10
PREDICTING DIVORCE AT MARITAL THERAPY INTAKE: A PRELIMINARY MODEL D. RUSSELL CRrZlVE. JEAN N. SODERQUIST, and RICHARD L. FRANK The purpose qf this study was to predict divorce in a marital-distressed and the~apy-seeking population. The sample was from the case records of the Marriage nnd Famlly Therapy Clinic at Brigham Yoting Univer- sity. Demogrnphic and psychological data as well as Marital Status Inventory (.MSI) and Marital Adjustment Test (MAT) scores were available for analysis. Statistical procedures successfully predicted mnrital status in a high percentage ofcases. The most importnntfind- ings were that wlves' variables were more important in divorce predic- tion than were htisbands'. In addition, marital quality was not found to be predictive of marital status. In recent years, a considerable amount of literature has been generated to identify variables that predispose individuals and couples to divorce (Larson & Holman, in press). Most studies to date have relied upon data sets from large national surveys such as the US Census, the General Social Surveys conducted by the National Opinion Research Center (NORC), and the Survey of Economic Opportunitv compiled by the United States Bureau of the Census. A few researchers have collected study-specific data on a number of divorce-related topics through the use of live observation and interviews, telephone surveys, or return mail questionnaires with, for the most part, relatively limited numbers of respondents. For example, Belsky, Spanier. and Rovine (1983) observed and interviewed 72 expectant volunteer cou- ples periodically over one year, begnning with the last trimester of preg- nancy, to investigate changes in the marital relationship across a transition Support for this article was prov~ded by a grant from the College ot Famtly, Home and Social Sciences of Brigham Young University. D. Russell Crane. Ph.D., is a Professor and Director of the Marriage and Family Therapy Graduate Programs in the Department of Family Sciences. 274 TLRB. Brigham Young University, Pmvo, UT 84602. Send reprint requests to Dr. Crane. Jean N. Soderquist, Ph.D., is a Marriage and Family Therapist in private practice. Salt Lake City, UT 84119. Richard L. Frank. Ph.D., is a Marriage and Family Therapist in private practice, Mount Vernon, WA 98273. The American Journal of Family Therapy, Vol. 23, No. 3, Fall 1995 Q Brunntrhfazel. Inc.

PREDICTING DIVORCE AT MARITAL THERAPY … Therapy Research/1995...PREDICTING DIVORCE AT MARITAL THERAPY INTAKE: A PRELIMINARY MODEL ... of the Marriage nnd Famlly Therapy Clinic at

Embed Size (px)

Citation preview

PREDICTING DIVORCE AT MARITAL THERAPY INTAKE: A PRELIMINARY MODEL

D. RUSSELL CRrZlVE. JEAN N. SODERQUIST, and RICHARD L. FRANK

The purpose qf this study was to predict divorce in a marital-distressed and the~apy-seeking population. The sample was from the case records of the Marriage nnd Famlly Therapy Clinic at Brigham Yoting Univer- sity. Demogrnphic and psychological data as well as Marital Status Inventory (.MSI) and Marital Adjustment Test ( M A T ) scores were available for analysis. Statistical procedures successfully predicted mnrital status i n a high percentage ofcases. The most importnntfind- ings were that wlves' variables were more important in divorce predic- tion than were htisbands'. In addition, marital quality was not found to be predictive of marital status.

In recent years, a considerable amount of literature has been generated to identify variables that predispose individuals and couples to divorce (Larson & Holman, in press). Most studies to date have relied upon data sets from large national surveys such as the US Census, the General Social Surveys conducted by the National Opinion Research Center (NORC), and the Survey of Economic Opportunitv compiled by the United States Bureau of the Census.

A few researchers have collected study-specific data on a number of divorce-related topics through the use of live observation and interviews, telephone surveys, or return mail questionnaires with, for the most part, relatively limited numbers of respondents. For example, Belsky, Spanier. and Rovine (1983) observed and interviewed 72 expectant volunteer cou- ples periodically over one year, begnning with the last trimester of preg- nancy, to investigate changes in the marital relationship across a transition

Support for this article was prov~ded by a grant from the College ot Famtly, Home and Social Sciences of Brigham Young University. D. Russell Crane. Ph.D., is a Professor and Director of the Marriage and Family Therapy Graduate Programs in the Department of Family Sciences. 274 TLRB. Brigham Young University, Pmvo, UT 84602. Send reprint requests to Dr. Crane. Jean N. Soderquist, Ph.D., is a Marriage and Family Therapist in private practice. Salt Lake City, UT 84119. Richard L. Frank. Ph.D., is a Marriage and Family Therapist in private practice, Mount Vernon, WA 98273.

The American Journal of Family Therapy, Vol. 23, No. 3, Fall 1995 Q Brunntrhfazel. Inc.

228 The Amencan Journai of Famiiy Therapy. Vol. 23. No. 3, Fall 1995

to parenthood. Larsen and Olson (1989) did a three-vear follow-up of 179 couples ~ v h o had taken a premarital i n v e n t o ~ (PREPARE) in order to demonstrate the abilitv of this instrument to predict marital satisfaction or divorce over time. ihough reiatively few in number, such studies pro- vide a sigruficant and varied source of information on factors relating to marital instability.

As important as the these types of investigations are, thev nevertheless have several significant limitations. To begin with, many of the surveys are cross-sectional rather than longitudinal. Among the problems associ- ated with using cross-sectional data to determine correlates of marital dissolution, one is that persons who divorce and quickly remarry are represented in such surveys as married. while those who are slower to remarry are overrepresented in the divorced groupings (Moore & Waite, 1981). The subjects overrepresented are usually women, because divorced men generally marrv more quicklv and are more difficult to locate (Kit- son & Raschke, l98i). In addition,'such surveys (because thev yield what amounts to a "snapshot" of current marital status) focus primarily on factors, such as educational level, that are measured at the time of the survey even though marital dissolution may have occurred much earlier (Kitson & Raschke, 1981). Thus, the temporal order of the variables is the opposite of what it should be for inferring the effect of such factors upon marital outcomes (Glenn & Supancic, 1984).

Second, nearly all of the studies to date have been based upon data collected from only female respondents (Schumm et al., 1989). Few have used information from men or couples. Since husbands and wives may not be affected equallv by the same variables and may be likely to view their marriages differently (Crane et al., 1995; Gottman, 1994; White, 1989), studies using only women are apt to present an incomplete picture.

Valuable information regarding the prediction of marital satisfaction has come from some recent researchers, a few of whom have designed longitudinal studies that included couples. Such studies have found differ- ences in marital satisfaction by comparing a couple's physiological re- sponses to stress and conflict. They have used observational coding of couples engaged in resolving high-conflict activities to yleld interesting data on future marital status and divorce outcome (Gottman & Levenson. 1985; Krokoff et al., 1989). These technology-intensive types of research projects are important in uncovering the complexities involved in the pre- diction of marital satisfaction and divorce potential. However, the re- sources. time. and skills required to make physiological measurements; to observe, record, and code interactions; or to follow and measure fami- lies for several years are not often available to researchers or clinicians. Also, distressed, therapy-seeking couples are not always willing or able to complete lengthy assessments or to participate in interactional activities. Therefore, it would seem pragmatic to establish a relatively simple pre- dictive model that could easily be used by clinicians until more compre- hensive assessment procedures are developed through continued research.

Predichng Divorce

Finallv, none of the studies on marital dissolution have dealt solely with a clinically distressed, treahnent-seeking population per se. Factors that may be useful in predicting divorce in the general population or in labora- tory research participants may or mav not be useful in making the same kind of prediction in clinicallv distressed couples in marital therapy.

The purpose of this study .was to begin to identify the characteristics that distinguished couples who remained married from couples who di- vorced and to predict divorce in a clinical sample. Given an accurate predictive model. clinicians would be able to identify which couples are likely to divorce and which are not. They would also have greater ability to (a) apply appropriate treatment to couples in both categories. and @) design treatments for those at higher risk for divorce.

METHOD

Subjects

The subjects in this study were 235 couples seeking therapy from the Brigham Young University Marriage and Family Therapy Clinic. All cou- ples completed an assessment packet prior to beginning therapy.

Assessment Instruments

The Marital Adjustment Test (MAT) assesses overall marital quality (Locke & Wallace, 1959). The instrument is internally consistent, and it has been demonstrated to accuratelv differentiate between distressed and nondistressed marital relationships'(Crane et al., 1990).

The Marital Status Inventory (MSI) (Weiss & Cerreto, 1980) is a 14-item hue I false test designed to measure divorce potential. The MSI has been proven to be a reliable and valid instrument. Crane and Mead (1980) and Crane. Newfield, and Armstrong (1984) calculated Spearmen-Brown split- half reliabilities ior the MSI at .86 and 37, respectively. In addition, both studies confirmed the MSI's discriminant validity.

Other Variables

Variables available from the clinic records included age at marriage. psychological distress, number of cMdren, number of preschool children, remarriage, number of stepchildren. religion (all were members of The Church of Jesus Christ of Latter-dav Saints [LDSI), wife's employment, education. and prior therapy experi;?nce. These variables represent most of those that have most often been empirically linked to divorce in nonclin- ical populations (Larson & Holman, in press).

Procedure

Using closed files from the clinic records. 235 couples were randomly chosen. Couples who divorced during treatment were identified from case

The Amencan Ioumal of Family Therapy, Lo1 23, No. 5. Fall 1995

notes. hvest~gators also used ~ u b l i c records from local district courts to identify those who had been granted divorces. Some coupies who had left the area were identiiied and traced through student alumni sources. Couples who could not be located were eliminated from rhe study.

RESLZTS

Reduction of Sample

Sixtyfour of the initial sam?le of 235 couples were found to have di- vorced. while 171 remained married. Equal groups of married and di- vorced husbands and wives were created by randomlv sampling the married couples to create a group equal in size to the group of divorced spouses. An additional random sample of 64 couples was set a s~de for a cross-validation analvsis.

Units or' Anaivs;~

Units of analvsis were husbands, wives, and couples. Previous investi- gations show that spouses are not equallv affected bv the same variables. Also, no acceptable techruque for combi4ng husband and wife scores has yet been devised (Booth et al., 1983). "His" and "her" global evaluations of marital interaction mav be verv different and, if averaged, rnav simply cancel each other out. ~hereiore, single-score comparisons provide the best information possible at this time.

Nevertheless. two sets of couples models, based on averaged and sepa- rate data, were created along with separate models for both husbands and wives. Including both types of couples models was done to determine if a combination of variables was more important than either husbands or wives variables alone.

This study used four procedures to create the prediction models: 1) multiple regression using maximum R2 (MAXR); 2 ) stepwise discriminant analysis (SDA); 3 ) descriptive discriminant analysis (DDA); and 4) pre- dictive discriminant analysis (PDA). These analyses were used to create the four models. one for husbands, one for wives, one for couples using averaged data, and one for couples using separate data.

Maximum R-square Regression

The MAXR was used to reduce the number of variables while assuring that the amount of variance accounted for was maximized (SAS User's Guide: Statistics, 1985). requires no sigruficance levels for factor entry into or removal from a given model. Selection of variables is based exclusively upon the size of the R' generated by the overall model. Thus, the "best" model of a given number of variables is "best" in the sense

Predictins Divorce

TABLE 1 Variables Selected by Stepwise Discriminant .-\nalvsis

Model Variable Partial R' F prob > i W i W Lambda

Husbands MSI 14 19.7 .0001 .87 Remarriage .04 4.8 0298 .83 Number oi Children 03 3.3 ,0713 81

Total Wives MSI Age at Marriage Prior Therapy

Total Couples (Averaged Data)

Total .\IS1 Score Age at Marriage Total Prior Therapy

Total Couples separate Data)

Wife's MSI .22 34.6 .I001 78 Wife's Age at Mamage 03 4.4 ,0383 76 Husband's MSI .04 5.2 .024i .73

Total .29

that it generates the highest RZ for any model with the same number of variables. In this case the "best" seven-variable models were selected. While eliminating some factors, these models accounted for nearly as much variance as did the corresponding complete models.

Stepwise Discriminant A~zaiysis

The 'best" seven-variable models were entered into SDA for further elimination of variables. Stepwise selection stopped \f.ith three variables in each model. These are listed, as ordered for each model, in Table 1. Total variance in divorce outcome explained for each model were 21% for husbands, 27% for wives, 28% for couples' averaged data and 29% for couples' separate data.

Descriptive Disciminant Analysis

The variables selected and ordered by SDA for each model were entered into DDA to describe the differences between married and divorced hus- bands and wives. Table 2 shows raw, standardized, and total structure coefficients for each variable in the models. In addition to the coefficients noted for each variable, DDA also produced class means for every variable in each of the disciminant functions created. Mahalanobis @ represents the differences between the mean vector of the divorced group and the mean vector of the mamed group, or the difference in group centroids, for each model. Mahalanobis D2 may be converted to an F-test in order

12 The Amencan loumal of Fanuly Therap". \-ol 3. No. 3. Fall 1995

TABLE ? Shucture Coefhdents and Class Means tor Descnphve Discnrmnant Models

Model Vanable Raw 5tandar- Total Means Mahdanob~s dlred Structure .Married D~r'orced D' F

Husbands MSI Remarnage Length of Marriage

Wives MSI Age at Marriage Prior Therapy

Couples (Averaged Data) Total hfSI Score Age at Marnage Total Prior Therapv

Couples (Separate Datal Wife's MSI Score Wife's Age at Marriage Husband's MSI Score

to determine the significance of differences between groups as is shown in Table 2.

Substantial differences between the mamed and divorced groups were found for MSI scores for all models. In addition, the husbands model showed the divorced group to have been married previousiv and to have been married a shorter time than their married counterparts. The wives model showed that in addition to having higher MSI scores, the wives who divorced were married at an older age than their mamed counter- parts. The couples models were characterized bv higher wives' MSI scores. wives' older age at marriage, and the husbands' MSI scores.

Predictioe Discriminant Anniusis

The variables used in DDA were then entered into FDA to determine if the models were able to classify subjects as to future marital status. For each model, classification was based on the within-group covariance matrices since the covariance matrices for husbands and wives were non- homogeneous.

Chi-square was used to test whether the classification was better than chance alone. In that analysis. the PDA functions for husbands and wives proved superior to chance alone in classifying spouses as to their future marital status. Table 3 indicates the percentages of subjects correctly and incorrectly classified by PDA.

Comparison with Cross-Vnlidation Snmpie

The cross-validation sample of couples, set aside earlier, was combined with the divorced sample already tested to check the accuracy of the

Predicting Divorce

TABLE 3 Percentazes of Subiects Correctlv Classified Divorced and Manied

Model Building Cross-Validation Samples Sample

Divorced Marned Married Husbands (N=M) 68.75- 70.31'" 76.56 Wives IN=&) 67.19.. 78.13' 79.69 Couples (Averaged Data) fN=128) 60.94 85.94"' 81.25 Couples (Separate Datal(N=128) 67.19" 78.13' 81.25

model in predicting future marital status. As indicated in Table 3, the models for husbands, wives, and couples (separate data) were actually better at predicting a continuation of marriage among spouses in the cross- validation sampie than they were in the original sample. Thus, the accu- racy of the predictive discriminant models in correctly classifying married subjects was confirmed.

DISCCSSION

Comparison with Nonciinicni Studies

Several of the factors frequently assodated with divorce in the nonclini- cal literature proved to be significant in this study. These included MSI scores, remarriage, length of mamage, age at mamage, revious therapy, and number of children. Overall, then it would seem t FI at there is some agreement between t h s clinical studv and other nonclinicai studies. How- ever, it is important to note that while other studies associated younger wives' age at marriage with divorce, in this study the direction was op 0

variable. 1. - site. It may be that this particular sample is somewhat unique on t a

One notable exception to the findings of previous research is the absence of marital quality as a predictor of divorce in the present study. Research- ers (Spanier & Lewis. 1980) have long posited a stron and direct relation- ship between marital qualitv and marital stability. file this may be true for the general population, the present study showed no such relationship. Several factors may account for this.

First, most of the studies linking marital qualitv with marital instabilit). have relied u on data collected retrospectively (~lbrecht & Kunz, 1980). K Spouses who ave divorced may tend to view their revious marital rela- tionship with a different eye than while married. T 1 ey may also simply lump all concerns about specific conflicts or problem areas into the single category of "marital quality." In this investigation, data were collected prior to the beginning of treatment while all marriages were intact.

Second, as mentioned earlier, previous studies have relied on data from only one spouse, most frequently the wife, to describe the quality of a

W The Amencan Journal of Family Therapy, Vol. 23. No. 3, Fall 1995

given relationship. Spouses often disagree about the quality ot their mar- riages (Larson dr Holman. in press). Data in the present studv were col- lected from both spouses.

Finally, there may well be a basement effect operating on this variable since couples entering treatment had low marital quality to begin with. The MAT then mav not be able to detect the small differences in marital quality between G o groups who both had low LMAT scores.

Individual versus Couples Models

Given the results of this study, it would appear that a couples model may be more useful in understanding divorce than the individual models of husbands and wives alone. The couples models accounted for more of the variance in divorce outcome than did either individual model. In addition, the couples model that was most important was the couples separate data model. It may be that the best picture of the divorce process is in the individual contributions of both partners, but studied as mutually duent ia l variables.

Wives' versus Husbands' Distress Levels

In considering the results of the couples models, one is drawn to the conclusion that the distress level of the wife is a more important predictor of divorce than is the distress level of her husband. This is consistent with previous research (Crane et al., 1984) that showed that the wives' distress level was the most important characteristic of divorced couples. Also. Gottman (1994), in an eight-year longitudinal study, found that only the wife's consideration of marital dissolution (based on MSI scores) predicted actual separation. Since many women know the costs of divorce in such terms as loss of social and economic status. increased responsibility for children, and custody disputes (Holden & Smock, 1991), their willingness to consider divorce should not be taken lightlv.

Divorce Prediction

The models created by FDA were all able to "predict," in an ex post facto manner, marital status with a significant level of accuracy. A chi- square analysis of each of the models demonstrated that, in every case, PDA was significantly more accurate than chance at assigning marital status. However, because most of the models were tested against the same set of data used to create them, there may have been an upward bias in their reported accuracy. Ideally, the models should be tested for accuracy against a completely new sample of married and divorced subjects.

When tested using the cross-validation sample, the models produced by PDA were as good or better at predicting a continuation of marriage among subjects than they were in the original sample. This appears to confirm the overall accuracy of the models in predicting future marital

Predicting Divorce

status for those rvho will remain mamed and argues logically for the accuracy of the models for predicting those who will divorce.

Conclusion and Future Directtons

Wives' distress as measured by Marital Status Inventor). scores are probably the most important variable identified in this study. It was the most influential factor identified on a consistent basis. Formnatelv, it is a variable that is readily accessible by most clinicians in the first few sessions of therapy.

Although the results of this investigation are interesting, the models accounted for relatively small amounts of the total variance in divorce outcome (21% to 29%). One reason mav be that the variables empirically linked to divorce in the nonclinical population are not entirely adequate for study in these investigations. More comprehensive intake procedures aimed at collecting information such as premarital pregnanq, education. etc. would be beneficial for future research. Data on some of these other variables might prove effective in enhancing the models already created. It should be noted, however, that in previous research, questionnaire mea- sures have typically accounted for only 10% of the variance in marital satisfaction, while observational studies have usually accounted for ap- proximately 25% of the variance (Levenson & Gottman, 1983). While it would be preferable to account for as much of the variance as possible, the results of this studv are well within the currentlv acceptable limits.

An additional limitation of this study mi ht be &e fact that all of the participants were members of the LDS chur 31 . While results were similar to findings in other studies. there was a difference in this group on the variable of wives' age at marriage. Perhaps that factor is unique to this population. Using this sample, however, &d point out an interesting pos- sibility. Since religious homogeneitv was previously found to be correlated to divorce (Chan & Heaton, 1989; orl lev &Woods, 1991), using a sample from a single religious denomination while achieving results similar to other studies may identify variables that are more important than reli- gion alone.

If the models created in this study were validated by further research, the results might be used to enhance the quality and effectiveness of mar- tial therapy. For example, therapists noting MSI scores and other variables that have been shown to be predictive of divorce may wish to discuss the likelihood of divorce with the clients and not assume that a continuation of the marriage is the goal. Also, knowing more about couples who even- tually may divorce might enable clinicians to develop, test, and implement more effective treatment aimed at preventing marital dissolution.

REFERENCES

Albrecht. S. L.. k K w . P. R. (1980). The decision to divorce: A social exchange perspective. loumnl of Divorce. 3, 319-337.

Bekky. I., Spanier. G. B., 6r Rovine. M. (1983). Stability and change in marriage across the transition to parenthood. Journal @Marriage and tke Family, 45, 567-577.

236 The Amencan Journal or Famly Therapy, Vol. 23. No. 3. Fall 1995

Booth. A,. Johnson. D.. & Edwards, J N. (1983). bleasunng marital instability lournal qf Llarrrage and the Famzly, 45, 387-394.

Chan. L.Y.. & Heaton. T. B. (1989). Demographic determinants of delaved divorce. iournai o f Diuorce. 13. 97-112.

Cor le . C. 1.. & Woods. A. Y. (1991). Socioecononuc. sociodemographic and attitudinal correlates oi the temoo of divorce. lournal of Diz~orce ond Remarnape. 16. 47-68.

~ ~ -~ ~ ~~~ ~ ~

CrXIe. D. R.. Allgood. S. h., Larson. J. H., & Griffin. W. (1990). Assessing marital quality with distressed and nondistressed couoles: A comoarison and eauivalencv table for three irequentlv used measures. [ourna; qf ,+larnaq;and the FamiiG 52. 57-93.

Crane. D. R., & Mead. D. E. (1980). The manta1 status inventory: Some prelimina~y data on an mtrument to measure marital dissolution potential. Amerrcan lournal o f Famriy T l r l m v . d . 31-35 r . . - . - -

Crane. D. R.. Newfield. N.. k Armstrong, D. (1984). Predicting divorce at marital therapy intake: Wives' distress and the manta1 status inventory. journal of Marrtal and Famliy Tkrnpy. 10. 305-312.

Crane. D. R.. Soderquist, I. N.. & Gardner, M. D. (1995). Gender differences in the c o p t i v e and behav~oral steps towards divorce. r\mmcan iournai o f Famiiy Tlrerapy, 23. 99-105.

Glenn. N. D.. & Suoanoc. M . (1984). Soc~oloprcal and demoeraohic correlates ot divorce and separation'm the United States: i\n u<date and reconsidbration. lournai qf Marriage and the Family, 46. 563476.

G o m a n J. .M. (1994). What predicts divorce? Hillsdale. NJ: Erlbaum. Gottman I. M.. k Levenson. R. W. (1985). A valid procedure for obtaining self-report of

affect in marital interaction, lournai of Consuitrng and Clinicnl Psychology, 2. 151-I@. Holden. K. C.. & Smock. P. J. (1991). The economic costs of marital dissolution: Why do

women bear a disproportionate cost? Annual Recinu of Sociology. 17, 51-78. Krokoft L. J.. Gottman J. M.. k Hass. S. D. (1989). Validation oi a global rapid couples

interamon scoring system. Behavrorai Assessment. 11. 65-79. Kitson. G. C.. k Raschke. H. J. (1981). Divorce research: What we know; what we need to

know. journal of Diuorce. 4(3), 1-37. L m n . A. 5.. k Olson. D. H. (1989). Predicting manta1 satisfaction using PREPARE: A

replication study. lournaiof Marztal and Family Therapy, 15, 311-322. Larson. J. H.. k Holman. T. 8. (in press). Premarital predictors of marital quality and

stability: iin applied literature review. Fumrly Rziatons. Levenson R. W.. k Gottman. J. M. (1983). Marital interaction: Phvsiological linkage and

affective exchange, lournal of Personairty and Social Psycholo-9. 45. 587-597. Locke. H. I.. & Wallace. K. M. (1959). Short marital-adjustment and prediction tests: Their

reliability and validity Marrirrge and Famrlu Licinq. 21. 25-255. Moore. K. A.. k Waite. L. 1. (1981). Marital dissolution, earlv motherhood and early rnar-

riage. Socral Forces. 60, 2040. SAS user's gurde: Statistrcs. (1985). C a y NC: SAS Institute. Schumm. W.R.. Obiorah. F. C., k Silliman, 8. (1989). Marital quality as a function of

conservative religous identification in a sample of Protestant and Catholic wives from the midwest. Psychological Rrports. 64, 124-126.

Spanier. G. 8.. k Lewis, R. A. (1980). Marital quality: A review of the seventies, lournni of Marrrage and the Famriy, 42 , 825-839.

Weiss. R. L. & Cerreto. M. (1980). The marital status mventory: Development of a measure of dissolution potential. Amnrcan lournal qf Famliy Therapy. 8, 8046.

White. B. 11989). Gender differences in marital communication patterns. Family Process. 28.89-106.