crisis, Social Selection, and Role Explanations of the Relationship Between Marital Disruption and Psychological Distress: A Pooled Time-Series Analysis of Four-Wave