Prediction of Response to Treatment in a Randomized Clinical Trial of Couple Therapy: A 2-Year Follow-Up

Abstract

Many studies have examined pretreatment predictors of immediate posttreatment outcome, but few studies have examined prediction of long-term treatment response to couple therapies. Four groups of predictors (demographic, intrapersonal, communication, and other interpersonal) and 2 moderators (pretreatment severity and type of therapy) were explored as predictors of clinically significant change measured 2 years after treatment termination. Results demonstrated that power processes and expressed emotional arousal were the strongest predictors of 2-year response to treatment. Moderation analyses showed that these variables predicted differential treatment response to traditional versus integrative behavioral couple therapy and that more variables predicted 2-year response for couples who were less distressed when beginning treatment. Findings are discussed with regard to existing work on prediction of treatment response, and directions for further study are offered.

Full text

Prediction of Response to Treatment in a Randomized Clinical Trial of
Couple Therapy: A 2-Year Follow-Up
Brian R. Baucom
University of California, Los Angeles
David C. Atkins
Fuller Graduate School of Psychology
Lorelei E. Simpson
Southern Methodist University
Andrew Christensen
University of California, Los Angeles
Many studies have examined pretreatment predictors of immediate posttreatment outcome, but few
studies have examined prediction of long-term treatment response to couple therapies. Four groups of
predictors (demographic, intrapersonal, communication, and other interpersonal) and 2 moderators
(pretreatment severity and type of therapy) were explored as predictors of clinically significant change
measured 2 years after treatment termination. Results demonstrated that power processes and expressed
emotional arousal were the strongest predictors of 2-year response to treatment. Moderation analyses
showed that these variables predicted differential treatment response to traditional versus integrative
behavioral couple therapy and that more variables predicted 2-year response for couples who were less
distressed when beginning treatment. Findings are discussed with regard to existing work on prediction
of treatment response, and directions for further study are offered.
Keywords: couple therapy, long-term treatment response, arousal, power
Supplemental materials: http://dx.doi.org/10.1037/a0014405.supp
Substantial empirical research has documented the effectiveness
of behavioral couple therapies for improving couple functioning
over the course of treatment (Baucom, Shoham, Mueser, Daiuto, &
Stickle, 1998; Christensen et al., 2004; Snyder, Castellani, &
Whisman, 2006). Although some studies have found significant
declines in treatment gains over time after treatment termination
(Jacobson, Schmaling, & Holtzworth-Munroe, 1987; Snyder,
Wills, & Grady-Fletcher, 1991), recent work (Christensen, Atkins,
Yi, Baucom, & George, 2006) has documented the ability of
integrative behavioral couple therapy (IBCT; Jacobson & Chris-
tensen, 1998) and traditional behavioral couple therapy (TBCT;
Jacobson & Margolin, 1979) to create improvement in couple
functioning over the course of therapy that is largely maintained
over a 2-year period after treatment termination. However, little
research has examined what might predict successful 2-year out-
comes in couple therapy. It is important to know not only what
predicts success at the end of treatment (e.g., Atkins et al., 2005)
but also what predicts 2-year, posttherapy adjustment of couples.
We addressed this issue in the current study by examining predic-
tors of treatment response 2 years after treatment termination.
There has been a large number of short-term randomized clin-
ical trials of couple therapies, but to date only one study has
examined predictors of treatment response 2 years or longer after
treatment termination. Snyder, Mangrum, and Wills (1993) found
that couples were more likely to be intact but maritally distressed
or divorced 4 years after treatment termination if (a) they reported
higher levels of depressive symptomatology, lower psychological
resilience, lower emotional responsiveness, poorer problem-
solving skills, or higher levels of marital distress or (b) neither
spouse was employed at a semiskilled or higher level position prior
to beginning insight-oriented marital therapy (Snyder & Wills,
1989) or behavioral marital therapy (as cited in Snyder & Wills,
1989).
Most studies that have examined predictors of treatment re-
sponse have focused on prediction of couple functioning at treat-
ment termination, largely because most empirical studies of re-
sponse to couple therapy have not assessed posttermination
outcome at intervals of 2 years or more. Atkins et al. (2005)
synthesized and organized existing work on pretreatment predic-
tors of response to therapy at termination into three categories—
Brian R. Baucom, Department of Psychology, University of California,
Los Angeles; David C. Atkins, Travis Research Institute, Fuller Graduate
School of Psychology; Lorelei E. Simpson, Department of Psychology,
Southern Methodist University; Andrew Christensen, Department of Psy-
chology, University of California, Los Angeles.
David C. Atkins is now at the Department of Psychiatry and Behavioral
Science, University of Washington. Additional materials are available on
the Web at http://dx.doi.org/10.1037/a0014405
This research was supported by National Institute of Mental Health
Grant MH56223, awarded to Andrew Christensen at UCLA, and National
Institute of Mental Health Grant MH56165, awarded to Neil S. Jacobson at
the University of Washington. A methodological supplement was awarded
to Andrew Christensen and David C. Atkins. After the death of Neil S.
Jacobsen in 1999, William George served as principal investigator at the
University of Washington. We are grateful for the enormous contributions
that Neil S. Jacobson made to this research.
Correspondence concerning this article should be addressed to Andrew
Christensen, who is now at the Department of Psychology, University of
Southern California, Los Angeles, Box 156304, Los Angeles, CA 90095-
1563. E-mail: christensen@psych.ucla.edu
Journal of Consulting and Clinical Psychology © 2009 American Psychological Association
2009, Vol. 77, No. 1, 160–173 0022-006X/09/$12.00 DOI: 10.1037/a0014405
160

demographic variables, interpersonal variables, and intrapersonal
variables—that were used to predict treatment response to IBCT
and TBCT over the course of therapy.
1
TBCT couples (relative to
IBCT couples), men (relative to women), and couples who had
been married for a longer period of time (relative to couples who
had been married for a shorter period of time) all experienced
greater gains in relationship satisfaction early in treatment, but
their rate of improvement slowed more rapidly later in therapy
than did that of their counterparts. Severely distressed couples
(relative to moderately distressed couples) experienced greater
deceleration in their rate of improvement of relationship satisfac-
tion toward the end of their course of therapy. Very sexually
dissatisfied IBCT couples improved more slowly in the beginning
of treatment but continued to improve over the course of therapy,
whereas very sexually dissatisfied TBCT couples improved rap-
idly in the beginning of therapy and then began to lose some of
their early improvement toward the end of therapy. Our primary
aim in the current study was to examine whether the predictors
used in Atkins et al. (2005), with some important additions, were
associated with treatment response in the same set of couples 2
years after treatment termination.
Given the paucity of studies that have examined pretreatment
predictors of long-term posttermination treatment response, we
used all of the pretreatment variables that were used to predict
treatment response at termination in Atkins et al. (2005) but added
some pretreatment variables (parental divorce, power, demand/
withdraw, and encoded arousal) to ensure that a wide range of
indices of couple functioning were included. The first of these
variables, parental divorce, is clearly related to later relationship
functioning. Empirical findings suggest that parental divorce may
shape children’s relationship schema, so children view relation-
ships as arrangements that are fragile, temporary, and best resolved
by dissolution when significant conflict occurs (e.g., Amato, 1996;
Amato & DeBoer, 2001). Though this mechanism of intergenera-
tional transmission of divorce is relevant to therapeutic interven-
tion, it has not been considered in previous research on couple
therapy outcomes.
Power bases, or desired resources that spouses bring to or hold
within a relationship (e.g., age, education, income; Cromwell &
Olson, 1975), have been linked to treatment response in couple
therapy. For example, Freeman, Leavens, and McCulloch (1969)
found that a greater absolute discrepancy in education was asso-
ciated with significantly greater improvement in a study of general
marital counseling. An array of power base indices (difference in
age, education, and income) was included in the current study.
Power processes, or methods that partners use to exert power
within the context of an interaction (Cromwell & Olson, 1975),
have been empirically linked to response to couple therapy.
Spouses who are more similar in the degree to which they deter-
mined the content of an interaction, either by setting the topic and
doing all of the talking or by refusing to talk about or changing the
topic, have been found to respond more favorably to behavioral
marital therapy at treatment termination and at a 6-month
follow-up (Whisman & Jacobson, 1990). The current study con-
ceptualized power processes in terms of influence tactics, defined
as the amount of freedom spouses leave each other to respond to
a request for change. Couples who allowed large amounts of
freedom were referred to as using soft influence tactics, whereas
couples who allowed minimal amounts of freedom were referred
to as using hard influence tactics (Kipnis, Schmidt, & Wilkinson,
1980).
The demand/withdraw interaction pattern occurs when one part-
ner requests change by nagging, criticizing or complaining and the
other partner avoids the request or withdraws from the discussion
in response (Christensen, 1988). Higher levels of demand/
withdraw have been consistently linked to lower levels of relation-
ship satisfaction (Eldridge & Christensen, 2002) and have been
linked to response to couple therapy, though in conflicting direc-
tions. Studies have found demand/withdraw to be associated with
negative outcomes (Shoham, Rohrbaugh, Stickle, & Jacob, 1998)
and with positive outcomes (Whisman & Jacobson, 1990). The
current study examined wife demand/husband withdraw and hus-
band demand/wife withdraw in an attempt to resolve these con-
flicting findings.
Pretreatment emotional arousal during problem-solving discus-
sions has yet to be studied in terms of response to couple therapies.
However, results from several converging lines of research suggest
that it may be an important avenue for investigation. Numerous
studies have documented associations from problematic commu-
nication and relationship distress to poorer therapy outcomes (e.g.,
Atkins et al., 2005; Shoham et al., 1998). Both higher levels of
problematic communication and higher levels of relationship dis-
tress are known to be associated with higher levels of physiolog-
ical arousal (see Kiecolt-Glaser & Newton, 2001, for review).
Finally, higher levels of physiological arousal have been shown to
be stronger predictors of long-term relationship functioning than
are interpersonal and intrapersonal variables (Gottman & Leven-
son, 1992).
With procedures consistent with those used by Atkins et al.
(2005), the current study examined whether four categories of
variables predicted response to two behaviorally based couple
therapies 2 years after treatment termination. The categories of
predictors included demographic variables (age, education, in-
come, parental marital status, length of marriage, and whether the
couple has children or not); intrapersonal variables (overall mental
health, presence or absence of diagnoses from the Diagnostic and
Statistical Manual of Mental Disorders [4th ed.; DSM–IV; Amer-
ican Psychiatric Association, 1994], neuroticism, and family his-
tory of distress); communication (demand/withdraw, affective
communication, constructive communication, encoded arousal,
and power processes); and other interpersonal variables (commit-
ment, influence in decision making, desired closeness, sexual
satisfaction, and power bases). Our primary aim in the current
study was to determine significant predictors of treatment outcome
within each of the four categories of variables. Additionally, the
current study explored whether the predictive ability of any of
1
Both IBCT and TBCT are behaviorally based couples therapy; how-
ever, there are distinct differences in the primary foci and interventions of
each therapy. TBCT uses behavioral exchange and communication skills
training to resolve conflict and increase relationship quality. In contrast,
IBCT employs a number of acceptance-based techniques to help couples
better understand and cope with their differences and to increase empathy
and support between partners. IBCT includes communication skills train-
ing similar to that used in TBCT but does not emphasize this training as
much. Interested readers are referred to Jacobson and Christensen (1998)
and Jacobson and Margolin (1979) for more detailed information about
IBCT and TBCT, respectively.
161
PREDICTION OF RESPONSE TO TREATMENT

these variables differs across gender, treatment condition, and
initial distress level.
Method
Participants
Participants (N⫽130 couples) were a subsample of 134 sig-
nificantly and stably distressed married, heterosexual couples who
were recruited for participation in a two-site study of couple
therapy conducted at the University of California at Los Angeles
and the University of Washington. Participants were required to be
legally married, to be living together, and to be seriously and
consistently distressed (i.e., to have tested in the significantly
distressed range on three separate measures of marital distress
completed during three consecutive screening assessments con-
ducted prior to participation in this study). Participants were not
allowed to receive any other forms of psychotherapy while they
were in the active treatment phase of this study, and individuals
who were taking psychotropic medications were required to be at
a stable dosage prior to beginning participation in this study. All
participants were given a diagnostic psychiatric interview (the
Structured Clinical Interviews for DSM–IV for Axis I and II;
SCID; First, Spitzer, Gibbon, & Williams, 1994). Participants who
met diagnostic criteria for current Axis I disorders of substance
abuse or dependence, schizophrenia, bipolar disorder, or current
Axis II disorders of borderline, schizotypal, or antisocial person-
ality disorder were excluded from the study. Finally, couples
characterized by moderate-to-severe husband-to-wife physical ag-
gression were excluded from participation. The subsample of
participants completed all relevant measures, and final marital
status (married vs. separated/divorced) was known for all couples.
See Christensen et al. (2004) for a complete description of recruit-
ment procedures, inclusion criteria, and study protocol and Atkins
et al. (2005) for the CONSORT flowchart for this study.
Participants in this sample ranged from 22 to 72 years of age at
pretreatment; the median age was 43 years (SD ⫽8.8) for men and
42 years (SD ⫽8.7) for women. Participants were, on average,
college educated (median level of education for both men and
women was 17 years, SD ⫽3.2). Median annual income was
$48,000 for the men and $36,000 for the women. Couples had been
married an average of 10.0 years (SD ⫽7.7). The sample was 77%
Caucasian, 8% African American, 5% Asian or Pacific Islander,
5% Latino/Latina, 1% Native American, and 4% other ethnicity.
Procedures
All couples participated in two assessments that were analyzed
for the current study. In the first of these assessments, conducted
prior to beginning therapy, couples completed a battery of self-
report questionnaires and participated in four 10-min videotaped
discussions. All measures of predictors were taken from this as-
sessment. Two of the interactions focused on a relationship prob-
lem (problem-solving discussions), and the other two focused on
an individual problem that was not a source of relationship distress
(social support discussions). Each spouse determined the topic for
one interaction of each type of discussion. The second assessment
occurred approximately 2 years after treatment termination and
was very similar to the pretherapy assessment in structure and
content.
Couples were classified as being either moderately distressed
(66 couples) or severely distressed (68 couples) prior to treatment
on the basis of a median split of a combined score on the Dyadic
Adjustment Scale (DAS; Spanier, 1976) and the Global Distress
Scale of the Marital Satisfaction Inventory—Revised (MSI–R;
Snyder, 1997). They were randomly assigned within these catego-
ries to receive up to 26 sessions of one of two behavioral therapies,
either TBCT (68 couples) or IBCT (66 couples). Details regarding
the procedure used for random assignment to therapy are presented
in Christensen et al. (2004). After completing treatment, couples
were not allowed to receive treatment from their study therapist for
a period of 2 years. We discouraged them from seeking couple
therapy elsewhere to prevent influences on posttreatment out-
comes by unknown or unspecified interventions and to allow them
to consolidate the gains they had made in treatment.
Measures
Demographic variables. Scores on all demographic variables
(age, education, income, presence of children, parental marital
status, years married, and gender) were obtained with a demo-
graphic questionnaire. Age and education were measured in years.
Income was measured in thousands of dollars of pretax monthly
income. Presence of children, parental marital status, and gender
were scored as dichotomous items. When there were discrepancies
between partners in the number of years married, wives’ reports
were used.
Intrapersonal Variables
Neuroticism. Neuroticism was assessed with the NEO Five-
Factor Inventory (NEO–FFI; Costa & McCrae, 1989). The NEO–
FFI is a widely used and well-validated, 60-item, short form of the
NEO Personality Inventory (NEO–PI; Costa & McCrae, 1985).
Cronbach’s alphas for men and women were .88 and .85, respec-
tively.
Mental Health Index. Overall mental health was assessed with
the Compass Outpatient Treatment Assessment System (COMPASS;
Sperry, Brill, Howard, & Grissom, 1996). The Mental Health
Index is a summary measure of the three subscales of the
COMPASS: Subjective Well-Being, Current Symptoms, and Cur-
rent Life Functioning. Higher scores indicate better mental health.
Cronbach’s alphas for men and women were .86 and .88, respec-
tively.
Psychological diagnoses. Presence of a psychological diagno-
sis was assessed with a dichomotomous score that indicates
whether or not spouses met criteria for a current psychological
disorder as defined by the DSM–IV (American Psychiatric Asso-
ciation, 1994) and determined by the SCID (First et al., 1994;
Spitzer, Williams, Gibbon, & First, 1994). See Christensen et al.
(2004) for details regarding interinterviewer reliability of this
measure. Of the 130 couples in the current study, 1 spouse had a
current, diagnosable disorder in 31 couples (24%) and both
spouses had a current, diagnosable disorder in 2 couples (2%). In
these couples, 8 husbands (6%) and 9 wives (7%) met criteria for
a mood disorder; 1 husband (1%) met criteria for a substance abuse
162 BAUCOM, ATKINS, SIMPSON, AND CHRISTENSEN

disorder;
2
9 husbands (7%) and 6 wives (5%) met criteria for an
anxiety disorder; and 2 wives (2%) met criteria for an adjustment
disorder. Of the 18 husbands and 17 wives who met criteria for a
current diagnosable disorder, 4 husbands and 9 wives met criteria
for a comorbid psychiatric disorder.
Family history of distress. Family history of distress was as-
sessed with the Family History of Distress Scale of the MSI–R
(Snyder, 1997). This scale assesses the level of distress in rela-
tionships within the family of origin; higher scores indicate more
distressed relationships. Cronbach’s alphas for men and women
were .76 and .79, respectively.
Communication
Affective communication. Affective communication was as-
sessed with the Affective Communication Scale from the MSI–R
(Snyder, 1997). The scale comprises 13 true/false items that indi-
cate the respondent’s degree of dissatisfaction with the amount of
affection and understanding expressed by his or her partner; higher
scores indicate greater dissatisfaction. Cronbach’s alphas for men
and women were.76 and .77, respectively.
Constructive communication. Constructive communication
was assessed with 7 Likert scale items (3 assessing constructive
behaviors and 4 assessing destructive behaviors) from the 35-item
Communication Patterns Questionnaire (CPQ; Christensen & Sul-
laway, 1984). The 4 destructive items are reverse scored and added
to the 3 constructive items to create a single index; higher scores
indicate higher levels of positive communication. Cronbach’s al-
phas for men and women were .69 and .67, respectively.
Demand/withdraw. Christensen & Sullaway, 1984): husband
demand/wife withdraw (HDWW) and wife demand/husband with-
draw (WDHW). Each subscale comprises 3 Likert scale items
from the 35-item CPQ. These 3 items index the likelihood of one
partner demanding and the other partner withdrawing across a
number of situations. Higher scores on both subscales indicate
greater engagement in the demand/withdraw interaction pattern
when discussing a problem. Husband and wife reports of both
subscales were highly correlated (HDWW, r⫽.43, p⬍.01;
WDHW, r⫽.44, p⬍.01); therefore, husband and wife reports
were averaged to create a single HDWW and WDHW score for
each couple. Cronbach’s alpha was .62 for HDWW and .54 for
WDHW.
Encoded arousal. Encoded arousal was assessed with range of
f
0
(maximum f
0
⫺minimum f
0
, measured in hertz) and was gen-
erated by analyzing audiotaped pretreatment problem-solving in-
teractions with the Praat computer program (Boersma & Weenink,
2005). One problem with the inclusion of arousal as a predictor of
treatment response is that many measures of arousal are compli-
cated, expensive, time consuming, and invasive. Fundamental fre-
quency (f
0
) is a noninvasive, alternative measure of encoded
arousal that is less expensive and complicated to capture than are
standard physiological measures of arousal. Fundamental fre-
quency refers to the pattern of vibration created by the vocal folds
during phonation, when the larynx regulates the outward flow of
air from the lungs and produces quasi-periodic vibrations with the
vocal folds (Kappas, Hess, & Scherer, 1990). Fundamental fre-
quency is very highly correlated (up to r⫽.9) with perceived pitch
(Kappas et al., 1990); higher fundamental frequency values corre-
spond to higher pitch. Recent reviews agree that f
0
measures are
robust and reliable indicators of encoded arousal (e.g., Juslin &
Sherer, 2005). Scores were averaged across both interactions for a
couple to generate one score for each spouse. Higher f
0
scores
indicate higher levels of encoded arousal.
Power processes. We assessed power processes with latent
semantic analysis (LSA; Landauer & Dumais, 1997) in order to
analyze transcriptions of pretreatment interactions separately for
the amount of hard and soft influence tactic language. LSA is a
quantitative method for assessing semantic content of text; in the
present study, we used it to assess the amount of hard and soft
influence tactics during couple discussions (see Baucom, 2008, for
a detailed description of the process of generating these scores).
We analyzed the language of both partners simultaneously, be-
cause LSA provides a more accurate index with larger amounts of
text. Some spouses did not generate enough content during a topic
to result in a reliable LSA score, so the decision was made to
analyze language at the level of the couple in order to maximize
sample size. Scores were averaged across both interactions for a
couple to generate one score for each couple.
Higher hard tactic scores indicate that partners used more ma-
nipulative language that allowed for less freedom in response to
their statements, and higher soft influence scores indicate that
partners used more collaborative language that allowed for more
freedom in response to their statements. For example, a request for
more time spent together could be made with hard language
(“Don’t you want to spend more time with me? You know it is
what I want and that it will be good for us”) or with soft language
(“I would really like it if we could find a way to spend more time
together. Would you be open to that?”).
Other Interpersonal Variables
Closeness/independence. Closeness/independence was as-
sessed with the Closeness and Independence Inventory (Heavey &
Christensen, 1991). This measure is a 9-item Likert scale of each
spouse’s desired level of closeness in the marital relationship.
Higher scores indicate a desire for greater closeness. Cronbach’s
alphas for men and women were.76 and .75, respectively.
Commitment. Commitment was assessed with the Marital Sta-
tus Inventory (MSI; Weiss & Cerreto, 1980). The MSI is a 14-
item, true/false response scale measure of steps taken toward
separation or divorce. Higher scores indicate lower commitment.
Cronbach’s alpha was .80 for both men and women.
Sexual satisfaction. Sexual satisfaction was assessed with the
Sexual Dissatisfaction Scale of the MSI–R (Snyder, 1997). The
scale measures the level of discontent with the frequency and
2
During the pilot study (see Christensen et al., 2004, for details of the
study design), we excluded potential participants for substance dependence
but not for substance abuse. In 1 couple, the husband met criteria for
substance abuse. We used the internal consistency of the DAS to calculate
the Reliable Change Index in creating the clinically significant change
categories. As noted by Bauer, Lambert, and Nielsen (2004), in clinical
samples, test–retest coefficients are deflated by actual individual differ-
ences in change. Bauer et al. and others (Martinovich, Saunders, &
Howard, 1996; Tingey, Lambert, Burlingame, & Hansen, 1996) have
concluded that internal consistency is a better measure of the psychometric
qualities of the outcome scale because of the deflation of test–retest
coefficients and because outcome scales are designed to test characteristics
that will change over time, rather than stable traits.
163
PREDICTION OF RESPONSE TO TREATMENT

quality of sexual activities. Higher scores indicate higher levels of
dissatisfaction with sexual activity in the relationship. Cronbach’s
alpha was .84 for both men and women.
Influence in decision making. Influence in decision making
was assessed with the Influence in Decision Making Questionnaire
(IDM; Kollock, Blumstein, & Schwartz, 1985). The IDM is an
8-item, Likert scale measure of perceptions by partners of their
relative influence over decisions across a number of categories
(e.g., how to spend money, where to go on vacation). All items are
summed to create a single index that indicates the overall amount
of influence that the responding spouse perceives himself or her-
self to have over the spouse. Higher scores indicate greater per-
ceived influence. Cronbach’s alphas for men and women were .73
and .78, respectively.
Power bases. We used individual reports of age, education,
and income to calculate absolute difference scores on all three
power bases. Higher scores indicate a greater absolute discrepancy
between partners in a given area of power.
Distress severity and treatment condition. Distress severity
was a dichotomous variable indicating either moderate or severe
marital dissatisfaction. (See Christensen et al., 2004, for a descrip-
tion of assignment to moderate and severe categories.) Treatment
condition was a dichotomous variable indicating assignment to
IBCT (Jacobson & Christensen, 1998) or TBCT (Jacobson &
Margolin, 1979). Both distress and treatment were used as poten-
tial moderating variables.
Clinical significance. The dependent variable was clinically
significant change category, which we assessed with an ordinal
variable that indicated four categories of change: deterioration
(reliable change in the direction of greater dissatisfaction), no
change, improvement (reliable change in the direction of greater
satisfaction), or recovery (reliable improvement and movement
into the nondistressed range) between pretreatment and the 2-year
follow-up. We calculated clinical significance with the procedures
outlined by Jacobson and Truax (1991) and used DAS scores
averaged across spouses.
3
The use of this index is a departure from
Atkins et al. (2005), who used trajectories of change on the DAS
as the treatment response index. Clinically significant change
categories were chosen as the index of treatment response in the
current study for three main reasons. First, there were a significant
number of divorces (n⫽24, 18% of the total sample) in the 2-year
period after treatment termination. It is not possible to collect
satisfaction data from a divorced couple; however, a divorced
couple clearly falls within the deteriorated relationship category.
Second, clinically significant change categories provide a more
stringent test of prediction. A measure may predict minor changes
in distress level but not in change categories. Finally, the use of
clinically significant change categories provides a more readily
3
We used the internal consistency of the DAS to calculate the Reliable
Change Index in creating the clinically significant change categories. As noted
by Bauer, Lambert, and Nielsen (2004), in clinical samples, test–retest coef-
ficients are deflated by actual individual differences in change. Bauer et al. and
others (Martinovich, Saunders, & Howard, 1996; Tingey, Lambert, Burlin-
game, & Hansen, 1996) have concluded that internal consistency is a better
measure of the psychometric qualities of the outcome scale because of the
deflation of test–retest coefficients and because outcome scales are designed to
test characteristics that will change over time, rather than stable traits.
Table 1
Means, Standard Deviations, and Correlations for Predictors
Variable MSD 1234567891011
1. Age (years) 42.6 8.3 —
2. Education (years) 17.0 2.6 ⫺0.03 —
3. Income 3.8 3.2 0.03 0.25 —
4. Parental marital status 0.4 0.4 ⫺0.12 ⫺0.21 0.01 —
5. Presence of children in marriage 0.7 0.5 ⫺0.16 ⫺0.09 0.02 0.19 —
6. Years married 10.1 7.6 0.61 ⫺0.08 0.01 ⫺0.06 0.09 —
7. Difference in age 4.2 3.7 0.10 ⫺0.05 0.13 0.09 0.18 ⫺0.04 —
8. Difference in education 2.8 2.6 0.14 0.19 0.21 ⫺0.02 0.13 0.02 0.10 —
9. Difference in income 3.2 3.7 0.12 0.05 0.53 0.01 ⫺0.03 0.18 0.27 0.04 —
10. Neuroticism 51.5 8.4 ⫺0.17 0.04 ⫺0.02 0.03 ⫺0.08 ⫺0.15 ⫺0.05 ⫺0.14 ⫺0.06 —
11. Mental Health Index 61.5 6.4 0.03 0.16 0.07 0.00 0.11 0.03 ⫺0.06 0.10 ⫺0.01 ⫺0.55 —
12. Current DSM–IV diagnosis 0.2 0.3 ⫺0.06 ⫺0.07 0.15 ⫺0.09 ⫺0.12 ⫺0.02 ⫺0.01 ⫺0.09 0.01 0.45 ⫺0.57
13. Family history of distress 55.8 6.1 ⫺0.01 0.02 0.00 0.27 0.04 ⫺0.11 ⫺0.05 0.03 0.08 0.17 ⫺0.09
14. Closeness/independence 31.0 4.6 ⫺0.01 ⫺0.14 ⫺0.02 ⫺0.03 0.12 0.03 0.03 0.14 ⫺0.07 ⫺0.07 0.06
15. Sexual Dissatisfaction Scale 60.3 7.9 0.17 ⫺0.05 ⫺0.16 ⫺0.02 0.03 0.14 0.18 ⫺0.01 0.03 ⫺0.06 ⫺0.07
16. MSI–R 4.0 2.4 ⫺0.17 ⫺0.11 0.00 0.14 0.10 ⫺0.07 ⫺0.01 ⫺0.10 0.00 0.14 ⫺0.25
17. Influence in decision making 40.4 6.4 ⫺0.15 0.02 0.14 0.05 ⫺0.03 ⫺0.10 ⫺0.04 0.21 ⫺0.15 0.14 0.04
18. Constructive communication ⫺2.5 7.4 0.01 0.07 ⫺0.09 ⫺0.15 ⫺0.18 ⫺0.06 ⫺0.06 0.02 ⫺0.13 ⫺0.12 0.06
19. Affective communication 63.4 5.0 0.16 ⫺0.06 ⫺0.05 ⫺0.11 0.23 0.20 ⫺0.06 0.06 ⫺0.08 ⫺0.10 0.08
20. Husband demand/wife withdraw 12.8 4.9 ⫺0.10 0.04 ⫺0.07 0.05 ⫺0.01 ⫺0.07 0.06 ⫺0.21 ⫺0.10 0.10 ⫺0.01
21. Wife demand/husband withdraw 17.9 4.4 ⫺0.01 ⫺0.07 0.05 0.09 0.05 0.11 ⫺0.09 0.04 0.17 0.18 ⫺0.25
22. Encoded arousal 431.1 15.8 ⫺0.25 0.02 ⫺0.06 ⫺0.09 0.08 ⫺0.13 ⫺0.06 ⫺0.01 0.14 0.15 ⫺0.02
23. Soft influence tactics 0.0 1.0 0.20 0.22 0.04 0.00 0.03 0.08 0.04 ⫺0.02 ⫺0.08 ⫺0.06 0.03
24. Hard influence tactics 0.0 1.0 0.29 0.10 ⫺0.01 0.01 ⫺0.13 0.10 0.00 ⫺0.06 ⫺0.03 0.01 ⫺0.13
25. Pretreatment severity 0.5 0.5 0.04 0.03 ⫺0.05 0.09 0.21 0.02 0.07 0.04 ⫺0.06 0.13 ⫺0.18
26. Therapy 0.5 0.5 ⫺0.04 0.00 0.02 ⫺0.08 0.03 0.00 ⫺0.10 ⫺0.04 ⫺0.07 0.08 0.00
Note. DSM–IV ⫽Diagnostic and Statistical Manual of Mental Disorders (4th ed., American Psychiatric Association, 1994); MSI–R ⫽Marital
Satisfaction Inventory—Revised.
164 BAUCOM, ATKINS, SIMPSON, AND CHRISTENSEN

interpretable metric for therapists, as it indicates the overall suc-
cess or failure in addition to improvement or decline.
Data Analyses
Predictors of response to treatment were identified with a series
of ordinal logistic regressions (Harrell, 2001). Groups of predictors
were analyzed in the following blocks: demographic, intraper-
sonal, communication, and other interpersonal variables. Each
block of variables included (a) main effects for treatment condi-
tion, pretreatment severity, and all predictors; (b) two-way inter-
actions between treatment condition and pretreatment severity as
well as between each predictor and treatment condition and pre-
treatment severity separately; and; (c) three-way interactions be-
tween each predictor, treatment condition, and pretreatment sever-
ity. All continuous predictors were centered prior to analysis and
calculation of interactions.
For each block, we used an automatic variable selection proce-
dure based on the Bayesian information criterion (BIC; Raftery,
1995) to identify optimal subsets of predictors. In choosing pre-
dictors, BIC weights the decision by sample size and number of
predictors; previous research has shown that typical stepwise pro-
cedures maximize prediction in the sample, whereas BIC maxi-
mizes prediction out of sample (Raftery, 1995). We analyzed plots
of predicted probabilities of membership in the four treatment
response categories, so we could interpret significant interactions
identified with the automated BIC algorithm. Given the data re-
quirements of ordinal logistic regression,
4
the relatively large
number of predictors,
5
and the modest number of data points, we
also conducted bootstrapped BIC analyses using 1,000 bootstrap
resamples to test the stability of the findings generated using the
entire sample (i.e., 1,000 pseudo-samples were generated via sam-
pling with replacement, and BIC model selection was conducted
on each sample; see Austin & Tu, 2004). Thus, in addition to
confidence intervals, we report the percentage of bootstrap sam-
ples in which a given predictor was selected. It is important to note
4
An important assumption of proportional odds logistic regression (POLR)
models is that the predictors demonstrate a linear association with the catego-
ries of the dependent variables. To test this assumption, we examined means
of predictors across treatment response categories. This examination revealed
that several of the predictors did not show a strictly linear pattern of increasing
means across increasing levels of the dependent variable. In order to test the
appropriateness of using a POLR model with this data, we ran continuation
ratio and extended continuation ratio models (see Harrell, 2001). These alter-
native models relax the POLR assumptions by (a) allowing for nonlinearity in
the patterns of means and (b) allowing for different patterns of association
between the predictor and the dependent variable across different categories of
the dependent variable. Neither model provided a fit to the data superior to that
of the POLR model, and thus we report the POLR results.
5
Empirical studies have shown that one consequence of fitting complex
models to smaller data sets is that coefficients can be larger than what
would be found in the population (i.e., the coefficients are too optimistic;
Harrell, 2001). Several approaches exist to “shrink” coefficients toward the
population values. Examination of two such methods—Bayesian POLR
with diffuse priors (Gelman, Jakulin, Pittau, & Su, 2008) and penalized
maximum likelihood (Harrell, 2001)—revealed little optimism in the co-
efficients; therefore, the coefficient estimates reported below appear to be
reasonable and are unadjusted.
12 13 14 15 16 17 18 19 20 21 22 23 24 25
—
0.10 —
⫺0.06 ⫺0.09 —
⫺0.08 0.01 0.00 —
0.23 0.04 ⫺0.29 0.15 —
⫺0.10 ⫺0.09 0.14 ⫺0.09 ⫺0.11 —
⫺0.13 ⫺0.08 ⫺0.10 0.00 ⫺0.30 ⫺0.17 —
⫺0.02 ⫺0.13 ⫺0.07 0.29 0.40 ⫺0.11 ⫺0.40 —
0.05 0.02 ⫺0.20 ⫺0.05 0.09 0.05 ⫺0.20 0.08 —
0.14 0.11 0.16 0.04 0.00 0.09 ⫺0.28 0.02 ⫺0.50 —
0.02 ⫺0.08 ⫺0.04 0.01 ⫺0.03 0.05 0.03 ⫺0.04 0.13 0.03 —
0.05 0.03 ⫺0.07 0.04 0.10 ⫺0.05 0.03 0.06 ⫺0.02 0.06 ⫺0.08 —
0.16 0.09 ⫺0.12 0.20 0.13 ⫺0.07 ⫺0.15 0.25 ⫺0.01 0.17 ⫺0.07 0.70 —
0.04 0.05 ⫺0.14 0.13 0.50 ⫺0.11 ⫺0.40 0.52 0.05 0.15 ⫺0.05 0.05 0.13 —
0.22 0.15 ⫺0.06 ⫺0.14 ⫺0.11 0.16 0.14 ⫺0.09 0.00 0.01 0.10 0.12 0.10 ⫺0.13
165
PREDICTION OF RESPONSE TO TREATMENT

that this statistic is different than statistical significance. Austin
and Tu (2004) noted, “One would expect that variables that truly
were independent predictors of the outcome would be identified as
predictors in a majority of the bootstrap samples, whereas noise
variables would be identified as predictors in only a minority of
samples” (p. 132). All analyses were conducted with R Version
2.6.0 (R Development Core Team, 2007).
Results
Means, standard deviations, and correlations for predictors,
moderators, and treatment response category are presented in
Table 1. In our examination of demographic predictors, the num-
ber of years that a couple had been married was significantly
associated with a positive response to treatment (B⫽0.13, p⬍
.01; see Table 2). For each additional year of marriage, the odds of
being in a better outcome category went up by 1.13. Bootstrap
analyses revealed that numbers of years married was a highly
reliable predictor of treatment response category, as it was selected
in 98% of the resamples. None of the other demographic variables
(age, education, income, presence of children, and parental marital
status) or any of the two- or three-way interactions involving any
of the demographic variables (including years married) emerged as
significant predictors of response to treatment.
None of the intrapersonal variables (overall mental health, pres-
ence or absence of DSM–IV diagnoses, neuroticism, and family of
origin environment), the other interpersonal variables (commit-
ment, influence in decision making, desired closeness, sexual
satisfaction, and power bases), or the self-reported communication
variables (demand/withdraw, affective communication, construc-
tive communication) emerged as significant predictors of response
to treatment. In addition, none of the two- or three-way interac-
tions involving intrapersonal, other interpersonal, or self-reported
communication variables emerged as significant predictors of re-
sponse to treatment.
There were, however, several interactions involving behavior-
ally based communication variables that emerged as significant
predictors of response to treatment (see Table 2). The interactions
of use of soft influence tactics with type of treatment (B⫽1.46,
p⬍.001) and wife’s encoded arousal with type of treatment (B⫽
0.07, p⬍.001) were significantly associated with treatment re-
sponse category. Use of soft influence tactics was significantly
associated with treatment response only for couples who received
IBCT, and the greatest differentiation was seen between couples
who were classified as recovered relative to couples who were
classified as deteriorated 2 years after treatment termination (see
Figure 1). In particular, higher levels of soft influence tactic use
were associated with a greater likelihood of being in a higher
treatment response category and lower levels of soft influence
tactic use were associated with a greater likelihood of being in a
lower treatment response category for couples who had received
IBCT (odds ratio [OR] ⫽4.29). Bootstrap analyses revealed this
interaction to be a reliable predictor of treatment response cate-
gory, as it was selected in 76% of the resamples.
Wife’s encoded arousal was significantly associated with treat-
ment response for couples who had received both IBCT and TBCT
(OR ⫽1.07); however, stronger effects were seen for couples who
had received TBCT than for couples who had received IBCT. In
particular, couples in which the wife had high encoded arousal
prior to beginning therapy were much more likely to be classified
as deteriorated relative to all other treatment response groups 2
years after treatment termination if they had received TBCT; if
they had received IBCT, they were marginally more likely to be
classified as recovered or improved than as unchanged or deteri-
orated. Couples in which the wife had low encoded arousal were
much more likely to be classified as recovered 2 years after
treatment termination regardless of whether they had received
IBCT or TBCT (see Figure 2). Bootstrap analyses revealed this
interaction to be a reliable predictor of treatment response cate-
gory, as it was selected in 70% of the resamples.
Two interactions with pretreatment severity emerged: hard in-
fluence tactics with pretreatment severity and wife’s encoded
arousal with pretreatment severity (B⫽1.55, p⬍.01; B⫽0.11,
p⬍.001, respectively). Use of hard influence tactics was signif-
icantly associated with treatment response category only for cou-
ples categorized as moderately distressed prior to treatment, and
the greatest differentiation was seen between couples who were
classified as recovered relative to couples who were classified as
deteriorated 2 years after treatment termination (see Figure 3). In
particular, using lower levels of hard influence tactics was asso-
ciated with a greater likelihood of being in a higher treatment
response category for couples who were classified as moderately
distressed prior to treatment (OR ⫽4.70). Bootstrap analyses
revealed that this interaction was a reliable predictor of treatment
response category, as it was selected in 82% of the resamples.
Wife’s encoded arousal was associated with treatment response
category mainly for couples who were classified as moderately
distressed prior to treatment, and the greatest differentiation was
seen between couples who were classified as improved or recov-
ered relative to all other treatment response categories 2 years after
treatment termination (see Figure 4). In particular, lower levels of
wife’s encoded arousal were associated with a greater likelihood of
being in a higher response category for couples who were classi-
fied as moderately distressed prior to treatment (OR ⫽1.12).
Bootstrap analyses revealed that this interaction is a reliable pre-
dictor of treatment response category, as it was selected in 92% of
the resamples. There is some evidence that is suggestive of a
crossover effect for wife’s encoded arousal across pretreatment
Table 2
Parameters for Significant Predictors of Treatment Response
Category
Variable BSEOR 95% CI for OR
Therapy 0.24 0.39 1.28 0.61, 2.82
Pretreatment severity ⫺1.04 0.40 0.35 0.15, 0.74
Years married 0.13 0.03 1.13 1.07, 1.21
Soft influence tactics 0.65 0.30 1.92 1.14, 3.64
Hard influence tactics ⫺0.78 0.30 0.46 0.25, 0.81
Wife’s encoded arousal ⫺0.01 0.01 0.99 0.96, 1.01
Soft Influence Tactics ⫻
Therapy 1.46 0.50 4.29 1.75, 12.47
Hard Influence Tactics ⫻
Pretreatment Severity 1.55 0.48 4.70 1.86, 12.33
Wife’s Encoded Arousal ⫻
Pretreatment Severity 0.11 0.03 1.12 1.06, 1.19
Wife’s Encoded Arousal ⫻
Therapy 0.07 0.03 1.07 1.02, 1.14
Note. OR ⫽odds ratio; CI ⫽confidence interval.
166 BAUCOM, ATKINS, SIMPSON, AND CHRISTENSEN

distress categories. Severely distressed couples appeared to be
more likely to respond more favorably to treatment when wife’s
display of encoded arousal was higher.
Discussion
This study examined a broad spectrum of variables measured
prior to treatment (demographic, intrapersonal, communication,
and other interpersonal variables) as predictors of categories of
response (deteriorated, no change, improved, and recovered) to
two couple therapies (IBCT and TBCT) 2 years after treatment
termination. Few studies have examined long-term response to
couple therapy using pretreatment predictors, so no specific hy-
potheses were made and potential predictors were identified
largely from studies of prediction of treatment response to couple
therapy at treatment termination. Several communication variables
and one demographic variable emerged as significant predictors of
treatment response category; a number of these variables inter-
acted with pretreatment severity or treatment type to differentially
predict outcome. It is noteworthy that three of the four significant
predictors of 2-year outcome were not self-report measures and
that almost all predictors focus on dynamic processes (emotional
arousal and influence tactics) rather than stable individual or
couple characteristics.
A number of communication variables emerged as significant
predictors of treatment response. This finding is in contrast to the
single demographic predictor (i.e., years married) and the lack of
intrapersonal variables that emerged as significant predictors of
treatment response. A considerable amount of empirical work has
documented the importance of demographic and intrapersonal
factors for overall marital functioning. For example, neuroticism
has been found to be one of the most robust and reliable cross-
sectional and longitudinal predictors of marital stability and qual-
ity (Karney & Bradbury, 1995; Kelly & Conley, 1987). However,
this study, as well as the Atkins et al. (2005) study of prediction of
treatment response at therapy termination, failed to find any evi-
dence that neuroticism predicts response to couple therapy. It is
unlikely that this null finding is a result of sample characteristics,
as the means and standard deviations for both spouses’ neuroticism
tscores were consistent with normative data for the NEO (hus-
bands, M⫽52.6, SD ⫽11.3; wives, M⫽50.3, SD ⫽10.7).
However, it is also important to consider that exclusionary criteria
included several psychological diagnoses. Though there is no
direct evidence of a restriction of range or inconsistency with what
would be expected from a random sample, it is possible that
exclusionary criteria impacted the possibility that neuroticism
would have predicted treatment response in the current study. This
lack of findings may be good news for couple therapy. Individual
psychological disturbance (at least within the current study of
relatively high-functioning participants) does not condemn couples
to treatment failure, though it may put them at risk for needing
couple therapy in the first place.
Predictors of Treatment Response for All Couples
Length of marriage predicted treatment response for all couples,
with spouses being more likely to respond favorably to therapy if
they had been married for longer periods of time. This finding is
Figure 1. Plot of predicted probability of treatment response (ClinSig) category for interaction between soft
influence tactic use by the couple and treatment. Pre-Tx ⫽pretreatment.
167
PREDICTION OF RESPONSE TO TREATMENT

in contrast to those of several previous studies and requires expla-
nation. Though relationship duration has not been linked to re-
sponse to couple therapy in prior studies, younger couples have
been found to benefit more from couple therapy than do older
couples (Baucom & Aiken, 1984; Bennun, 1985; Hahlweg, Schin-
dler, Revenstorf, & Brangelmann, 1984; O’Leary & Turkewitz,
1981). Age and relationship duration are clearly not the same
variable, but it is reasonable to think that younger couples have
shorter relationship durations. On the basis of this logic, greater
relationship duration predicting better treatment response is the
opposite of what would reasonably be hypothesized. It is important
to note that this study specifically recruited seriously and stably
distressed couples; prior studies have included a larger proportion
of mildly and moderately distressed couples. It may be that
spouses who are seriously and stably distressed but who have been
together for longer periods of time are more committed and have
already passed through or sorted out more problematic issues
(Atkins et al., 2005). A measure of behavioral commitment (the
MSI) was included as a predictor in the current study but was not
found to be a significant predictor of treatment response. However,
this measure is somewhat limited in that it indexes steps taken
toward divorce, or the bottom end of the range of commitment. In
addition, although it is logical to assume that younger couples will
have been married for a shorter period of time, it may not be
appropriate to assume that older spouses have longer relationship
durations. The average age of marriage has increased since the
publication of studies that linked younger age to more positive
response to treatment (United Nations, 2000). This societal trend
may have impacted the association between age of spouses and
relationship duration and thus have made them less viable proxies
for one another.
Pretreatment Distress ⫻Predictor Interactions
Hard influence tactics and wife’s encoded arousal were found to
predict response to treatment but only for couples who were
classified as moderately distressed prior to beginning treatment.
Other studies have found pretreatment satisfaction level to be
predictive of response to treatment (Snyder et al., 1993), as was
found in the current study, but this study is the first that we are
aware of that has found an interaction between pretreatment sat-
isfaction level and predictors of outcome. Why is it more difficult
to predict treatment response for more distressed couples? There
are two likely reasons. The first reason is that more distressed
couples have higher scores on multiple variables known to be
associated with relationship distress. For highly distressed couples,
the high scores on several of these risk factors may combine to
represent higher cumulative risk. Though individual risk factors
contribute to this cumulative risk, it is likely that individual vari-
ables become secondary to the overall gestalt they combine to
create. It is possible that this cumulative risk is why pretreatment
severity emerged as a significant risk factor and why fewer indi-
vidual variables predicted treatment response for severely dis-
tressed couples. The second related reason is that more severely
distressed couples’ scores tend to fall within a restricted range and
therefore have less statistical predictive power.
No specific hypotheses were made regarding hard influence
tactics and encoded arousal, though it is logical that lower levels of
Figure 2. Plot of predicted probability of treatment response (ClinSig) category for interaction between wife’s
encoded arousal and treatment.
168 BAUCOM, ATKINS, SIMPSON, AND CHRISTENSEN

hard influence tactics and lower levels of wife’s encoded arousal
are associated with positive response to treatment. Hard influence
tactics are characterized by high levels of emotional manipulation
and pressure and decreased room for spouses to discuss requests
for change. This finding is in line with Jacobson and Christensen’s
(1998) suggestion that a collaborative set, which is a shared sense
of investment in working on the relationship and a willingness to
compromise in order to strengthen the relationship, is a crucial
ingredient for successful couple therapy. To the best of our knowl-
edge, our study is the first to link pretreatment arousal to outcome
in couple therapy. This finding is consistent with those of studies
that have documented long-term associations between higher lev-
els of conflict related to arousal and an increased likelihood of
divorce (e.g., Gottman & Levenson, 1992) and with recent empir-
ical work documenting the importance of emotion regulation for
overall relationship functioning (Snyder, Simpson, & Hughes,
2006).
Though emotion regulation was not directly assessed, it is likely
that high levels of emotional arousal were indicative of intra-
and/or interpersonal difficulties with emotion regulation. Although
the pathways that link arousal to treatment outcome are currently
unknown, it may be that when spouses are highly aroused by their
own sources of distress, they are unable to join effectively with
their partners in processing old wounds or in problem solving
current conflicts during therapy sessions. The emotional and cog-
nitive demands of therapy may be too great when spouses are
already using all of their available coping skills to handle being in
a highly aroused emotional state. There was a minor suggestion in
the data that higher levels of arousal were associated with better
response to treatment for severely distressed couples. This effect
was small and unexpected; however, it may be that higher levels of
arousal are associated with more positive treatment response for
severely distressed couples because arousal is in part an index of
continued engagement of the relationship. Severely distressed cou-
ples who are unaroused during discussion of the problems in that
relationship may have disengaged beyond the point of repair.
Future studies should seek to replicate and further explore this
finding.
Therapy ⫻Predictor Interactions
Identification of prescriptive variables has long been one of the
aims of treatment outcome research on couple therapy. Uncovering
these variables has been difficult, and currently little is known about
what qualities of a couple or spouses would recommend that they
receive one couple therapy over another (see Snyder, Castellani, &
Whisman, 2006, for a review). Soft influence tactics and wife’s
encoded arousal emerged as two such variables in the current study.
Couples who used higher levels of soft influence tactics were sig-
nificantly more likely to respond well to treatment if they received
IBCT, whereas couples who used low levels of soft influence
tactics were significantly more likely to respond poorly to treat-
ment if they received IBCT. Soft influence tactics were unrelated
to treatment response for couples that received TBCT. One of the
primary interventions in IBCT is empathic joining, a technique that
aims to get spouses to share vulnerable emotions related to ongo-
ing distress (Jacobson & Christensen, 1998). Use of higher levels
of soft influence tactics, which are characterized by collaboration,
Figure 3. Plot of predicted probability of treatment response (ClinSig) category for interaction between hard
influence tactic use by the couple and pretreatment (Pre-Tx) severity.
169
PREDICTION OF RESPONSE TO TREATMENT

connection, and shared power, likely makes it easier for couples to
engage in empathic joining and thus to be more responsive to
IBCT. These findings suggest that couples that use high levels of
soft influence tactics may already be more responsive to IBCT,
whereas couples that use low levels of soft influence tactics would
benefit from a more skills-based treatment.
Couples in which the wife had higher levels of pretreatment
encoded arousal were more clearly at risk for being classified as
deteriorated 2 years after treatment termination if they had re-
ceived TBCT than if they had received IBCT. It is possible that
much of the encoded arousal expressed by wives before they began
therapy was associated with anger, frustration, and irritation. Al-
though this possibility is admittedly speculative, wives in this
sample were far more likely than husbands to initiate participation
in this study (Doss, Atkins, & Christensen, 2003), so there is some
indirect evidence that wives were discontented with the state of
their marriages and were motivated to pursue change. In the
lexicon of IBCT, anger, frustration, and irritation are all hard
emotions that include significant messages of blame. The primary
IBCT intervention techniques, such as empathic joining, seek to
shift from expressions of hard emotion to expressions of soft
emotion; TBCT does not contain techniques that explicitly target a
shift in emotional expression. Though additional study is required
to confirm this possibility, it is likely that interventions that spe-
cifically targeted emotion helped highly aroused wives shift the
emotions associated with their arousal over the course of therapy
and display greater benefits from the therapy they received. A shift
in emotions may also help wives to integrate their experience of
long-standing conflicts. A substantial literature links moderate-to-
high levels of emotional arousal associated with novel stimulation
to enhanced long-term memory consolidation (McGaugh, 2000). It
may be that arousal associated with new or different emotional
experiences enhanced learning ideas, concepts, and/or skills asso-
ciated with the new emotional experience, whereas arousal asso-
ciated with existing emotional experiences may have impaired
learning ability for new concepts, ideas, and skills delivered over
the course of therapy (Maroun & Akirav, 2008).
Pretreatment Severity
Pretreatment severity emerged as a moderately robust predictor
of treatment response for all couples. As was discussed above, this
variable may represent cumulative risk for severely distressed
couples. However, it is important to examine the possibility that
this finding is somewhat artifactual in nature. By definition, mod-
erately distressed couples are closer to the recovered category of
clinical significance, so it could be easier for them to get there over
the course of treatment. It also may be easier for severely dis-
tressed couples to be in the improved category due to regression to
the mean. To examine this possibility, we collapsed treatment
response categories into two groups: those who improved (recov-
ered and improved) and those who did not (no change and dete-
riorated). A roughly equal percentage of severely (61%) and mod-
erately (69%) distressed couples improved. The percentage of
couples that did not improve over the course of treatment was also
highly similar for moderately (31%) and severely (39%) distressed
couples. However, the percentage of couples that improved in
therapy and were in the recovered range by the end of treatment
Figure 4. Plot of predicted probability of treatment response (ClinSig) category for interaction between wife’s
encoded arousal and pretreatment severity.
170 BAUCOM, ATKINS, SIMPSON, AND CHRISTENSEN

was notably different between severely (50%) and moderately
(80%) distressed couples. Thus, it is likely that a proportion of the
effect of pretreatment severity is somewhat artifactual and is due to
the nature of clinical significance as an outcome.
Limitations
The current study is subject to several limitations. First, there is
a relative lack of research on long-term prediction of treatment
response to couple therapies, and, as a result, the current study was
guided by prediction studies of treatment response at therapy
termination. It is possible that important predictors were not in-
cluded in the current study. We took steps to counteract this
possibility by including theoretically viable predictors that had not
been previously explored, and a number of these variables turned
out to be significant (i.e., influence tactics and encoded arousal).
Second, a relatively large number of variables was explored for the
amount of data that was available. We used bootstrap resampling
techniques to minimize the possibility that the findings of the
current study reflect capitalization on chance and that they con-
verge to suggest that any effects due to chance are minimal. Third,
alternative methods for analyzing individual variables and com-
bining individual scores into couple-level indices may have re-
sulted in different findings. For example, weak- or strong-link
approaches could have been used in place of average individual
scores. Alternative methods for analyzing individual variables and
combining individual level scores into couple-level scores were
explored with several variables, and they did not reveal alternative
results.
6
Fourth, the use of some summary variables (such as the
presence or absence of a psychological diagnosis on the SCID and
overall psychopathology [the Mental Health Index] on the
COMPASS) may have limited the sensitivity of analyses per-
formed in this study. It is possible that continuous measures of
specific psychopathology, such as depression and anxiety, may
predict treatment outcome. Fifth, the demographic composition of
couples in the current study may limit the generalizability of these
results. Despite outreach efforts to obtain a diverse sample, cou-
ples were largely Caucasian and college educated. Finally, 4
couples who participated in the study were unable to be assessed
2 years after treatment termination and were excluded as a result.
There is no evidence that there was anything unique about these
couples that separated them from the couples that were assessed
and included; however, it is possible that exclusion of these 4
couples may have slightly altered the results of the current study.
Summary and Future Directions
The findings of the current study are cause both for optimism for
the field of couple therapy and for a call for continued efforts.
Optimism springs from the importance of malleable communica-
tion variables for treatment response combined with the relative
insignificance of static demographic and intrapersonal variables, as
well as from the Treatment ⫻Predictor interactions that suggest
that particular couples may differentially benefit from specific
treatments. Some of these predictors, such as encoded arousal and
influence tactics, have not been previously explored in prediction
studies. These findings necessitate replication but suggest fruitful
avenues for exploration by those seeking to understand the mech-
anisms of couple therapy, why particular treatments work better
for certain couples, and how treatments may be modified for the
particular needs of any given couple. Encoded arousal appears to
hold particular promise in this regard. One of the most powerful
effects identified by the treatment outcome research literature is
exposure. Exposure has yet to be studied as a mechanism of
change in couple therapy, but it is clearly implicated in the strat-
egies of IBCT that promote emotional acceptance. It is possible
that couples who benefit from IBCT will show significant declines
in arousal associated with conflict over the course of treatment.
The findings of the current study demonstrate that arousal plays an
important role in the outcome of couple therapy and suggest that
exploration of exposure as a mechanism of change in couple
therapy is likely to be fruitful. Some findings of the current study
(i.e., the positive effects of being married for longer periods of
time) require further study. In particular, additional study is needed
to help us understand the relative lack of predictors of treatment
outcome for couples who were severely distressed prior to treat-
ment. However, despite these caveats, the current study shows that
variables measured prior to treatment can substantially predict
outcome over 2 years later.
6
We conducted several additional analyses to test alternative methods
for analyzing individual variables and combining individual variables into
couple level indices. These additional analyses included the addition of a
variable representing the square of encoded arousal, the addition of average
couple level scores for each of the power bases, and the use of signed
difference scores for power bases rather than unsigned scores. None of
these additional analyses revealed any alternative findings or changed any
of the findings presented in the Discussion section.
References
Amato, P. R. (1996). Explaining the intergenerational transmission of
divorce. Journal of Marriage and the Family, 58, 628 – 640.
Amato, P. R., & DeBoer, D. D. (2001). The transmission of marital
instability across generations: Relationship skills or commitment to
marriage? Journal of Marriage and the Family, 63, 1038 –1051.
American Psychiatric Association. (1994). Diagnostic and statistical man-
ual of mental disorders (4th ed.). Washington, DC: Author.
Atkins, D. C., Berns, S. B., George, W., Doss, B., Gattis, K., & Chris-
tensen, A. (2005). Prediction of response to treatment in a randomized
clinical trial of marital therapy. Journal of Consulting and Clinical
Psychology, 73, 893–903.
Austin, P. C., & Tu, J. V. (2004). Bootstrap methods for developing
predictive models. American Statistician, 54, 131–137.
Baucom, B. (2008). The language of demand/withdraw: Verbal and vocal
channels of expression in dyadic interaction. Unpublished doctoral
dissertation, University of California, Los Angeles.
Baucom, D. H., & Aiken, P. A. (1984). Sex role identity, marital satisfac-
tion, and response to behavioral marital therapy. Journal of Consulting
and Clinical Psychology, 52, 438 – 444.
Baucom, D. H., Shoham, V., Mueser, K. T., Daiuto, A. D., & Stickle, T. R.
(1998). Empirically supported couple and family interventions for mar-
ital distress and adult mental health problems. Journal of Consulting and
Clinical Psychology, 66, 53– 88.
Bauer, S., Lambert, M. J., & Howard, K. I. (1996). Clinical significance
methods: A comparison of statistical techniques. Journal of Personality
Assessment, 82, 60 –70.
Bennun, I. (1985). Prediction and responsiveness in behavioural marital
therapy. Behavioural Psychotherapy, 13, 186 –201.
Boersma, P., & Weenink, D. (2005). Praat: Doing phonetics by computer
171
PREDICTION OF RESPONSE TO TREATMENT

(Version 4.3.27) [Computer program]. Retrieved October 7, 2005, from
http://www.praat.org/
Christensen, A. (1988). Dysfunctional interaction patterns in couples. In P.
Noller & M. A. Fitzpatrick (Eds.), Perspectives on marital interactions
(pp. 31–52). Clevedon, England: Multilingual Matters.
Christensen, A., Atkins, D. C., Berns, S., Wheeler, J., Baucom, D. H., &
Simpson, L. E. (2004). Traditional versus integrative behavioral couple
therapy for significantly and chronically distressed married couples,
Journal of Consulting and Clinical Psychology, 72, 176 –191.
Christensen, A., Atkins, D. C., Yi, J., Baucom, D. H., & George, W. H.
(2006). Couple and individual adjustment for 2 years following a ran-
domized clinical trial comparing traditional versus integrative behavioral
couple therapy. Journal of Consulting and Clinical Psychology, 74,
1180 –1191.
Christensen, A., & Sullaway, M. (1984). Communication Patterns Ques-
tionnaire. Unpublished manuscript, University of California, Los Ange-
les.
Costa, P. T., & McRae, R. R. (1985). The NEO Personality Inventory:
Manual. Lutz, FL: Psychological Assessment Resources.
Costa, P. T., & McRae, R. R. (1989). NEO/PI manual supplement. Lutz,
FL: Psychological Assessment Resources.
Cromwell, R. E., & Olson, D. H. (1975). Multidisciplinary perspectives of
power. In R. E. Cromwell & D. H. Olson (Eds.), Power in families (pp.
15–37). New York: Wiley.
Doss, B. D., Atkins, D. C., & Christensen, A. (2003). Who’s dragging their
feet? Husbands and wives seeking marital therapy. Journal of Marital
and Family Therapy, 29, 165–177.
Eldridge, K. A., & Christensen, A. (2002). Demand–withdraw communi-
cation during couple conflict: A review and analysis. In P. Noller & J. A.
Feeney (Eds.), Understanding marriage: Developments in the study of
couple interaction. (pp. 289 –322). New York: Cambridge University
Press.
First, M. B., Spitzer, R. L., Gibbon, M., & Williams, J. B. W. (1994).
Structured Clinical Interview for DSM–IV Axis Disorders (SCID–I/P)—
Patient Edition (Version 2.0). Arlington, VA: American Psychiatric
Publishing.
Freeman, S. J., Leavens, E. J., & McCulloch, D. J. (1969). Factors asso-
ciated with success or failure in marital counseling. Family Coordinator,
18, 125–134.
Gelman, A., Jakulin, A., Pittau, M. G., & Su, Y. (2008). A default prior
distribution for logistic and other regression models. Unpublished
manuscript.
Gottman, J. M., & Levenson, R. W. (1992). Marital processes predictive of
later dissolution: Behavior, physiology, and health. Journal of Person-
ality and Social Psychology, 63, 221–233.
Hahlweg, K., Schindler, L., Revenstorf, D., & Brangelmann, J. C. (1984).
The Munich marital therapy study. In K. Hahlweg & N. S. Jacobson
(Eds.), Marital interaction: Analysis and modification (pp. 3–26). New
York: Guilford Press.
Harrell, F. E. (2001). Regression modeling strategies: With applications to
linear models, logistic regression, and survival analysis. New York:
Springer.
Heavey, C. L., & Christensen, A. (1991). Closeness and Independence
Inventory. Unpublished manuscript.
Jacobson, N. S., & Christensen, A. (1998). Acceptance and change in
couple therapy: A therapist’s guide to transforming relationships. New
York: Norton.
Jacobson, N. S., & Margolin, G. (1979). Marital therapy: Strategies based
on social learning and behavior exchange principles. New York: Brun-
ner/Mazel.
Jacobson, N. S., Schmaling, K. B., & Holtzworth-Munroe, A. (1987).
Component analysis of behavioral marital therapy: 2-year follow-up and
prediction of relapse. Journal of Marital and Family Therapy, 13,
187–195.
Jacobson, N. S., & Truax, P. (1991). Clinical significance: A statistical
approach to defining meaningful change in psychotherapy research.
Journal of Consulting and Clinical Psychology, 59, 12–19.
Juslin, P., & Sherer, K. (2005). Vocal expression of affect. In J. Harrigan,
R. Rosenthal, & K. Scherer (Eds.), The new handbook of methods in
nonverbal behavioral research (pp. 65–136). New York: Oxford Uni-
versity Press.
Kappas, A., Hess, U., & Scherer, K. (1990). Voice and emotion. In B. Rime
& R. Feldman (Eds.), Fundamentals of nonverbal behavior (pp. 201–
238). New York: Cambridge University Press.
Karney, B. R., & Bradbury, T. N. (1995). The longitudinal course of
marital quality and stability: A review of theory, method, and research.
Psychological Bulletin, 118, 3–34.
Kelley, E. L., & Conley, J. J. (1987). Personality and compatibility: A
prospective analysis of marital stability and marital satisfaction. Journal
of Personality and Social Psychology, 52, 27– 40.
Kiecolt-Glaser, J. K., & Newton, T. L. (2001). Marriage and health: His
and hers. Psychological Bulletin, 127, 472–503.
Kipnis, D., Schmidt, S. M., & Wilkinson, I. (1980). Intraorganizational
influence tactics: Explorations in getting one’s way. Journal of Applied
Psychology, 65, 440 – 452.
Kollock, P., Blumstein, P., & Schwartz, P. (1985). Sex and power in
interaction: Conversational privileges and duties. American Sociological
Review, 50, 34 – 46.
Landauer, T. K., & Dumais, S. T. (1997). A solution to Plato’s problem:
The latent semantic analysis theory of the acquisition, induction, and
representation of knowledge. Psychological Review, 104, 211–240.
Maroun, M., & Akirav, I. (2008). Arousal and stress effects on consolida-
tion and reconsolidation of recognition memory. Neuropsychopharma-
cology, 33, 394 – 405.
Martinovich, Z., Saunders, S., & Howard, K. I. (1996). Some comments on
“Assessing clinical significance.” Psychotherapy Research, 6, 124 –132.
McGaugh, J. (2000, January 14). Memory: A century of consolidation.
Science, 287, 248 –251.
O’Leary, K. D., & Turkewitz, H. (1981). A comparative outcome study of
behavioral marital therapy and communication therapy. Journal of Mar-
ital and Family Therapy, 7, 159 –169.
Raftery, A. E. (1995). Bayesian model selection in social research (with
discussion). Sociological Methodology, 25, 111–196.
R Development Core Team (2007). R: A language and environment for
statistical computing. Vienna, Austria: R Foundation for Statistical
Computing. Retrieved from http://www.R-project.org
Shoham, V., Rohrbaugh, M. J., Stickle, T. R., & Jacob, T. (1998).
Demand–withdraw couple interaction moderates retention in cognitive–
behavioral versus family-systems treatments for alcoholism. Journal of
Family Psychology, 12, 557–577.
Snyder, D. K. (1997). Marital Satisfaction Inventory—Revised (MSI–R)
manual. Los Angeles: Western Psychological Services.
Snyder, D. K., Castellani, A. M., & Whisman, M. A. (2006). Current status
and future directions in couple therapy. Annual Review of Psychology,
57, 317–344.
Snyder, D. K., Mangrum, L. F., & Wills, R. M. (1993). Predicting couples’
response to marital therapy: A comparison of short- and long-term
predictors. Journal of Consulting and Clinical Psychology, 61, 61– 69.
Snyder, D. K., Simpson, J., & Hughes, J. N. (2006). Emotion regulation in
couples and families: Pathways to dysfunction and health. Washington,
DC: American Psychological Association.
Snyder, D. K., & Wills, R. M. (1989). Behavioral versus insight-oriented
marital therapy: Effects on individual and interspousal functioning.
Journal of Consulting and Clinical Psychology, 57, 39 – 46.
Snyder, D. K., Wills, R. M., & Grady-Fletcher, A. (1991). Long-term
effectiveness of behavioral versus insight-oriented marital therapy: A
4-year follow-up study. Journal of Consulting and Clinical Psychology,
59, 138 –141.
172 BAUCOM, ATKINS, SIMPSON, AND CHRISTENSEN

Spanier, G. B. (1976). Measuring dyadic adjustment: New scales for
assessing the quality of marriage and similar dyads. Journal of Marriage
and the Family, 38, 15–28.
Sperry, L., Brill, P. L., Howard, K. I., & Grissom, G. R. (1996). Treatment
outcomes in psychotherapy and psychiatric interventions. New York:
Brunner/Mazel.
Spitzer, R. L., Williams, J. B. W., Gibbon, M., & First, M. B. (1994).
Structured Clinical Interview for DSM–IV Personality Disorders
(SCID–II). Washington, DC: American Psychiatric Press.
Tingey, R. C., Lambert, M. J., Burlingame, G. M., & Hansen, N. B. (1996).
Clinically significant change: Practical indicators for evaluating psycho-
therapy outcome. Psychotherapy Research, 6, 144 –153.
United Nations. (2000). World Marriage Patterns 2000. Retrieved July 9,
2008, from http://www.un.org/esa/population/publications/worldmarriage/
worldmarriage.htm
Weiss, R. L., & Cerruto, M. (1980). The Marital Status Inventory: Devel-
opment of a measure of dissolution potential. American Journal of
Family Therapy, 8, 80 – 86.
Whisman, M. A., & Jacobson, N. S. (1990). Power, marital satisfaction,
and response to marital therapy. Journal of Family Psychology, 4,
202–212.
Received April 4, 2008
Revision received October 8, 2008
Accepted October 10, 2008 䡲
173
PREDICTION OF RESPONSE TO TREATMENT