Commitment, Pro-Relationship Behavior, and Trust in Close Relationships

Abstract

The present work advances and tests an interdependence-based model of the associations among commitment, pro-relationship behavior, and trust. Findings from two longitudinal studies revealed good support for model predictions. Commitment-inspired acts such as accommodation and willingness to sacrifice provide diagnostic information regarding a partner's pro-relationship motives. Individuals come to trust their partners when they perceive that their partners have enacted pro-relationship behaviors, departing from their direct self-interest for the good of the relationship. The results of mediation analyses are consistent with a model of mutual cyclical growth in which (a) dependence promotes strong commitment, (b) commitment promotes pro-relationship acts, (c) pro-relationship acts are perceived by the partner, (d) the perception of pro-relationship acts enhances the partner's trust, and (e) trust increases the partner's willingness to become dependent on the relationship. Auxiliary analyses revealed that self-reported attachment style does not account for substantial variance beyond the features of interdependence that form the basis for the present model.

Full text

INTERPERSONAL RELATIONS AND GROUP PROCESSES
Commitment, Pro-Relationship Behavior, and Trust in Close Relationships
Jennifer Wieselquist, Caryl E. Rusbult, and
Craig A. Foster
University of North Carolina at Chapel Hill
Christopher R. Agnew
Purdue University
The present work advances and tests an interdependence-based model of the associations among
commitment, pro-relationship behavior, and trust. Findings from two longitudinal studies revealed good
support for model predictions. Commitment-inspired acts such as accommodation and willingness to
sacrifice provide diagnostic information regarding a partner's pro-relationship motives. Individuals come
to trust their partners when they perceive that their partners have enacted pro-relationship behaviors,
departing from their direct self-interest for the good of the relationship. The results of mediation analyses
are consistent with a model of mutual cyclical growth in which (a) dependence promotes strong
commitment, (b) commitment promotes pro-relationship acts, (c) pro-relationship acts are perceived by
the partner, (d) the perception of pro-relationship acts enhances the partner's trust, and (e) trust increases
the partner's willingness to become dependent on the relationship. Auxiliary analyses revealed that
self-reported attachment style does not account for substantial variance beyond the features of interde-
pendence that form the basis for the present model.
The more he treated her as though she were really very nice, the more
Lotty expanded and became really very nice, and the more he,
affected in his turn, became really very nice
himself;
so that they went
round and round, not in a vicious but in a highly virtuous circle.
—Elizabeth von Arnim, The Enchanted April
Sometimes involvement with a close partner is simple. When
partners' goals correspond and their behavioral preferences are
compatible, partners can readily achieve desirable outcomes such
as intimacy, companionship, and security. It is easy to behave well
and do the right thing when interdependence structure is congenial.
The true test of a relationship arises when circumstances are not so
congenial—when partners encounter dilemmas involving con-
flicted interaction, incompatible preferences, or extrarelationship
temptation. In dilemmas of this sort the immediate interests of the
individual are incompatible with the interests of the relationship,
and something must give.
Jennifer Wieselquist, Caryl E. Rusbult, and Craig A. Foster, Department
of Psychology, University of North Carolina at Chapel Hill; Christopher R.
Agnew, Department of Psychology, Purdue University.
This research was supported in part by National Institute of Mental
Health Grant BSR-1-R01-MH-45417 and National Science Foundation
Grant BNS-9023817. We thank Steven Drigotas for making it possible for
us to gather the Study 1 data. We thank John Holmes for both direct and
-indirect contributions to our general model.
Correspondence concerning this article should be addressed to Jennifer
Wieselquist, Department of Psychology, University of North Carolina,
Chapel Hill, North Carolina 27599-3270. Electronic mail may be sent to
wiesel@email.unc.edu.
What leads partners to behave well when they encounter inter-
dependence dilemmas? Why are some partners willing to forego
immediate self-interest and promote the interests of their relation-
ships whereas others are disinclined to do so? What are the
consequences of pursuing self-interest rather than the interests of
one's relationship? This article proposes that two constructs—
commitment and trust—play central roles in shaping motivation
and behavior in ongoing relationships. Our model combines con-
cepts identified by Rusbult and her colleagues in their analysis of
commitment processes (Rusbult, Drigotas, & Verette, 1994) with
concepts identified by Holmes and Rempel (1989) in their analysis
of trust. These theoretical analyses are highly complementary, in
that both analyses rest on concepts from interdependence theory
(Kelley & Thibaut, 1978; Thibaut & Kelley, 1959).
We begin by suggesting that transformation of motivation
shapes preferences and behavior in everyday interdependence di-
lemmas. Next we introduce the concept of commitment, describing
the bases of commitment and outlining the role of commitment in
promoting pro-relationship behavior. Then we describe the emer-
gence of trust, characterizing trust as a reflection of the partner's
commitment and benevolent intentions; that
is,
we describe trust as
an implicit gauge of the extent and reliability of a partner's
pro-relationship motivation. Finally, we present the results of two
longitudinal studies designed to test our general model.
Transformation of Motivation and Pro-Relationship
Behavior
An interdependence dilemma is a "dilemma" because it involves
conflicting motives. On the one hand, there may be compelling
reasons to pursue immediate self-interest. On the other hand, there
Journal of Personality and Social Psychology, 1999, Vol. 77, No. 5, 942-966
Copyright 1999 by the American Psychological Association, Inc. 0O22-3514/99/$3.0O
942

COMMITMENT AND TRUST943
may be compelling reasons to promote the interests of one's
relationship. Resolving interdependence dilemmas therefore en-
tails some degree of effort or personal cost. For example, if Mary
enacts a rude or hostile behavior, John's immediate impulse may
be to behave rudely in return. John may feel demeaned, he may
wish to defend his dignity, or he may seek to gain some measure
of revenge. The impulse to reciprocate negativity—to defend one-
self in the face of attack—appears to be quite strong (Rusbult,
Verette, Whitney, Slovik, & Lipkus, 1991; Yovetich & Rusbult,
1994).
However, a retaliative act on John's part is likely to escalate
conflict, producing a hostile interaction that could harm his rela-
tionship with Mary. Thus, John's direct, self-interested impulses
are at odds with the interests of his relationship. From a strictly
personal point of view, the loss of pride John would suffer if he
were to swallow Mary's insult may seem more unpleasant than the
unpleasantness associated with retaliating, further irritating Mary,
and harming their relationship. Thus, if John is to behave in such
a manner as to benefit his relationship, he must swallow his pride,
control his impulse toward retaliation, and find it in himself to
behave in a conciliatory manner.
The interdependence theory distinction between the given situ-
ation and the effective situation provides a framework for under-
standing what makes some partners willing to endure cost or exert
effort to ensure the well-being of their relationships (Kelley &
Thibaut, 1978). The given situation refers to each partner's imme-
diate well-being in a specific situation, describing what we assume
to be each person's "gut level," self-centered preferences (e.g., the
impulse to retaliate when Mary behaves in a hostile manner). It
should be clear that people do not necessarily pursue their given
preferences. Frequently, behavior is shaped by broader concerns,
including strategic considerations, long-term goals, or desire to
promote both one's own and a partner's well-being. Movement
away from given preferences results from transformation of mo-
tivation, a process that leads individuals to relinquish their imme-
diate self-interest and act on the basis of broader considerations.
The effective situation describes the modified preferences that are
assumed to result from the transformation process; reconceptual-
ized, effective preferences are argued to guide behavior (e.g.,
accommodating rather than retaliating when Mary behaves in a
hostile manner).
The transformation process may produce a variety of orienta-
tions (Kelley & Thibaut, 1978; McClintock, 1972). In well-
functioning relationships the transformation process frequently
yields pro-relationship motives, producing a shift from desire to
maximize one's immediate self-interest (MaxOwn) toward pro-
relationship orientations such as desire to maximize the partner's
interests (MaxOther) or joint interests (MaxJoint; Yovetich &
Rusbult, 1994). At the same time, antirelationship transformation
is also possible (e.g., desire to maximize the difference between
one's own and the partner's interests, or MaxRel).
In light of the variety of transformational tendencies that part-
ners might display, interdependence dilemmas can be construed as
diagnostic situations (Holmes & Rempel, 1989; such situations
have also been described as "strain tests"; cf. Kelley, 1983a).
Interdependence dilemmas are "diagnostic" in the sense that be-
havior in such situations is revealing of the individual's broader
goals,
values, and motives. Does John resolve a specific dilemma
by pursuing his immediate self-interest, or does he set aside
self-interest for the good of the relationship? If John reacts to
Mary's rudeness by swallowing his pride and accommodating,
such a departure from his gut-level self-interest demonstrates his
benevolent feelings for Mary and his constructive goals for their
relationship.
Relatively stable transformational tendencies are argued to
emerge as a result of adaptation. Specific dilemmas initially are
experienced as unique problems, the resolution of which may
result from either controlled or automatic consideration of the
dilemma (cf. Uleman & Bargh, 1989). For example, to resolve a
given dilemma John may engage in active thought, reviewing his
options, analyzing surrounding circumstances, considering his
goals for the future, and deciding whether to pursue self-interest or
the interests of the relationship. Alternatively, John may react in an
automatic or impulsive manner, on the basis of his immediate
preferences. Either way, experience is acquired.
Over time, some dilemmas will be experienced repeatedly. As a
result of adaptation, individuals presumably develop habitual ten-
dencies to react to specific dilemmas in specific ways, such that
transformation may occur quite rapidly, with little or no conscious
thought (Kelley, 1983b). At critical choice points individuals may
continue to engage in active decision making, but just as often
transformation of motivation may be guided by habit. In some
relationships, partners may routinely engage in pro-relationship
transformation, whereas in other relationships partners may rou-
tinely react selfishly or in a manner that is antithetical to the
well-being of the relationship. Thus, it becomes important to
identify the bases of stable transformational tendencies.
Commitment Processes
Consistent with other characterizations of the commitment pro-
cess (cf. M. P. Johnson, 1991; Levinger, 1979), we suggest that
commitment reliably promotes pro-relationship motivation and
behavior (Rusbult et al., 1994). Commitment level represents long-
term orientation toward a relationship, including intent to persist
and feelings of psychological attachment (Rusbult, 1983; Rusbult
et al., 1994). Commitment develops as a result of changes over
time in three aspects of dependence. Individuals become increas-
ingly dependent on their relationships—and become increasingly
committed—to the degree that (a) satisfaction level is high, or the
relationship gratifies the individual's most important needs (e.g.,
needs for intimacy, companionship, sexuality); (b) quality of al-
ternatives is poor, or the individual's most important needs could
not be gratified independent of the relationship (e.g., on one's own,
in alternative romantic involvements, by friends or family mem-
bers);
and (c) investment size is high, or numerous resources have
become attached to the relationship (e.g., time and effort, joint
possessions, shared friendship network).
Commitment reliably promotes persistence in a relationship
(e.g., Bui, Peplau, & Hill, 1996; Drigotas & Rusbult, 1992; Felm-
lee,
Sprecher, & Bassin, 1990; Rusbult, 1983). Moreover, com-
mitment promotes a variety of so-called maintenance acts, includ-
ing (a) disparagement of alternatives, or tendencies to drive away
or derogate tempting alternative partners (D. J. Johnson & Rusbult,
1989;
Miller, 1997); (b) willingness to sacrifice, or tendencies to
forego desired activities for the good of a relationship (Van Lange
et al., 1997); and (c) accommodative behavior, or tendencies to
accommodate rather than retaliate when a partner behaves poorly
(Rusbult et al., 1991). In addition, commitment promotes cognitive

944WIESELQUIST, RUSBULT, FOSTER, AND AGNEW
tendencies such as (d) cognitive interdependence, or tendencies to
think in terms of "we, us, our" rather than "I, me, mine" (Agnew,
Van Lange, Rusbult, & Langston, 1998); and (e) positive illusion,
or tendencies toward excessively favorable evaluations of one's
partner or relationship (Murray, Holmes, & Griffin, 1996; Rusbult,
Van Lange, Yovetich, Wildschut, & Verette, 1999).
Existing research is compatible with the claim that commitment-
inspired maintenance acts result from transformation of motivation
(e.g., Rusbult et al., 1991; Van Lange et al, 1997; Yovetich &
Rusbult, 1994). For example, when confronted with accommoda-
tive dilemmas, individuals given plentiful reaction time (i.e., time
for transformation) react more constructively than do those given
limited reaction time; this discrepancy is not evident for situations
that do not require transformation of motivation. In addition,
existing evidence suggests that maintenance acts are associated
with couple well-being (e.g., Murray et al., 1996; Rusbult, Bisson-
nette, Arriaga, & Cox, 1998; Van Lange et al., 1997), operationally
defined as (a) probability of persisting and (b) scores on the
Dyadic Adjustment Scale (Spanier, 1976), a frequently used mea-
sure of quality of couple functioning.
Four features of commitment may explain why this variable
reliably promotes pro-relationship transformation. First, commit-
ted individuals are dependent and literally need their relation-
ships—the relationship provides desirable outcomes, investments
are sizeable, or alternatives are poor. The more individuals stand to
lose,
the more effort they will exert to hold on to what they've got
(cf. Holmes, 1981). Second, commitment involves long-term ori-
entation. In long-term involvements it is beneficial to develop
patterns of reciprocal pro-relationship behavior. Thus, pro-
relationship acts may represent a means of encouraging reciproc-
ity, thereby maximizing long-term self-interest (cf. Axelrod,
1984).
Also, with long-term orientation the costs of foregoing
self-interest are aggregated over a longer time perspective and in
light of the partner's reciprocal departures from self-interest (cf.
Kelley, 1983a). Third, commitment involves psychological attach-
ment. The self and the partner may become linked to the extent that
a departure from self-interest benefitting the partner may not be
experienced as personally costly (cf. Aron & Aron, 1997). And
fourth, commitment may induce a collectivistic, communal orien-
tation, including tendencies to respond to a partner's needs in a
rather unconditional manner. In a committed, communally ori-
ented relationship, partners may endure costs or exert effort with-
out counting what they receive in return (cf. Clark & Mills, 1979).
Interpersonal Trust
Thus far we have emphasized individual-level processes: We
have argued that individuals who are highly committed exhibit
pro-relationship transformation of motivation, and we have sug-
gested that such motivation makes individuals willing to depart
from their immediate self-interest in interdependence dilemmas.
However, this individual-level description presents only half of the
picture in understanding behavior in ongoing relationships.
In developing an interdependence-based explanation of pro-
relationship behavior, it is equally important to understand how
such individual-level processes influence partners (cf. Hinde,
1979).
Given that commitment plays a central role in relationships, it
would be adaptive for partners to implicitly or explicitly attend to
one another's commitment levels. Why so? First, commitment and
dependence make individuals vulnerable; such vulnerability is
reduced when the partner is equally vulnerable (equal vulnerability
represents balance of
power).
Second, couple well-being has been
shown to rest on both level of commitment and mutuality of
commitment (Drigotas, Rusbult, & Verette, in press; cf. Holmes &
Rempel, 1989; Kelley, 1983a). Third, maintenance acts follow the
principle of reciprocity—we are more willing to enact pro-
relationship behaviors to the degree that the partner is expected to
do so (Rusbult, Bissonnette, et al., 1998; Van Lange et al., 1997).
To the extent that achieving and sustaining equal dependence,
mutual commitment, and reciprocity of pro-relationship acts rest
on knowledge of a partner's commitment, an implicit gauge of the
partner's commitment would seem to have considerable functional
value. We suggest that relationship-specific trust is such a gauge.
Most theory and research concerning trust has examined this
phenomenon as a disposition—as a relatively enduring personal
attribute that is assumed to yield considerable stability in cogni-
tion, affect, and behavior across a variety of situations and across
a variety of interaction partners. For example, Rotter (1980) de-
scribed trust as a personality trait, and attachment theory empha-
sizes the ways in which early attachment experiences influence
mental models of attachment, which in turn shape later inclinations
to trust close partners (cf. Hazan & Shaver, 1994; Reis & Patrick,
1996).
In the present context it is suitable to describe trust as an
interpersonal phenomenon, construing trust not as a stable trait but
as a quality that is specific to a particular relationship with a
particular partner (cf. Hinde, 1979). Toward this goal, Holmes and
his colleagues have conceptualized trust as a relationship-specific
phenomenon, defining trust level as the expectation that a given
partner can be relied on to behave in a benevolent manner and be
responsive to one's needs (Holmes, 1989; Rempel, Holmes, &
Zanna, 1985; Sorrentino, Holmes, Hanna, & Sharp, 1995).
Trust is said to include three components (Holmes & Rempel,
1989):
(a) predictability, or belief that the partner's behavior is
consistent; (b) dependability, or belief that the partner can be
counted on to be honest, reliable, and benevolent; and
(c)
faith, or
conviction that the partner is intrinsically motivated to be respon-
sive and caring—belief that the partner's motives go beyond
instrumental bases for benevolence. Each component is argued to
be a necessary feature of trust.
How do we develop conviction that our partners are predictable
and dependable, and how do we develop faith that our partners
consistently will be responsive to our needs? Over the course of
extended involvement, partners inevitably confront situations in
which personal interests are pitted against the interests of the
relationship; that is, individuals encounter the sorts of interdepen-
dence dilemmas or "strain tests" described earlier. In such situa-
tions,
individuals implicitly make a choice: "Should I do what's
good for me, or should I put my partner's needs before my own?"
As noted earlier, Holmes and Rempel (1989) suggested that the
emergence of trust rests on the manner in which individuals are
perceived to behave during such episodes. Episodes of this sort are
termed diagnostic situations, in recognition of the fact that behav-
ior in such situations is diagnostic of the individual's broader
goals,
values, and motives.
We suggest that commitment-inspired maintenance acts are
diagnostic of pro-relationship orientation. Accommodation and

COMMITMENT AND TRUST945
sacrifice may provide particularly unambiguous evidence of be-
nevolent motives, in that when individuals accommodate rather
than retaliate—and when they sacrifice otherwise desirable activ-
ities to solve problems of noncorrespondence—they demonstrate
that they are willing to behave toward the partner in a generous and
giving manner. To some extent, cognitive maintenance tendencies
may provide parallel evidence, especially insofar as such tenden-
cies are seen to involve exceptional effort or cost. If it is true that
the benevolent acts promoted by commitment provide evidence
regarding strength of pro-relationship orientation, then trust can be
construed as a mirror reflecting the strength of a partner's
commitment.
As partners develop increased trust in one another they are
likely to become increasingly dependent on one another; that is,
they are likely to become increasingly satisfied with the relation-
ship,
increasingly willing to forego alternatives, and increasingly
willing to invest in the relationship (Holmes & Rempel, 1989). As
John becomes increasingly confident that Mary will be responsive
to his needs, he is likely to experience enhanced satisfaction with
their relationship. Moreover, as John's trust grows, he should be
more willing to make himself vulnerable by cognitively or behav-
iorally driving away alternatives, and he should be more willing to
throw in his lot with Mary by investing in their relationship,
emotionally or behaviorally. Such increased dependence will yield
strengthened commitment, which in turn should produce increased
willingness to engage in generous, pro-relationship acts.
In short, we are describing a mutual cyclical growth model in
which (a) Partner A's high dependence (high satisfaction, poor alter-
natives, high investments) increases A's commitment; (b) Partner A's
strong commitment motivates A's pro-relationship behavior; (c) Part-
ner B's observation of A's pro-relationship behavior increases B's
perceived partner pro-relationship behavior; (d) Partner B's perceived
partner pro-relationship behavior increases B's trust; and (e) Partner
B's strong trust makes B increasingly willing to become dependent,
which in turn enhances B's commitment, and so on. Thus, over the
course of a well-functioning, long-term relationship, each person's
movement toward increased commitment should be accompanied by
enhanced trust and parallel increases in commitment on the part of the
partner.
Research Overview and Hypotheses
Figure 1 presents a schematic representation of our model. We
conducted two studies to obtain evidence relevant to this model. In
A's
Dependence
Level
f \
B's
Dependence
Level
^ J
HIA's
Commitment
Level
V J
H5A's
Trust
Level
t
i
H5B's
Trust
Level
v. J
HIB's
Commitment
Level
^ J
H2A's
Prorelationship
Behavior
\ J
H3
H3
H2
A's
Perception of B's
Prorelationship
Behavior
v J
1
H4
1
B's
Perception of A's
Prorelationship
Behavior
^. J
B's
Proreiationship
Behavior
^ J
Figure 1. A mutual cyclical growth model of the associations among commitment, pro-relationship behavior,
and trust (HI through H5 refer to hypothesized associations among model variables).

946WIESELQUIST, RUSBULT, FOSTER, AND AGNEW
each study we obtained data from both partners in ongoing rela-
tionships—Study 1 was a three-wave longitudinal study of dating
relationships, and Study 2 was a two-wave longitudinal study of
marital relationships. We used two operational definitions of
pro-
relationship behavior, examining (a) accommodative behavior
(Studies 1 and 2) and (b) willingness to sacrifice (Study 1 only).
Given that our model is partially dyadic in character, some
predictions involve within-individual associations whereas other
predictions involve across-partner associations. For example, our
model predicts that increases in Partner A's commitment cause
increases in Partner A's pro-relationship behavior (a within-
individual link) and predicts that increases in Partner A's pro-
relationship behavior cause increases in Partner B's perceived
partner pro-relationship behavior (an across-partner link). To dis-
tinguish between within-individual and across-partner links, all
hypotheses are described as associations between variables for
Partner A and Partner B.
The model depicted in Figure 1 is symmetrical, with five con-
structs for Partner A and five parallel constructs for Partner B. In
light of this symmetry, our hypotheses describe the Figure 1 links
that begin with Partner A's dependence and end with Partner B's
dependence. Within-individual predictions describe both the asso-
ciations of men's predictors with their own criteria and the asso-
ciations of women's predictors with their own criteria. Across-
partner predictions describe both the associations of men's
predictors with women's criteria and the associations of women's
predictors with men's criteria.1
We obtained measures of relevant model variables on multiple
research occasions so that we could test our hypotheses in both
concurrent analyses and in analyses examining change over time in
each model variable. We anticipated that each of the links dis-
played in Figure 1 would be evident not only (a) in analyses
examining the associations among model variables at a single
point in time (i.e., in concurrent analyses) but also (b) in analyses
examining the associations of earlier predictors with later criteria,
controlling for earlier levels of the criterion (i.e., in residualized
lagged analyses). To the extent that a given predictor variable is
associated not only with (a) concurrent scores for a given criterion
but also with (b) change over time in that criterion, we are in a
better position to rule out plausible alternative explanations of our
findings (e.g., mood effects or self-report bias).
On the basis of the model described above, we advanced six
hypotheses. Hypotheses 1 through 5 represent each of the specific
links in the Figure 1 model. An additional hypothesis—termed the
"Commitment-Trust Hypothesis"—examines an association that
is not a direct link in the Figure 1 model. The Commitment-Trust
Hypothesis predicts that Partner A's commitment will be associ-
ated with Partner B's trust. Although A's commitment and B's
trust are not directly linked in Figure 1, given that we have
characterized trust as an implicit gauge of the strength of a part-
ner's commitment, this association should be evident if our
interdependence-based model is valid.
Commitment-trust
hypothesis:
Partner A's commitment level will be
positively associated with Partner B's trust level.
The first direct link in our model predicts that commitment
develops as a consequence of changes over time in dependence.
Numerous empirical investigations have revealed support for this
hypothesis, documenting the independent associations of satisfac-
tion level, quality of alternatives, and investment size with com-
mitment level (e.g., Bui et al., 1996; Felmlee et al., 1990; Rusbult,
1983;
Rusbult, Martz, & Agnew, 1998; Simpson, 1987). Accord-
ingly, in the present work we will examine a composite depen-
dence variable—a variable reflecting total dependence on a rela-
tionship.
Hypothesis
1: Partner A's dependence level will be positively asso-
ciated with Partner A's commitment level.
Second, and consistent with prior research regarding commit-
ment and pro-relationship acts, we predicted that strong commit-
ment would promote tendencies toward pro-relationship behaviors
such as accommodative behavior and willingness to sacrifice (e.g.,
Rusbult, Bissonnette, et al., 1998; Rusbult et al., 1991; Van Lange
et al., 1997).
Hypothesis
2:
Partner A's commitment level will be positively asso-
ciated with Partner A's pro-relationship behavior.
Our third hypothesis rests on the assumption that interdepen-
dence structure to some degree is "real" and therefore perceivable
by all involved parties. In contrast to the social constructionist
point of view (cf. Gergen, 1985), we anticipated that partners'
perceptions of one another's actions to some degree would be
veridical (at least insofar as we are in a position to judge that which
is "veridical"). Specifically, we predicted that there would be a
link between one person's self-reported tendencies toward pro-
relationship behavior and the partner's perceived partner pro-
relationship behavior (e.g., when I say that I accommodate, my
partner should perceive that I accommodate).
Hypothesis
3:
Partner A's self-report of pro-relationship behavior will
be positively associated with Partner B's perceived partner pro-
relationship behavior.
Fourth, we predicted that the perception of partner pro-
relationship acts such as accommodation and willingness to sac-
rifice would yield enhanced trust in the partner.
Hypothesis
4:
Partner B's perceived partner pro-relationship behavior
will be positively associated with Partner B's trust level.
Finally, in line with our model of mutual cyclical growth, we
predicted that trusting individuals will be willing to take risks and
make themselves vulnerable by becoming increasingly dependent
on their relationships (i.e., trust should be associated with high
satisfaction, poor alternatives, and high investments).
Hypothesis
5:
Partner B's trust level will be positively associated with
Partner B's dependence level.
1 We did not anticipate any substantively meaningful differences in the
strength of association among model variables for women and men. Ac-
cordingly, we refer to the partners in a given relationship as "Partner A"
and "Partner B," and do not distinguish between partners in our hypothe-
ses.
However, to explore the possibility of sex differences (and time
differences) in levels of
variables
and in the strength of association among
variables, we tested our hypotheses in regression analyses that included
main effects and interactions for both partner sex and time.

COMMITMENT AND TRUST947
Importantly, in addition to performing tests of the simple asso-
ciations predicted in Hypothesis 1 through 5, we also performed
parallel first-order mediation analyses (cf. Baron & Kenny, 1986).
If our model is valid, we should find that in analyses examining the
associations of each criterion with its presumed proximal and
next-most-distal predictor (a) the presumed proximal predictor
should continue to account for substantial variance in the criterion,
whereas (b) the presumed distal predictor should account for
reduced variance in the criterion. To the extent that tests of
mediation are consistent with expectations, we can feel more
confident of the validity of our model. For example, we will be in
a better position to rule out common method variance or socially
desirable responding as explanations of the observed simple asso-
ciations among model variables.
We also examined the simple associations of all model variables
with scores on the Dyadic Adjustment Scale (Spanier, 1976), a
frequently used measure of couple well-being. To the extent that a
process of mutual cyclical growth characterizes the associations
among model variables, we should find that all model variables
exhibit positive associations with quality of couple functioning.
Finally, in addition to examining the role of interdependence
structure in shaping the development of trust, we also explored the
associations of model variables with attachment style. A priori, it
might seem plausible that attachment style would play a role in the
process depicted in Figure 1. For example, given that the concept
of felt security resides at the heart of attachment theory, it might
seem reasonable to anticipate that trust would be positively asso-
ciated with secure attachment and negatively associated with
anxious-ambivalent and avoidant attachment. Also, given that
one's interdependence history is assumed to play a role in shaping
current transformational tendencies, it might seem plausible that
commitment and inclinations toward pro-relationship behavior
would exhibit parallel associations with the three attachment
styles. The extant literature provides support for this line of rea-
soning, suggesting that mental models of attachment exert mean-
ingful effects on the capacity to trust a partner and on willingness
to depart from self-interest on behalf of that person (for reviews,
see Hazan & Shaver, 1994; Reis & Patrick, 1996). We held the
general expectation that although such associations might be evi-
dent, our interdependence-based model would account for unique
variance above and beyond an attachment-based model.
Method
Study
1
Participants and recruitment. Study
1
participants were 53 couples (53
women,
53 men) who
participated
in a
three-wave longitudinal study
of
romantic relationships; Times
1 and 3
spanned about
a
10-week period
in
couples' relationships.2 We recruited participants
via a
two-stage process:
(a) descriptions
of the
project were posted
on the
notice board
for the
University
of
North Carolina research participant pool
as
well
as in the
campus newspaper;
and (b)
interested couples contacted Jennifer
Wie-
selquist to receive additional information about the project and to volunteer
for participation.
At Time 1 participants were 19.94 years old
on
average (17% freshmen,
28%
sophomores,
24%
juniors, 24% seniors, 8% other). The majority were
Caucasian (6% African American,
1%
Asian American, 91% Caucasian,
1%
Latino,
2%
other).
At
Time
1
participants
had
been involved
for an
average
of 19.17
months.
The
majority described their involvements
as
steady dating relationships (5% dating casually, 10% dating regularly, 74%
dating steadily,
11%
engaged or married). Most participants described their
relationships
as
monogamous (91% said neither partner dated others,
5%
said one partner dated others,
5%
said both partners dated others).
There
was
some attrition over
the
course
of the
study.
We
obtained
complete data—data
for
both partners
for all
model variables
for all
three
research occasions—for
40 of
53 couples (with
the
exception
of
missing
data
for the
occasional variable). Thirteen couples separated over
the
course
of
the study;
for
these couples,
we
obtained complete data
for the
research occasions prior
to
breakup.
The
sample included
53
couples
at
Time
1, 45
couples
at
Time
2, and 40
couples
at
Time
3. We
performed
preliminary analyses
to
determine whether couple attrition might have
influenced
the
obtained findings;
the
results
of
these analyses suggested
that attrition exerted little
or no
substantively meaningful effects
on the
observed associations among model variables.3
Research design and procedure.
We
obtained data from each couple
on three occasions over the course
of
an academic semester—once every
4
to
5
weeks.
At
each research occasion partners attended sessions during
which they completed questionnaires
and
participated
in
laboratory tasks
relevant
to
broader project goals. While completing their questionnaires,
partners were separated
to
prevent them from viewing
one
anothers'
responses.
We
assured participants that their responses would remain
confidential
and
that their partners would never
be
informed
of
their
responses.
At the end of
each research occasion
we
partially debriefed,
paid,
and
thanked couples
for
their assistance. Couples were paid $10
for
participation
at
each research occasion;
in
addition, individuals who were
recruited through
the
research participant pool received credit toward
partial fulfillment of the requirements
for
introductory psychology courses.
Questionnaires. The questionnaires included measures
of
dependence
(satisfaction, alternatives, investments), commitment level, accommoda-
tive behavior, perceived partner accommodation, willingness
to
sacrifice,
perceived partner willingness
to
sacrifice, trust level,
and
dyadic adjust-
2 Data from this project were also used
in (a)
Drigotas, Rusbult,
and
Verette
(in
press), which examined
the
association
of
mutuality
of
com-
mitment with couple well-being (Study
1
used data from Times
1, 2, and
3);
(b)
Drigotas, Rusbult, Wieselquist,
and
Whitton (1999), which exam-
ined
the
associations among partner affirmation, movement toward
the
ideal
self,
and
dyadic adjustment (Study 1 used data from Times
1, 2, and
3);
and (c) Van
Lange
et al.
(1997), which examined
the
association
of
commitment with willingness to sacrifice (Study
4
used data from Time 2).
3 We performed preliminary analyses
to
determine whether attrition
might have influenced
our
results. First,
we
categorized couples into
two
groups:
(a)
full-data couples (40 couples
for
whom
we had
complete data
at Times 1,
2,
and 3) and (b) partial-data couples (13 couples
for
whom
we
did not have complete data because
of
breakup).
We performed analyses
of
variance
on the
Time
1
data
to
assess initial differences between
the
groups. The groups
did not
initially differ
in
age, duration
of
relationship,
or Dyadic Adjustment. Importantly, separately for full-data and partial-data
couples,
we
calculated
the
correlations among
all
Time
1
variables—
Dependence, Commitment, Accommodation, Perceived Partner Accom-
modation, Sacrifice, Perceived Partner Sacrifice, Trust,
and
Dyadic
Ad-
justment. Out
of
28 correlations,
7
differed significantly
or
marginally;
in
six
of
seven instances, the association was weaker among full-data couples
(comparing
rs for
full- vs. partial-data couples, average
z =
-0.42). Such
differences
may be
attributable
to
greater variance
in
model variables
for
partial-data couples, who were
on
the road
to
breakup. Thus,
to the
extent
that attrition-based bias existed, such bias
at
worst was prejudiced against
our analysis,
in
that (a) we have fewer observations for partial-data couples
and (b) the associations among variables were
a
bit stronger for partial-data
couples. Although
it is
difficult
to
fully assess the extent
to
which attrition
influenced
our
findings, these analyses give
us
some confidence that such
influence
was
minimal.

948WIESELQUIST, RUSBULT, FOSTER, AND AGNEW
ment, along with additional measures that are irrelevant to the goals of the
present research (e.g., partner affirmation, correspondence of outcomes).
Following previous research regarding the investment model (Rusbult,
1983;
Rusbult, Martz, & Agnew, 1998), to measure dependence level we
included four items each to assess the three bases of dependence-
satisfaction level (e.g., "All things considered, to what degree do you feel
satisfied with your relationship?" 0 = not at all satisfied, 8 = completely
satisfied),
quality of alternatives (e.g., "All things considered, how attrac-
tive are the people other than your partner with whom you could become
involved" 0 = alternatives not at all appealing, 8 = alternatives are
extremely appealing; reverse scored, such that high values reflect high
dependence), and investment size (e.g., "Have you personally invested
things in your relationship that you would in some sense lose if the
relationship were to end—have you invested time or energy, have you
disclosed secrets to one another, etc.?" 0 = put nothing into relationship,
8 = put a great deal into relationship). We developed a single measure of
dependence by averaging responses to all scale items (a at Times 1, 2, and
3 = .76, .79, and .86).
Following previous research regarding commitment processes (Rusbult,
1983;
Rusbult, Martz, & Agnew, 1998), we included five items to measure
commitment level (e.g., "To what degree do you feel committed to main-
taining your relationship?" 0 = not very committed, 8 = completely
committed; a = .82, .84, and .90). To determine whether partners could
actually perceive one another's commitment, we included a parallel instru-
ment to measure perceived partner commitment (e.g., "To what degree
does your partner feel committed to maintaining your relationship?" 0 =
not very committed, 8 = completely committed; a = .82, .84, and .87). We
developed a single measure of each construct by averaging the items
designed to tap each variable.
Following previous research regarding accommodation processes (Rus-
bult et al., 1991), we included 16 items to measure accommodative behav-
ior. The instrument included four stems describing accommodative dilem-
mas—situations in which the partner enacted a potentially destructive exit
or neglect behavior (e.g., "When my partner is upset and says something
mean to me or snaps at me..."). Each stem was followed by four
items—one each for exit, voice, loyalty, and neglect (e.g., for exit, "I feel
so angry that I want to walk right out the door"). Participants reported the
frequency with which they engaged in each of 16 responses (four responses
for each of four stems; 0 = never, 8 = always; a = .71, .70, and .69). A
parallel instrument measured perceived partner accommodation (e.g.,
"When I'm upset and say something mean to my partner or snap at
him/her .. ., my partner talks to me about what's going on, trying to work
out a solution," 0 = partner never does this, 8 = partner always does this;
a = .80, .65, and .73). We reverse scored items where necessary so that
high numbers consistently reflected tendencies toward accommodation
(i.e.,
destructive items were reverse scored), and we developed averaged
measures of accommodative behavior and perceived partner
accommodation.
Following previous research regarding willingness to sacrifice (Van
Lange et al., 1997), we asked participants to list the four "most important
activities in your life, other than your relationship." Participants listed life
domains such as parents and siblings, career, education, religion, friends, or
pastimes (e.g., playing soccer). We pitted personal well-being against
relationship well-being using the logic of forced-choice methodology—for
each activity, the participant was asked to "Imagine that it was not possible
to engage in Activity No. 1 and maintain your relationship with your
partner. To what extent would you consider giving up this activity for the
good of your relationship?" (0 = definitely would not give up activity, 8 =
definitely would give up activity; a = .76, .82, and .86). A parallel
instrument measured perceived partner willingness to sacrifice (i.e., par-
ticipants described the four most important activities in the partner's life,
reporting the extent to which the partner would be willing to give up each
activity; a = .76, .85, and .86). We developed a single measure of each
construct by averaging the items designed to tap each variable.
To measure trust we identified the four most reliable items from the
predictability, dependability, and faith subscales of the Rempel et al.
(1985) instrument to develop a 12-item measure of trust level (see reli-
ability coefficients in Rempel et al., 1985, p. 103; e.g., "Though times may
change and the future is uncertain, I know my partner will always be ready
and willing to offer me strength and support," 0 = agree not at all, 8 =
agree completely; a = .89, .88, and .89). We developed a single measure
of trust by averaging responses to all scale items.
In order to examine the simple associations of all model variables with
couple well-being we included a version of Spanier's (1976) Dyadic
Adjustment Scale that is suitable for dating relationships. This 32-item
instrument assesses qualities of adjustment such as intimacy, agreement,
effective problem solving, and shared activities (e.g., "Do you and your
partner engage in outside interests together?" 0 = none of
them,
4 = all of
them;
"Do you kiss your partner?" 0 = never, 5 = every day; a = .90, .92,
and .94). We developed a single measure of adjustment by summing
responses to all scale items.
Study 2
Participants and recruitment. Study 2 participants were 65 couples (65
women, 65 men) who participated in Times 3 and 5 of
a
six-wave longitudinal
study of marital relationships; Times 3 and
5
spanned about a 12-month period
in couples' marriages.4 We recruited participants via a three-stage process: (a)
Over a 3-year period we located couples who applied for marriage licenses at
the Orange County, North Carolina, Courthouse; (b) research assistants tele-
phoned couples to determine whether they wished to receive project informa-
tion (interested couples were mailed such information); and (c) the principal
investigator telephoned couples to solicit their participation. The analyses
reported in this article are based on all couples who completed research
activities at Times 3 and 5 of the study, at which occasions we measured all
relevant model variables.
At Time 3 participants were 32.51 years old on average. All participants
had completed high school, 45% had bachelor's degrees, and 37% had
graduate degrees. Participants' personal annual salary was around $25,000.
The majority were Caucasian (2% African American, 3% Asian American,
94%
Caucasian). Forty-seven percent were Protestant, 18% were Catholic,
2%
were Jewish, and
33%
had other religious orientations (e.g., atheist). At
Time 3 participants had been married for about 2 years (24.02 months), and
38%
had given birth to one or more children.
There was some attrition over the course of the study. We obtained
complete data—data for both partners for all model variables for both
Time 3 and Time
5—for
65 couples (with the exception of missing data for
4 Data from this project were also used in (a) Arriaga and Rusbult
(1998),
which examined the association of accommodation with partner
perspective taking (Study 1 used data from Times 2, 4, and 6); (b)
Bissonnette, Rusbult, and Kilpatrick (1997), which examined the associa-
tions among commitment, empathic accuracy, and accommodation (data
from Times 2 and 4 were used); (c) Drigotas, Rusbult, and Verette (in
press),
which examined the association of mutuality of commitment with
couple well-being (Study 2 used data from Times 1, 3, and 5); (d) Drigotas,
Rusbult, Wieselquist, and Whitton (1999), which examined the associa-
tions among partner affirmation, movement toward the ideal
self,
and
dyadic adjustment (Study 4 used data from Time 6); (e) Gaines et al.
(1997),
which examined the association of attachment style with accom-
modation (Study 4 used data from Time 3); (f) Rusbult, Bissonnette, et al.
(1998),
which examined the association of accommodation with both
commitment and couple well-being (data from Times 1, 2, and 3 were
used);
(g) Rusbult et al. (1999), which examined the association of com-
mitment with positive illusion (Study 3 used data from Times 2 and 5); and
(h) Van Lange et al. (1997), which examined the association of commit-
ment with willingness to sacrifice (Study 6 used data from Times 3, 4,
and 5).

COMMITMENT AND TRUST949
the occasional variable).
A
total
of
123 couples participated
in
the study
at
Time
1, 88
couples participated
at
Time
3, and 65
couples participated
at
Time
5.
We performed preliminary analyses
to
determine whether couple
attrition might have influenced
the
obtained findings;
the
results
of
these
analyses suggested that attrition exerted little
or no
substantively mean-
ingful effects
on the
observed associations among model variables.5
Research design
and
procedure.
The
project
was a
six-wave lagged
longitudinal study: Couples commenced participation
at
different times but
engaged
in
parallel activities
at a
parallel pace, completing research activ-
ities
at
approximately 6-month intervals.
At
Times
3 and 5 we
mailed
partners copies
of
questionnaires, asking them
to
complete their question-
naires independently. Completed questionnaires were returned via the mail.
We assured participants that their responses would remain confidential and
that their partners would never
be
informed
of
their responses. At the end
of each research occasion we partially debriefed, paid, and thanked couples
for their assistance. Couples were paid
$25 for
participation
at
each
research occasion.
Questionnaires. The Time
3
and Time
5
questionnaires included mea-
sures
of
dependence level (satisfaction, alternatives, investments), commit-
ment level, accommodative behavior, perceived partner accommodation,
trust level, dyadic adjustment,
and
attachment, along with additional mea-
sures that are irrelevant to the goals
of
the present research (e.g., centrality
of relationship). The questionnaires included items that were similar to
(or
identical
to)
those used
in
Study 1
to
measure dependence level (13 items;
a at Times 3 and 5
=
.77 and .80), commitment level (5 items;
a =
.78 and
.84),
accommodative behavior
(16
items;
a = .81 and .83),
perceived
partner accommodation (16 items;
a =
.84
and
.87), trust level (12 items;
a
= .88 and .91), and
dyadic adjustment
(a
version suitable
for
marital
relationships;
32
items;
a =
.91
and
.94).
In
addition, participants read
the
Hazan
and
Shaver (1987) paragraph descriptions
of
secure, anxious-
ambivalent, and avoidant attachment and reported the degree to which they
agreed with each paragraph
(0 = not at all
characteristic
of
me,
8 =
extremely characteristic
of
me).
Results
Reliability
and
Validity
of
Measures
Test-retest correlations calculated for each variable revealed
good consistency between earlier and later measures for depen-
dence level (for Study 1, average r for Time
1-to-Time
2 and Time
2-to-Time 3 lags = .84; for Study 2, r for Time 3-to-Time 5 lag =
.77),
commitment level (rs = .80 and .75), perceived partner
commitment (for Study 1, r = .85; this variable was not measured
in Study 2), own accommodation (rs = .71 and .83), perceived
partner accommodation (rs = .81 and .74), own willingness to
sacrifice (for Study 1, r = .82), perceived partner willingness to
sacrifice (for Study 1, r = .80; sacrifice was not measured in Study
2),
trust level (rs = .86 and .75), and dyadic adjustment (rs = .68
and .86), as well as for secure, anxious-ambivalent, and avoidant
attachment (for Study 2, rs = .47, .60, and .61; allps < .01; these
variables were not measured in Study I).6
In Study 1 we used two operational definitions of pro-
relationship behavior, examining self-reported tendencies toward
both accommodation and sacrifice. We calculated concurrent cor-
relations between parallel measures to ensure that these variables
to some degree were distinguishable. As expected, measures of
own accommodation and own willingness to sacrifice were mod-
erately positively correlated (average r for Times 1, 2, and 3 =
.29),
as were measures of perceived partner accommodation and
perceived partner willingness to sacrifice (average r = .24; both
ps < .05).
Concurrent across-partner correlations revealed moderate-to-
good convergence in partners' reports of dyadic adjustment (for
Study 1, average r for Times 1, 2, and 3 = .42; for Study 2,
average r for Times 3 and 5 = .76). Also, men's reports of own
accommodation were correlated with their partners' reports of
perceived partner accommodation (rs = .48 and .49); parallel
associations were evident for women (rs = .52 and .61). Men's
reports of willingness to sacrifice were correlated with their part-
ners'
reports of perceived partner willingness to sacrifice (r = .51);
a parallel association was evident for women (r = .46). In addition,
men's reports of commitment were correlated with their partners'
reports of perceived partner commitment (for Study 1, r = .70); a
parallel association was evident for women (r = .77; allps < .01).
These analyses provide good support for the validity of our mea-
sures. More generally, these findings are compatible with the
assumption that interdependence structure to some degree is "real"
and therefore perceivable by both partners. Although there is some
subjectivity, the observed associations of self-report with partner-
report suggest that the data to a considerable degree reflect real
circumstances of interdependence: partners agree in their descrip-
tions of one another's commitment and pro-relationship acts and
agree in their descriptions of couple well-being.
5 We performed preliminary analyses
to
determine whether attrition
might have influenced
our
results. Using
the
full Time
3
data
set, we
categorized couples into two groups:
(a)
participating couples (65 couples
whose data are used
in
the present study,
for
whom we have complete data
at Times
3
and 5) and (b) discontinued couples (23 couples whose data
are
not included
in the
present study because they separated
or
discontinued
participation between Times
3 and
5). We performed analyses
of
variance
on
the
Time
3
data
to
assess differences between
the
groups. Compared
with discontinued couples, participating couples exhibited greater depen-
dence, trust, and dyadic adjustment. Importantly, separately
for
participat-
ing
and
discontinued couples,
we
calculated
the
correlations among
all
Time
3
variables.
Out of 36
correlations, only
3
differed significantly
or
marginally;
2
correlations among interdependence variables were weaker
among participating couples (comparing
rs for
participating vs. discontin-
ued couples, average
z =
—0.73)
and one
correlation with
an
attachment
variable
was
stronger among participating couples (average
z =
0.63).
Thus,
to the
extent that attrition-based bias existed,
our
analyses slightly
underestimate the strength
of
association among interdependence variables
and slightly overestimate
the
strength
of
association with attachment vari-
ables.
Although
it is
difficult
to
fully assess
the
extent
to
which attrition
influenced our findings, these analyses give
us
some confidence that such
influence was minimal.
6 We calculated concurrent correlations
of all
model variables with
duration of relationship—number
of
months dating
in
Study 1, and number
of months married
in
Study
2. The
duration measure was skewed
in
both
studies, so we performed log-linear transformations of
this
variable prior to
performing analyses.
In
Study
1
duration
was
positively correlated with
dependence, commitment,
and
perceived partner willingness
to
sacrifice
(rs
= .17, .25, and .19) and was
negatively correlated with perceived
partner sacrifice
(r =
-.15,
all ps <
.05); duration
was not
significantly
correlated with other model variables.
In
Study
2,
duration was negatively
correlated with dyadic adjustment
(rs =
—.19,
p <
.05); duration was
not
significantly correlated with other model variables
or
with measures
of
attachment style. (Given that we recruited Study
2
couples
at
the beginning
of their marriage—and given that couples proceeded through
the
study
more
or
less
in
tandem—in Study
2
there was relatively little across-couple
variability
in
duration
of
relationship.)

950WIESELQUIST, RUSBULT, FOSTER, AND AGNEW
Across-partner correlations for other model variables are not
directly relevant to assessing measure validity, in that these qual-
ities logically could differ for the partners in a given relationship.
For example, if husbands' and wives' commitment levels were
uncorrelated, this could be because partners' levels of commitment
frequently differ. Nevertheless, it was gratifying to discover that in
general, wives and husbands exhibited mutuality—partners re-
ported similar levels of dependence (for Studies 1 and 2, average
rs = .52 and .55), commitment (rs = .66 and .61), perceived
partner commitment (Study 1, r = .67), accommodation (rs = .42
and .56), perceived partner accommodation (rs = .32 and .43),
willingness to sacrifice (Study 1, r = .39), perceived partner
willingness to sacrifice (Study 1, r = .46), and trust (rs = .41 and
.55;
all ps < .05). More generally, these findings are compatible
with an assumption underlying our model of mutual cyclical
growth—the claim that over the course of extended involvement,
each person's movement toward increased dependence, commit-
ment, pro-relationship behavior, and trust will be accompanied by
parallel movement on the part of the partner. As one would
anticipate, partners' self-reported tendencies toward secure,
anxious-ambivalent, and avoidant attachment were essentially un-
correlated (for Study 2, rs = .12, .16, and .17).
Analysis Strategy
Separately for Study 1 and Study 2, we used a two-step analysis
strategy to estimate effect sizes and significance levels in simul-
taneous regression analyses. In Step
1
we calculated the proportion
of variance accounted for by each model variable, pooling the data
for male and female partners at all research occasions. All Step 1
analyses included main effects of time and sex, the Time X Sex
interaction, the interaction of time with each model variable, the
interaction of sex with each model variable, and the three-factor
interaction of Time X Sex with each model variable. Step 1
analyses provided estimates of (a) the beta for each variable and
(b) the proportion of variance accounted for by each variable
(Step 1 effect SS + Step 1 total 55).7
The data from male and female partners at multiple research
occasions are nonindependent, so the error terms in Step
1
analyses
are inappropriate (i.e., they are based on pooled nonindependent
data) and the degrees of freedom are inflated (e.g., the df reflect
observations from men and women on multiple research occa-
sions).
In Step 2 we performed analyses relevant to assessing the
significance of each Step 1 effect. In Step 2 we replicated the
Step 1 analyses separately for male and female partners at each
research occasion. We used the median error term from these
analyses in calculating significance levels for Step 1 effects. Thus,
the Step 2 procedures identified (a) the error term for calculating
the significance of each Step 1 effect (1 — [Step 2 model SS +
Step 2 total 55]), and (b) the appropriate degrees of freedom for
calculating the significance of each Step 1 effect (the Step 2
numerator and denominator df for each effect).8
On the basis of the results of the Step 1 and Step 2 analyses, we
calculated an F for each effect using the following equation:
(Step 1 effect 55 ^ Step 1 total 55)
(1 - [Step 2 model 55 + Step 2 total SSj)
Step 2 denominator effect df
Step 2 numerator effect df
To report ft for model variables, we calculated the (signed) square-
root of the F for each effect. Given that our analysis strategy is
relatively conservative, we report one-tailed significance tests for
all associations for which we advanced a priori hypotheses.
Note that the logic underlying our calculations parallels the
logic underlying traditional procedures for calculating effect sizes
and significance levels. The difference is that in estimating the
proportion of variance accounted for by a given variable (i.e., in
estimating numerator effects), we pooled data across multiple data
sets.
It is reassuring to note that the ?s obtained using this proce-
dure are very close to those obtained by averaging the rs from the
individual Step 2 analyses. In all instances, the significance versus
nonsignificance of the obtained t was identical to the significance
versus nonsignificance of the average t from the individual Step 2
analyses.
In testing each hypothesis we performed four types of analy-
sis—we calculated both concurrent and residualized lagged simple
associations, and we performed both concurrent and residualized
lagged mediation analyses (cf. Baron & Kenny, 1986). To facili-
tate interpretation of our results, when a given analysis includes
two or more model variables (i.e., in all analyses except those
7 In Study 1 six sets of data were pooled in the Step 1 analyses
examining concurrent associations among model variables (data for men's
and women's Time 1, 2, and 3 variables); in Study 2 four sets of data were
pooled in the Step 1 concurrent analyses (data for men's and women's
Time 3 and 5 variables). In Study 1 four sets of data were pooled in the
analyses examining lagged associations among variables (data for predict-
ing men's and women's Time 2 criteria from Time 1 predictors, as well as
data for predicting Time 3 criteria from Time 2 predictors); in Study 2 two
sets of data were pooled in the analyses examining lagged associations
(data for predicting men's and women's Time 5 criteria from Time 3
predictors). All variables were centered. In both studies, Step 1 analyses
examining the simple concurrent association of a given predictor (VarA)
with a given criterion included 7 variables: Time, Sex, Time X Sex, VarA,
VarA X Time, VarA X Sex, and VarA X Time X Sex. Analyses exam-
ining the simultaneous concurrent associations of two predictors (VarA and
VarB) with a given criterion included 11 variables—in addition to the
variables listed above, two-predictor models also included VarB, VarB X
Time, VarB X Sex, and VarB X Time X Sex. In Study 1, Step 1 analyses
examining the residualized lagged association of an earlier predictor
(Earlier-VarA) with a later criterion (Later-X) included 8 variables (or 12
variables, for a two-predictor model); in Study 2, Step 1 analyses exam-
ining the residualized lagged association of an earlier predictor with a later
criterion included 4 variables (given that there was just one time lag, there
were no main effects or interactions involving Time). In residualized
lagged analyses, later measures of the criterion were regressed onto earlier
measures of predictors, controlling for earlier levels of the criterion—in
addition to the variables listed above, each analysis also included as a
predictor variable the earlier measure of the criterion, Earlier-X. In essence,
this type of analysis examines the association of earlier predictors with
change over time in the criterion.
8 Step 2 analyses examining the association of
a
single predictor variable
with a given criterion included just one variable (i.e., there were no main
effects or interactions involving Time or Sex because Step 2 analyses were
performed separately for each sex at each time). In Study 1, six analyses
were performed in the Step 2 analyses examining concurrent effects, and
four analyses were performed in the Step 2 analyses examining lagged
effects; in Study 2, four analyses were performed in the analyses examining
concurrent effects, and two analyses were performed in the analyses
examining lagged effects.

COMMITMENT AND TRUST951
Table 1
Commitment-Trust Hypothesis: Predicting Partner Trust Level From Individual Commitment
Level—Concurrent and Lagged Associations, Studies I and 2
(Association of
A's
Commitment with B's Trust)
Variable
B's concurrent trust level
Study 1
A's commitment level
Study 2
A's commitment level
B's later trust level
Study 1
A's earlier commitment level
B's earlier trust level
Study 2
A's earlier commitment level
B's earlier trust level
Simple
j3
.45
.87
.40
.76
association
t
2.93**
9.95**
3.19**
8.74**
J3
.43
.46
.12
.82
.13
.73
Regression analysis
t
3.03**
3.80**
1.36t
8.40**
1.34t
7.39**
R2
.21**
.21**
.75**
.57**
Note. Table values are betas from regression analyses; significance levels are based on one-tailed tests. In
addition to the variables listed above, all analyses also included main effects of time and sex, the Time X Sex
interaction, and all two- and three-factor interactions of time and sex with each model variable.
t p < .10 (marginally significant).
**
p < .01.
examining concurrent simple associations) we also present the
simple association of each variable with the criterion (see statistics
under "Simple Association"). The "simple association" reflects the
association of a given model variable with the criterion in an
analysis that takes account of variance attributable to time, sex,
and interactions with time and sex. We present this type of simple
association rather than a correlation coefficient because this type
of simple association parallels the regression results and accord-
ingly represents a suitable baseline for interpreting unique associ-
ations. Thus, all statistics presented below are based on analyses
that included main effects and interactions for time and sex.9
Commitment-Trust Hypothesis: Is Partner A's
Commitment Level Associated With Partner B's
Trust Level?
Our model proposes that trust is an implicit gauge of the
strength of a partner's commitment. Although this association is
not a direct link in the Figure 1 model, the Commitment-Trust
Hypothesis suggested that if our interdependence-based model is
correct, we should observe evidence of a simple association be-
tween Partner A's commitment level and Partner B's trust level.
As can be seen in Table 1, the concurrent analyses revealed that in
both Study 1 and Study 2, Partner A's commitment was signifi-
cantly positively associated with Partner B's trust (see Table 1,
statistics under "B's concurrent trust level"; /3 = .43 and .46). The
residualized lagged analyses revealed that in both Study 1 and
Study 2, Partner A's earlier commitment was marginally associ-
ated with Partner B's later trust, controlling for B's earlier trust
(see Table 1, statistics under "B's later trust level"; /3 = .12 and
.13).
That is, Partner A's earlier commitment level accounted for
marginal unique variance in Partner B's later trust, above and
beyond Partner B's earlier trust—each individual's earlier com-
mitment predicted change over time in the partner's trust.
Hypotheses 1 Through 5: Concurrent and Residualized
Lagged Simple Associations
Table 2 summarizes the results of regression analyses examin-
ing the concurrent simple associations predicted in Hypotheses 1
through 5, and Table 3 summarizes the results of analyses exam-
ining the residualized lagged simple associations predicted in
Hypotheses 1 through 5. These analyses provide relatively good
support for our model. All 13 of the predicted concurrent simple
associations were statistically significant (see Table 2), all 13 of
the lagged simple associations were significant (see Table 3,
statistics under "Simple association"), and 9 of 13 residualized
lagged simple associations were significant or marginal (see Ta-
ble 3, statistics under "Regression analysis"). Below, we review
evidence relevant to each predicted association.
Hypothesis 1. Our model proposes that commitment is
strengthened to the extent that individuals are more dependent on
their relationships. Specifically, Hypothesis 1 suggested that de-
pendence level would be positively associated with commitment
level. The concurrent analyses revealed that in both Study 1 and
Study 2, dependence was significantly positively associated with
commitment (see under "Hypothesis 1" in Table 2; /3 = .75 and
.73).
Also consistent with Hypothesis 1, the residualized lagged
analyses revealed that in both Study 1 and Study 2, earlier depen-
dence predicted marginal change over time in commitment (see
under "Hypothesis 1" in Table 3; /3 = .16 and .19).
9 Out of a total of 54 regression analyses, there were only three margin-
ally significant main effects of time, four significant or marginal main
effects of sex, and two marginal interactions of time with other model
variables. (We used two-tailed significance tests in these analyses because
we did not advance a priori hypotheses regarding these variables.) Given
that this number of effects could easily have emerged by chance, we will
not address these scattered findings.

952WIESELQUIST, RUSBULT, FOSTER, AND AGNEW
Table 2
Hypotheses 1 Through 5: Concurrent Regression Analyses for
All Model Variables, Studies 1 and 2
Variable
Regression analysis
t R2
Hypothesis 1:
A's dependence level
—»
A's commitment level
A's commitment level
Study 1
A's dependence level
Study 2
A's dependence level
.75
.73
7.12**
7.89**
.58*
.56*
Hypothesis 2:
A's commitment level
—»
A's pro-relationship behavior
As accommodation
Study 1
A's commitment level
Study 2
A's commitment level
A's willingness to sacrifice
Study 1
A's commitment level
.48
.39
.52
3.37**
3.05**
3.92**
.23**
.16**
.28**
Hypothesis 3:
A's pro-relationship behavior -> B's perceived pro-relationship behavior
B's perceived accommodation
Study 1
A's accommodation
Study 2
A's accommodation
B's perceived willingness to sacrifice
Study 1
A's willingness to sacrifice
.48
.55
.49
3.63**
4.57**
3.33**
.27**
.32**
.26**
Hypothesis 4:
B's perceived pro-relationship behavior
—»
B's trust level
B's trust level
Study 1
B's perceived accommodation .72 6.29** .54**
Study 2
B's perceived accommodation .62
6.01**
.41**
B's trust level
Study 1
B's perceived willingness to
sacrifice .27 1.86* .12*
B's trust
B's dependence level
Study 1
B's trust level
Study 2
B's trust level
Hypothesis 5:
level
—>
B's dependence
.63
.53
level
4.69**
4.56**
.37**
.30**
Note. Table values are betas from regression analyses; significance levels
are based on one-tailed tests. In addition to the variables listed above, all
analyses also included main effects of time and sex, the Time X Sex
interaction, and all two- and three-factor interactions of time and sex with
each model variable.
*p<.05.
**/?<.01.
Hypothesis 2. Our model proposes that when individuals en-
counter interdependence dilemmas, they should be more willing to
engage in pro-relationship behaviors to the extent that they are
more strongly committed to their relationships. Specifically, Hy-
pothesis 2 suggested that commitment would be positively asso-
ciated with both accommodative behavior and willingness to sac-
rifice. The concurrent analyses revealed that all three tests of this
hypothesis were significant—the simple association of commit-
ment with accommodative behavior was significant in both stud-
ies,
and the association of commitment with willingness to sacri-
fice was significant in Study 1 (see under "Hypothesis 2" in Table
2;
sacrifice was not measured in Study 2). The residualized lagged
analyses revealed that two of three tests of this hypothesis were
significant or marginal—in Study 1, earlier commitment predicted .
significant change over time in accommodation and predicted
marginal change over time in willingness to sacrifice (see under
"Hypothesis 2" in Table 3). However, the residualized lagged
association of earlier commitment with later accommodation was
not significant in Study 2, possibly because of insufficient change
over time in accommodation (for the simple association of earlier
accommodation with later accommodation, /3 = .84).
Hypothesis 3. We assume that interdependence structure and
processes to some degree are observable and accordingly should
be perceived similarly by close partners; that
is,-
that there should
be an association between Partner A's self-reported tendencies
toward pro-relationship behavior and Partner B's perceived partner
pro-relationship behavior. The concurrent analyses revealed that
all three tests of this hypothesis were significant—for both accom-
modation and willingness to sacrifice, each individual's descrip-
tion of his or her pro-relationship behavior was associated with the
partner's perception of the individual's pro-relationship behavior
(see under "Hypothesis 3" in Table 2). The residualized lagged
analyses, too, revealed that all three tests were significant or
marginal—each individual's description of his or her pro-
relationship behavior was associated with change over time in the
partner's perception of the individual's pro-relationship behavior
(see Table 3).
Hypothesis 4. Our model suggests that individuals develop
trust in their partners when they observe the partner behave well in
diagnostic situations; that is, individuals develop increased trust
when they observe the partner depart from his or her immediate
self-interest for the good of the relationship. Accordingly, Hypoth-
esis 4 suggested that perceived partner accommodation and per-
ceived partner willingness to sacrifice would be positively associ-
ated with trust level. The concurrent analyses revealed that all
three tests of this hypothesis were significant (see "Hypothesis 4"
in Table 2). However, the residualized lagged analyses revealed
that only one of three tests of this hypothesis was even mar-
ginal—in Study 1, earlier perceived partner accommodation pre-
dicted marginal change over time in trust level (see Table 3). The
other two residualized lagged associations were not significant,
possibly because of insufficient change over time in trust (for the
simple association of the earlier and later measures of each crite-
rion, j3 = .76 and .87).
Hypothesis 5. Finally, we anticipated that trusting individuals
would be willing to make themselves vulnerable by becoming
increasingly dependent on their relationships. Therefore, Hypoth-
esis 5 suggested that trust level would be positively associated with
dependence level. The concurrent analyses revealed that in both

COMMITMENT AND TRUST953
Table 3
Hypotheses 1 Through 5: Residualized Lagged Regression Analyses
for All Model Variables, Studies 1 and 2
Simple
association
Variable 8 t
Regression analysis
t R2
Hypothesis 1:
A's dependence level
—»
A's commitment level
A's later commitment level
Study 1
A's earlier dependence level
A's earlier commitment level
Study 2
A's earlier dependence level
A's earlier commitment level
A's commitment 1
A's later accommodation
Study 1
A's earlier commitment level
A's earlier accommodation
Study 2
A's earlier commitment level
A's earlier accommodation
A's later willingness to sacrifice
Study 1
A's earlier commitment level
A's earlier willingness to sacrifice
.65
.79
.63
.73
Hypothesis 2:
5.35**
8.70**
5.94**
8.03**
.16
.68
.19
.61
evel
—»
A's pro-relationship behavior
.45
.72
.35
.84
.50
.83
Hypothesis 3:
A's pro-relationship behavior
—*
B's perceived
B's later perceived accommodation
Study 1
A's earlier accommodation
B's earlier perceived accommodation
Study 2
A's earlier accommodation
B's earlier perceived accommodation
B's later perceived willingness to sacrifice
Study 1
A's earlier willingness to sacrifice
.52
.85
.48
.74
.52
B's earlier perceived willingness to sacrifice .81
Hypothesis 4:
B's perceived pro-relationship behavior
B's later trust level
Study 1
B's earlier perceived accommodation
B's earlier trust level
Study 2
B's earlier perceived accommodation
B's earlier trust level
B's later trust level
Study 1
.71
.87
.52
.76
B's earlier perceived willingness to sacrifice .35
B's earlier trust level.87
2.94**
6.66**
2.79**
10.82**
3.38**
8.15**
.19
.62
-.05
.86
.19
.77
1.37t
5.42**
1.44t
4.75**
1.68*
4.78**
-0.55
9.79**
1.36t
6.87**
pro-relationship behavior
3.69**
3.49**
4.00**
7.11**
3.52**
7.46**
.14
.75
.15
.66
.20
.72
—>
B's trust level
5.65**
9.95**
4.54**
8.75**
2.15*
9.95**
.16
.77
.11
.68
.11
.82
1.32f
7.06**
1.39t
6.03**
1.71*
6.09**
1.38t
6.09**
0.94
5.85**
1.25
9.49**
.67**
.59**
.54**
.90**
.68**
.69**
.57**
.68**
.78**
.56**
.76**
Hypothesis 5:
B's trust level
—>
B's dependence level
B's later dependence level
Study 1
B's earlier trust level
B's earlier dependence level
Study 2
B's earlier trust level
B's earlier dependence level
.60
.84
.36
.73
4.11**
9.41**
2.97**
8.24**
.17
.751.51t
7.19**
-.09 -0.84
.80 7.66**
.73*
.61*
Note. Table values are betas from regression analyses; significance levels are based on one-tailed tests. In
addition to the variables listed above, all analyses also included main effects of time and sex, the Time X Sex
interaction, and all two- and three-factor interactions of time and sex with each model variable,
tp < .10 (marginally significant). *p < .05.
**
p < .01.

954WIESELQUIST, RUSBULT, FOSTER,
AND
AGNEW
studies, trust level
was
significantly positively associated with
dependence level
(see
"Hypothesis 5"
in
Table 2). The residualized
lagged analyses revealed that
in
Study
1,
earlier trust predicted
marginal change over time
in
dependence (see Table 3). However,
the residualized lagged association
was
nonsignificant
in
Study
2
because
of
either
(a)
insufficient change over time
in
dependence
(for
the
simple association
of
earlier dependence with later depen-
dence,
/3
=
.73),
or
(b) suppression (the beta
for
earlier dependence
was .73
in the
simple association
but .80 in
the regression analysis,
and
the
beta
for
earlier trust
was .36 in the
simple association
but
— .09
in the
regression analysis).
Hypotheses
1
Through
5:
Concurrent
and
Residualized
Lagged Mediation Analyses
The simple associations reported above provide very good
sup-
port
for our
model. However, there
are
plausible alternative
ex-
planations
of
these findings—the observed associations might
be
attributable
to
common method variance
or to
participants'
ten-
dencies
to
describe well-functioning relationships
in a
generally
favorable light.
As a
more stringent test
of the
plausibility
of the
model advanced
in the
introduction
and
displayed
in
Figure
1, we
performed
a
series
of
first-order concurrent
and
residualized
lagged mediation analyses
(cf.
Baron
&
Kenny, 1986).
To the
extent that
our
model
is
valid,
we
should find that
in
simultaneous
regression analyses examining
the
associations
of
each criterion
with
its
presumed proximal
and
next-most-distal predictor
(a) the
presumed proximal predictor should continue
to
account
for sub-
stantial variance
in the
criterion, whereas
(b) the
presumed distal
predictor should account
for
reduced variance
in the
criterion.
Concurrent mediation analyses. Table
4
summarizes
the re-
sults
of
concurrent mediation analyses that correspond
to
Hypoth-
eses 1 through
5. As
noted earlier, each criterion
was
significantly
positively associated with
its
presumed proximal predictor—all
simple associations were significant
(see
statistics under "Simple
association"
in
Table
4; the
presumed proximal predictor
is pre-
sented
in the
first
row for
each model).
And
consistent with
expectations, most criteria were positively associated with their
presumed distal predictors—13
of 14
simple associations were
significant
or
marginal (see statistics under "Simple association"
in
Table
4; the
presumed distal predictor
is
presented
in the
second
row
for
each model).
In all
14 cases, the association
of
the criterion
with
its
presumed proximal predictor was descriptively larger than
the association
of the
criterion with
its
presumed distal predictor.
Consistent with expectations, when
we
performed simultaneous
regression analyses examining
the
associations
of
each criterion
with
its
presumed proximal
and
next-most-distal predictor,
we
found that coefficients
for
the presumed proximal predictor tended
to remain relatively strong—for presumed proximal predictors,
the
betas observed
in the
mediation analyses declined
by an
average
of
only
.07
compared with
the
betas observed
in the
simple associa-
tions
(see
statistics under "Regression analysis"
in
Table
4). In
all
14
analyses,
the
presumed proximal predictor accounted
for
significant
or
marginal variance
in the
criterion above
and
beyond
the presumed distal predictor. Also consistent with expectations,
the simultaneous regression analyses revealed that coefficients
for
the presumed distal predictor tended
to
decline rather substantial-
ly—for presumed distal predictors,
the
betas observed
in the me-
diation analyses declined
by an
average
of .27
compared with
the
betas observed
in the
simple associations
(see
statistics under
"Regression analysis"
in
Table 4). Only
2 of 14
analyses revealed
that
the
presumed distal predictor accounted
for
even marginal
variance
in the
criterion above
and
beyond
the
presumed proximal
predictor.
Consistent with Hypothesis
1,
when
we
regressed commitment
simultaneously onto
its
presumed proximal
and
distal predictors,
associations with dependence
(the
presumed proximal predictor)
remained relatively strong, whereas associations with trust level
(the presumed distal predictor) declined substantially. Consistent
with Hypothesis
2,
when
we
regressed indexes
of
pro-relationship
behavior—both accommodation
and
sacrifice—simultaneously
onto their presumed proximal
and
distal predictors, associations
with commitment tended
to
remain strong, whereas associations
with dependence declined considerably. Consistent with Hypoth-
esis
3,
regression analyses examining Partner
B's
perceptions
of
the partner's pro-relationship behavior—both perceived partner
accommodation
and
perceived partner willingness
to
sacrifice—
revealed that associations with Partner
A's
self-reported
pro-
relationship behavior tended
to
remain quite strong whereas asso-
ciations with Partner
A's
commitment declined substantially.
Consistent with Hypothesis
4,
regressions
of
Partner
B's
trust onto
Partner
B's
perceptions
of the
partner's pro-relationship behav-
ior—along with Partner
A's
self-report
of
pro-relationship behav-
ior—revealed that associations with the former variables remained
strong whereas associations with
the
latter variables declined
con-
siderably.
And
consistent with Hypothesis
5,
when
we
regressed
dependence onto
its
presumed proximal
and
distal predictors,
associations with trust tended
to
remain strong whereas associa-
tions with perceptions
of the
partner's pro-relationship behavior
declined substantially. These results provide excellent support
for
the fairly precise model
of
associations among variables that
we
outlined
in the
introduction.
Is
it
possible that
the
association
of
perceived partner
pro-
relationship behavior with trust results from projection?
In
report-
ing
on
perceived partner accommodation
and
willingness
to sac-
rifice, it
is
possible that individuals operate
on the
assumption that
their partners engage
in
pro-relationship acts
to the
same degree
as
they themselves
do. If
this line
of
reasoning
is
valid,
it is
possible
that
the
observed association
of
perceived partner accommodation
(and perceived sacrifice) with trust is attributable
to
the association
of perceived partner accommodation with
own
accommodation
(and
to the
association
of
perceived partner sacrifice with
own
sacrifice).
We
performed auxiliary analyses
to
explore this possi-
bility. First,
we
examined
the
simple associations
of
trust level
with both perceptions
of the
partner's pro-relationship behavior
(|3 = .72 and
.62
for
perceived partner accommodation
and .27 for
perceived partner sacrifice)
and
self-reports
of own pro-
relationship behavior
(/3 = .62 and .58 for own
accommodation
and
.22 for own
sacrifice;
all ps < .10).
Thus, both perceived
partner pro-relationship behavior
and own
pro-relationship behav-
ior exhibit simple associations with trust. When
we
regressed trust
level onto both perceived partner accommodation
and self-
reported
own
accommodation, coefficients remained strong
for
perceived partner accommodation
(jS = .54 and .45,
both
ps <
.01),
whereas coefficients declined considerably
for own
accom-
modation (j8
= .27 and
.24, bothps
<
.10). Results were less clear
for sacrifice,
in
that when
we
regressed trust level onto both
perceived partner sacrifice
and
self-reported
own
sacrifice,
coef-

COMMITMENT AND TRUST
955
Table
4
Hypotheses
1
Through
5:
Concurrent Mediation Analyses for
All
Model
Variables,
Studies
1 and 2
Variable
A's dependence
A's commitment level
Study 1
A's dependence level
A's trust level
Study 2
A's dependence level
A's trust level
Simple
association
j3
Hypothesis
level
—>
A':
.75
.59
.73
.49
Hypothesis
t
1:
5 commitment level
7.12**
4.17**
7.89**
4.35**
2:
.65
.18
.65
.15
Regression analysis
t
5.03**
1.20
6.07**
1.36t
A's commitment level
—>
A's pro-relationship behavior
A's accommodation
Study 1
A's commitment level
A's dependence level
Study 2
A's commitment level
A's dependence level
A's willingness to sacrifice
Study 1
A's commitment level
A's dependence level
A's pro-relationship behavioi
B's perceived accommodation
Study 1
A's accommodation
B's commitment level
Study 2
A's accommodation
B's commitment level
B's perceived willingness to sacrifice
Study 1
A's willingness to sacrifice
B's commitment level
.48
.39
.39
.38
.52
.47
Hypothesis
3.37**
2.66**
3.05**
3.13**
3.92**
3.46**
3:
.39
.11
.25
.20
.34
.22
'
—»
B's perceived pro-relationship
.48
.32
.55
.28
.49
.40
3.63**
1.95*
4.57**
2.13*
3.33**
2.77**
Hypothesis 4:
B's perceived pro-relationship behavior
—>
B's trust level
Study I
B's perceived accommodation
A's accommodation
Study 2
B's perceived accommodation
A's accommodation
B's trust level
Study 1
B's perceived willingness to sacrifice
A's willingness to sacrifice
.72
.46
.62
.47
.27
.15
Hypothesis
6.29**
3.30**
6.01**
3.93**
1.86*
0.92
5:
B's trust level
—»
B's dependence
B's dependence level
Study 1
B's trust level
B's perceived accommodation
Study 2
B's trust level
B's perceived accommodation
B's dependence level
Study 1
B's trust level
B's perceived willingness to sacrifice
.63
.46
.53
.29
.63
.36
4.69**
3.08**
4.56**
2.26*
4.69**
2.32*
.43
.11
.51
.08
.36
.22
1.68*
0.49
1.38t
1.05
1.36t
1.06
behavior
2.52**
0.60
3.96**
0.62
2.01*
1.20
B's trust level
.64
.15
.53
.16
.29
.02
level
.63
.01
.56
-.05
.59
.20
4.85**
1.12
4.10**
1.23
1.61t
0.16
3.19**
0.02
3.61**
-0.32
4.14**
1.50t
R2
.61**
.58**
.25**
.18**
.37**
.27**
.34**
29**
.58**
.43**
.13*
.39**
.32**
.46**
Note. Table values are betas from regression analyses; significance levels are based on one-tailed tests. In
addition to the variables listed above, all analyses also included main effects of time and sex, the Time X Sex
interaction, and all two- and three-factor interactions of time and sex with each model variable,
tp < .10 (marginally significant). * p < .05.
**
p < .01.

956WIESELQUIST, RUSBULT, FOSTER, AND AGNEW
ficients declined for both variables (/3 = .13 and .13, both ns). On
balance, these findings suggest that our results for trust are not a
simple product of projection; that is, the association of trust with
perceived partner behavior is more than a reflection of individuals'
perceptions of their own behavior.
Residualized lagged mediation analyses. Table 5 summarizes
the results of residualized lagged mediation analyses that corre-
spond to Hypothesis 1 through 5. As noted earlier, all criteria
exhibited simple lagged associations with their presumed proximal
predictors—all simple lagged associations were significant (see
statistics under "Simple association" in Table 5; the presumed
proximal predictor is presented in the first row for each model).
And consistent with expectations, all criteria tended to exhibit
lagged associations with their presumed distal predictors—all 14
simple associations were significant or marginal (the presumed
distal predictor is presented in the second row for each model). In
all 14 cases, the simple association of the criterion with its pre-
sumed proximal predictor was descriptively larger than the asso-
ciation with its presumed distal predictor.
Unfortunately, when we performed simultaneous multiple re-
gression analyses examining residualized lagged associations,
most associations declined rather substantially for both presumed
proximal predictors and presumed distal predictors. The presumed
proximal predictors accounted for significant or marginal variance
in only 5 of 14 analyses; in no instance did the presumed distal
predictor account for even marginal variance above and beyond
the presumed proximal predictor. Once we took into account
variance attributable to the earlier measure of each criterion, there
was simply insufficient remaining variance to examine associa-
tions with other predictor variables (for the simple association of
the earlier and later measures of each criterion, average /3 = .80).
(Even in the residualized lagged simple associations presented
earlier [see Table 3], many presumed proximal predictors ac-
counted for only marginal change over time in the criteria.) These
findings are not inconsistent with our model, in that the presumed
distal predictors in no case accounted for greater change over time
in the criteria than did the presumed proximal predictors. At the
same time, these results are somewhat inconclusive, in that the
proximal predictors generally do not account for substantial vari-
ance above and beyond the distal predictors, presumably because
of insufficient change over time in the criteria.
Associations of Model Variables With Dyadic Adjustment
To provide additional evidence relevant to our model of mutual
cyclical growth, we examined the simple associations of all model
variables with dyadic adjustment.10 The concurrent analyses re-
vealed significant simple associations of all model variables with
dyadic adjustment, in both Study 1 (betas ranged from .20 to .64)
and Study 2 (betas ranged from .60 to .78; all ps < .05). Lagged
analyses revealed significant or marginal simple lagged associa-
tions of all earlier model variables with later dyadic adjustment in
both Study 1 (betas ranged from .23 to .66) and Study 2 (betas
ranged from .45 to .61; all ps < .10). However, in analyses
examining residualized lagged associations with dyadic adjust-
ment, that is, in analyses in which we regressed later adjustment
simultaneously onto earlier model variables along with earlier
adjustment, only four associations were significant or marginal: In
Study
1
earlier accommodation, earlier perceived accommodation,
and earlier trust predicted significant or marginal change over time
in adjustment, and in Study 2 earlier perceived partner accommo-
dation predicted significant change over time in adjustment (/3s =
.16,
.19, .21, and .16; all ps < .10). Although the residualized
lagged analyses are inconclusive, the results of concurrent and
simple lagged analyses provide moderately good evidence that
model variables such as dependence, commitment, pro-
relationship behavior, and trust are associated with healthy func-
tioning in ongoing close relationships.11
Associations of Model Variables With Attachment Style
In Study 2 we obtained measures of each partner's self-reported
tendencies toward secure attachment, anxious-ambivalent attach-
ment, and avoidant attachment. We performed a variety of con-
current analyses to examine the associations of attachment style
with model variables, the results of which are summarized in
Table 6. First, we examined the simple concurrent associations of
secure, anxious-ambivalent, and avoidant attachment with depen-
dence, commitment, accommodative behavior, perceived partner
accommodation, and trust, as well as with partner trust (see sta-
tistics under "Simple associations" in Table 6). For purposes of
comparison, we also present the simple association for each pre-
sumed proximal predictor from our interdependence-based model
(e.g., for the analysis predicting dependence, the presumed prox-
imal predictor was trust; see statistics under "Interdependence
predictor").
Most concurrent simple associations with attachment variables
were in the anticipated direction: Secure attachment tended to
exhibit positive associations with model variables, whereas asso-
ciations with anxious-ambivalent and avoidant attachment tended
to be negative. Out of a total of 18 concurrent links with the three
attachment variables, 10 associations were significant or marginal.
However, associations with the attachment variables tended to be
descriptively weak: Across the six regression analyses examining
10 The Dyadic Adjustment Scale assesses diverse aspects of couple
well-being, including intimacy, intent to persist, effective problem solving,
and positive affect. Accordingly, a few items in the instrument overlap with
key variables in our model. To ensure that our results were not influenced
by item overlap, we deleted key items from the instrument to develop (a)
a commitment-and-satisfaction-purged measure of adjustment and (b) a
conflict-purged measure of adjustment. Analyses performed using these
"purged" measures of adjustment parallel those reported below.
11 In support of the independence of the commitment and trust con-
structs, in concurrent analyses in which we regressed dyadic adjustment
simultaneously onto commitment and trust, both variables accounted for
unique variance in couple well-being (for commitment j8s = .29 and .31,
for trust /3s = .53 and .63, all ps < .05). We also performed concurrent
analyses in which we regressed adjustment simultaneously onto (a) own
accommodation and perceived partner accommodation and (b) own will-
ingness to sacrifice and perceived partner willingness to sacrifice. In
support of the independence of reports of one's own and perceptions of the
partner's pro-relationship acts, both own accommodation and perceived
partner accommodation accounted for unique variance in couple well-
being (for own accommodation j3s = .41 and .31, for perceived partner
accommodation /3s = .26 and .42, all ps < .05). In the Study 1 analyses
examining willingness to sacrifice, neither own sacrifice nor perceived
partner sacrifice accounted for significant unique variance in simultaneous
regression analyses (0s = .29 and —.05, both ns).

COMMITMENT AND TRUST957
simple associations, the models including secure attachment ac-
counted for an average of 6% of the variance in model variables,
models including anxious-ambivalent attachment accounted for an
average of 13% of the variance, and models including avoidant
attachment accounted for an average of 5% of the variance.
Next, we performed regression analyses examining the simul-
taneous associations of secure, anxious-ambivalent, and avoidant
attachment with each criterion (see statistics under "Three Attach-
ment Predictors" in Table 6). Anxious-ambivalent attachment
exhibited consistent negative associations with model variables,
accounting for significant unique variance in four of six regression
analyses; only one coefficient for secure attachment was marginal,
and only one coefficient for avoidant attachment was marginal.
Across the six regression analyses, the three attachment variables
collectively accounted for an average of 18% of the variance in
model variables. In contrast, parallel regression analyses including
just one interdependence predictor variable accounted for an av-
erage of 33% of the variance in model variables.
To determine whether attachment style accounted for unique
variance beyond interdependence variables, we performed regres-
sion analyses including four predictors, examining simultaneous
associations with secure, anxious-ambivalent, and avoidant attach-
ment, along with the appropriate proximal interdependence pre-
dictor (see statistics under "Interdependence predictor plus three
attachment predictors" in Table 6). Very few attachment coeffi-
cients were significant—two of six coefficients for anxious-
ambivalent attachment were significant, and one coefficient for
secure attachment was marginal. In contrast, all six coefficients for
the interdependence predictors were significant. Across the six
analyses, the four-factor models accounted for an average of 39%
of the variance in the criteria. Compared with regression models
including just one interdependence predictor, these four-factor
models accounted for an average of 6% additional explained
variance.
It is possible that the four-predictor regression models underes-
timate variance attributable to each attachment style, in that all
three attachment variables were included in each analysis. Accord-
ingly, we performed a final series of analyses, regressing each
criterion onto two predictor variables—the appropriate proximal
interdependence predictor along with just one attachment variable.
Analyses examining the interdependence predictor along with se-
cure attachment revealed that the coefficient for secure attachment
was significant in predicting accommodation (for the interdepen-
dence predictor and secure attachment, )3s = .37 and .28, both
ps < .05). These models accounted for an average of 34% of the
variance in model
criteria—1%
more than the models including
only the interdependence predictor. In analyses examining the
interdependence predictor along with anxious-ambivalent attach-
ment, the coefficient for anxious-ambivalent attachment was sig-
nificant in predicting accommodation (for the interdependence
predictor and anxious-ambivalent attachment, /3s = .33 and —.27)
and trust level (/3s = .53 and -.28, allps < .05). These models
accounted for an average of 37% of the variance in criteria—4%
more than the models including only the interdependence predic-
tor. In analyses examining the interdependence predictor along
with avoidant attachment, the coefficient for avoidant attachment
was marginal in predicting accommodation (for the interdepen-
dence predictor and avoidant attachment, j3s = .35 and
—.20,
both
ps < .10). These models accounted for an average of 34% of the
variance in
criteria—1%
more than the models including only the
interdependence predictor. Thus, attachment style typically does
not account for substantial variance beyond the features of inter-
dependence that reside at the heart of the present analysis. In
contrast—and consistent with expectations—variables from our
interdependence-based model consistently account for unique vari-
ance beyond an attachment-based model.
Discussion
Summary of Results
Findings from two longitudinal studies—one examining dating
relationships and one examining marital relationships—revealed
good support for an interdependence-based model of the within-
individual and across-partner associations among commitment,
pro-relationship behavior, and trust. Table 7 presents a summary of
findings regarding the concurrent and residualized lagged associ-
ations among contiguous model variables, Table 8 presents a
summary of findings from concurrent mediation analyses, and
Table 9 presents a summary of findings from residualized lagged
mediation analyses.
Simple associations. Analyses examining the concurrent sim-
ple associations between contiguous model variables revealed ex-
cellent support for our model: 15 of 15 analyses were consistent
with predictions (see statistics under "Concurrent association" in
Table 7). In parallel manner, analyses examining the simple lagged
associations between contiguous model variables also revealed
excellent support for our model: in 15 of 15 analyses, earlier
measures of presumed predictors exhibited significant associations
with later measures of presumed criteria (see statistics under
"Lagged association"). More importantly, residualized lagged
analyses revealed moderately good support for our model: in 11
of 15 analyses, we observed significant or marginal associations of
predictors with criteria when we examined change in criteria by
regressing later criteria onto earlier predictors, controlling for
earlier levels of each criterion (see statistics under "Residualized
lagged association"). Even when we limit ourselves to hypotheses
that are entirely new to the present research (i.e., the Commitment-
Trust Hypothesis, as well as Hypotheses 3, 4, and 5), our hypoth-
eses were supported in all 10 concurrent analyses and in 7 of 10
residualized lagged analyses. Results from the lagged analyses are
an important complement to the concurrent results, in that they
help us rule out alternative explanations of our findings. For
example, assuming that there is some day-to-day variability in
mood, it seems unlikely that mood effects could explain the
associations between earlier predictors and later criteria. Also, it
seems unlikely that self-report bias could explain our findings, in
that the residualized lagged analyses predict later criteria while
controlling for earlier levels of the criteria. Such stringent statis-
tical control arguably captures and controls for tendencies toward
consistency in responding, acquiescence, or tendencies toward
socially desirable responding.
Mediation analyses. Analyses examining first-order mediation
assessed the simultaneous associations of presumed proximal and
distal predictors with a given criterion. Preliminary concurrent
analyses revealed significant simple associations of predictor vari-
ables with their respective criteria for 13 of 13 proximal predictors
and for 13 of 14 distal predictors (see statistics under "Simple

958WIESELQUIST, RUSBULT, FOSTER, AND AGNEW
Table 5
Hypotheses 1 Through 5: Residualized Lagged Mediation Analyses
for All Model Variables, Studies 1 and 2
Simple
association Regression analysis
Variable B t 8 t
Hypothesis 1:
A's dependence level
—>
A's commitment level
A's later commitment level
Study 1
A's earlier dependence level
A's earlier trust level
A's earlier commitment level
Study 2
A's earlier dependence level
A's earlier trust level
A's earlier commitment level
.65
.50
.79
.63
.33
.73
5.35**
3.43**
8.70**
5.94**
2.68**
8.03**
.11
.10
.68
.21
-.08
.64
0.81
0.79
5.20**
1.53t
-0.69
4.73**
.68*
.59**
Hypothesis 2:
A's commitment level
—»
A's pro-relationship behavior
A's later accommodation
Study 1
A's earlier commitment level
A's earlier dependence level
A's earlier accommodation
Study 2
A's earlier commitment level
A's earlier dependence level
A's earlier accommodation
A's later willingness to sacrifice
Study 1
A's earlier commitment level
A's earlier dependence level
A's earlier willingness to sacrifice
45
40
72
35
33
84
50
46
83
2.94**
2.77**
6.66**
2.79**
2.50**
10.82**
3.38**
3.19**
8.15**
.09
.15
.63
-.01
-.08
.87
.10
.00
.76
0.50
0.86
4.75**
-0.09
-0.66
9.68**
0.62
0.02
5.74**
3.69**
2.11*
3.49**
4.00**
2.01*
7.11**
3.52**
2.54**
7.46**
.11
.03
.75
.16
-.03
.66
.20
.02
.73
0.89
0.27
6.54**
1.37t
-0.26
5.96**
1.55t
0.19
5.99**
.57**
.70**
.69**
Hypothesis 3:
A's pro-relationship behavior -> B's perceived pro-relationship behavior
B's later perceived accommodation
Study 1
A's earlier accommodation .52
A's earlier commitment level .33
B's earlier perceived accommodation .85
Study 2
A's earlier accommodation .48
A's earlier commitment level .27
B's earlier perceived accommodation .74
B's later perceived willingness to sacrifice
Study 1
A's earlier willingness to sacrifice .52
A's earlier commitment level .41
B's earlier perceived willingness to sacrifice .81
.67*
.57*
.69**
Hypothesis 4:
B's perceived pro-relationship behavior • • B's trust level
B's later trust level
Study 1
B's earlier perceived accommodation .71
A's earlier accommodation .40
B's earlier trust level .87
Study 2
B's earlier perceived accommodation .52
A's earlier accommodation .47
B's earlier trust level .76
B's later trust level
Study 1
B's earlier perceived willingness to sacrifice .35
A's earlier willingness to sacrifice .21
B's earlier trust level .87
5.65**
2.70**
9.95**
4.54**
3.85**
8.75**
2.15*
1.30t
9.95**
.16
.00
.77
.17
.05
.61
.08
.06
.82
1.14
0.00
5.87**
1.43t
0.36
4.91**
0.83
0.57
9.14**
.78*
.56*
.76*

COMMITMENT AND TRUST959
Table 5 (continued)
Simple
associationRegression analysis
Variable
Hypothesis 5:
B's trust level
—»
B's dependence level
B's later dependence level
Study 1
B's earlier trust level
B's earlier perceived accommodation
B's earlier dependence level
Study 2
B's earlier trust level
B's earlier perceived accommodation
B's earlier dependence level
B's later dependence level
Study 1
B's earlier trust level
B's earlier perceived willingness to sacrifice
B's earlier dependence level
60
45
84
36
20
73
60
37
84
4.11**
2.93**
9.41**
2
97**
1.53t
8.24**
4.11**
2.23*
9.41**
.11
.06
.76
-.13
.06
.80
.20
.07
.67
0.75
0.43
7.52**
-1.01
0.50
7.35**
1.57t
0.71
5.51**
.74*
.63**
.71*
Note. Table values are betas from regression analyses; significance levels are based on one-tailed tests. In
addition to the variables listed above, all analyses also included main effects of time and sex, the Time X Sex
interaction, and all two- and three-factor interactions of time and sex with each model variable.
•f
p < .10 (marginally significant). *p < .05.
**
p < .01.
associations" in Table 8 for both "Proximal predictor" and "Distal
predictor").12 In each instance, simple associations with criteria
were descriptively stronger for presumed proximal predictors than
for presumed distal predictors. When we regressed each criterion
simultaneously onto its presumed proximal and distal predictors,
we obtained excellent support for our model of mediation (see
statistics under "Mediation analyses" in Table 8 for both "Proxi-
mal predictor" and "Distal predictor"; we did not examine medi-
ation for the Commitment-Trust Hypothesis because this associ-
ation is not a direct link in our model): (a) In 14 of 14 analyses,
presumed distal predictors accounted for substantially reduced
variance beyond presumed proximal predictors (see "Distal pre-
dictor"; coefficients declined by an average of .27), whereas (b)
in 14 of 14 analyses, presumed proximal predictors accounted for
substantial variance beyond presumed distal predictors (see statis-
tics under "Proximal predictor"; coefficients declined by an aver-
age of only .07). Thus, the associations of presumed distal predic-
tors with criteria were rather thoroughly mediated by associations
with presumed proximal predictors. Consistent support for our
model was observed not only for hypotheses that have been
examined in previous research but also for hypotheses that are new
to the present research.
Preliminary lagged analyses revealed significant or marginal
simple associations of predictor variables with their respective
criteria for all 13 proximal predictors and for all 14 distal predic-
tors (see statistics under "Simple associations" in Table 9). In each
instance, simple lagged associations with criteria were descrip-
tively stronger for presumed proximal predictors than for pre-
sumed distal predictors. When we regressed later measures of each
criterion simultaneously onto its presumed proximal and distal
predictors, controlling for earlier levels of the criterion, we ob-
tained inconclusive evidence relevant to our model of mediation
(see statistics under "Mediation analyses" in Table 9): (a) In all 14
analyses, presumed distal predictors accounted for substantially
reduced variance beyond presumed proximal predictors (see "Dis-
tal predictor"), and (b) in only 5 of 14 analyses, presumed proxi-
mal predictors accounted for substantial variance beyond pre-
sumed distal predictors (see statistics under "Proximal predictor").
These results do not actively disconfirm our model, in that distal
predictors in no case accounted for substantial variance in the
presence of proximal predictors. At the same time, proximal pre-
dictors frequently did not account for substantial variance beyond
distal predictors. Presumably, these findings were weak because
residualized lagged mediation represents an enormously difficult
"test to pass"—once we control for the earlier level of each
criterion, there is insufficient remaining variance to obtain reliable
estimates of variance attributable to two additional predictor vari-
ables.
Indeed, it is somewhat remarkable that we observed mar-
ginally significant effects for presumed proximal variables in 5
of 14 analyses.
The evidence from mediation analyses—particularly the excel-
lent support revealed in concurrent mediation analyses—does
much to demonstrate the independence of our constructs, as well as
provide support for claims regarding the direct and indirect asso-
ciations among model variables. These results also help us rule out
alternative explanations of our findings. For example, it could be
12 For Study 1, two mediation analyses are relevant to Hypothesis 5: one
analysis in which accommodation is the presumed distal predictor of
dependence and a second analysis in which sacrifice is the presumed distal
predictor of dependence (see Tables 8 and 9, first and second rows under
"Hypothesis 5"). For both rows, the presumed proximal predictor is trust.
Accordingly, the simple association of trust with dependence is identical
for the first and second rows. For the second row the simple association is
identical to the association presented in the immediately preceding row.

960 WIESELQUIST, RUSBULT, FOSTER, AND AGNEW
Table 6
Predicting Each Model Variable From Attachment Style Variables:
Concurrent Analyses, Study 2
Predicting criteria
Simple associations
Dependence level
Commitment level
Accommodation
Perceived partner accommodation
Trust level
Partner's trust level
Three attachment predictors
Dependence level
Commitment level
Accommodation
Perceived partner accommodation
Trust level
Partner's trust level
Interdependence predictor plus three
attachment predictors
Dependence level
Commitment level
Accommodation
Perceived partner accommodation
Trust level
Partner's trust level
Interdependence
predictor
.53**
.73**
.39**
.55**
.62**
.46**
.46**
.73**
.32**
.56**
.52**
44**
Secure
.07
.07
.31**
.18t
.18t
.13
-.06
-.09
.20f
.07
.09
.02 •
-.11
-.05
.23t
-.02
.04
.06
Attachment style
Anxious-
ambivalent
-.33**
-.16
-.32**
-.34**
-.44**
-.17f
-.33**
-.15
-.27*
-.20*
_44**
-.16
-.13
.09
-.23*
.08
-.29**
-.08
Avoidant
-.13
-.18t
-.26*
-.15
-.15
-.16
-.12
-.21t
-.10
-.10
-.04
-.13
-.11
-.13
-.04
-.06
-.02
-.05
R2
.18
.12
.21
.19
.26
.10
.33
.59
.30
.34
.50
.26
Note. Table values are betas from regression analyses; significance levels are one-tailed tests. In addition to the
predictor variables listed above, all analyses also included main effects of time and sex, the Time X Sex
interaction, and all two- and three-factor interactions of time and sex with each predictor variable.
\p < .10 (marginally significant). *p < .05.
**
p < .01.
argued that the simple associations among model variables are
attributable to common method variance or to tendencies toward
socially desirable responding. Alternatively, it could be argued that
the simple associations are attributable to participants' tendencies
to describe well-functioning relationships in a generally favorable
light and to describe poorly functioning relationships in a dispar-
aging light. The fact that mediation analyses consistently revealed
stronger associations for presumed proximal causes than for pre-
sumed distal causes does much to enhance our confidence in the
specific model of cause-and-effect outlined in the introduction and
displayed in Figure 1. In the following paragraphs we address each
component of the model, reviewing findings relevant to each
hypothesis and discussing the broader implications of our results.
Commitment, Pro-Relationship Behavior, and Trust
Consistent with the Commitment-Trust Hypothesis, we ob-
tained very good support for the assertion that trust can be con-
strued as an implicit gauge of the strength of a partner's commit-
ment. Although the Figure 1 model does not include a direct link
between Partner A's commitment and Partner B's trust, we exam-
ined this association because it embodies a key assumption under-
lying our model: the claim that in relationships characterized by
strong interdependence, partners' experiences are inextricably
linked (Kelley & Thibaut, 1978; Rusbult & Van Lange, 1996).
Consistent with our interdependence analysis, the relationship-
specific variables that guide each individual's behavior appear to
be mirrored by complementary relationship-specific variables that
ultimately guide the partner's behavior. Moreover, it is striking
that although individual commitment and partner trust are strongly
linked in across-partner analyses, auxiliary analyses revealed that
these variables exhibit unique within-individual associations with
couple well-being (see Footnote 11); that is, although trust argu-
ably is a gauge of partner commitment, commitment and trust are
sufficiently distinct within each individual that they account for
independent variance in couple well-being.
Hypotheses 1 and 2 examined predictions that have received
support in prior research. Consistent with Hypothesis 1, we ob-
served very good support for the frequently tested prediction that
dependence on a relationship (operationally defined in the present
work as the "sum total" of high satisfaction, poor alternatives, and
high investments) is associated with strong commitment (e.g., Bui
et al., 1996; Felmlee et al., 1990; Rusbult, 1983; Rusbult, Martz, &
Agnew, 1998). Extending prior research relevant to Hypothesis 1,
the present research examined the role of dependence in mediating
the association of presumed distal variables with commitment
level. Consistent with expectations, we observed evidence consis-
tent with the claim that strong trust (the presumed distal predictor)
does not directly translate into enhanced commitment—trust argu-
ably yields enhanced commitment largely insofar as strong trust
produces enhanced dependence, which in turn strengthens com-
mitment. Moreover, and in accord with Hypothesis 2, we obtained
moderate support for the previously tested prediction that the

COMMITMENT AND TRUST961
Table 7
Summary of Concurrent and Lagged Simple Associations,
Studies 1 and 2
p values
Residualized
AnalysesConcurrent
associationLagged
associationassociation
Commitment-trust hypothesis:
A's commitment level
—>
B's trust level
Study 1
Study 2 .05
.05.05
.05.10
.10
Hypothesis 1:
A's dependence level
—»
A's commitment level
(supported in previous research)
Study 1
Study 2 .05
.05.05
.05.10
.10
Hypothesis 2:
A's commitment level
—»
A's pro-relationship behavior
(supported in previous research)
Study 1 accommodation .05 .05 .05
Study 1 sacrifice .05 .05 .10
Study 2 accommodation .05 .05 ns
Hypothesis 3:
A's pro-relationship behavior
—»
B's perceived pro-relationship behavior
Study 1 accommodation .05 .05
Study 1 sacrifice .05 .05
Study 2 accommodation .05 .05
.10
.05
.10
Hypothesis 4:
B's perceived pro-relationship behavior • •B's trust level
Study 1 accommodation .05 .05
Study 1 sacrifice .05 .05
Study 2 accommodation .05 .05
.10
ns
ns
Hypothesis 5:
B's trust level
—»
B's dependence level
Study I
Study 2 .05
.05.05
.05.10
ns
willingness to depart from one's immediate self-interest for the
good of a relationship is predicted by strong commitment (e.g.,
Arriaga & Rusbult, 1998; Rusbult, Bissonnette, et al, 1998; Rus-
bult et al., 1991; Van Lange et al., 1997). Importantly, in the
present research we used two operational definitions of pro-
relationship behavior, examining the associations of commitment
with both accommodative behavior and willingness to sacrifice.
In support of Hypothesis 3, we observed very good evidence
that partners perceive one another's pro-relationship acts in a
manner that to some extent agrees with each person's self-report.
A social constructionist might argue that such findings reflect
partners' shared understandings regarding their relationship and
that across-partner agreement does not mean that such perceptions
necessarily are veridical (cf. Gergen, 1985). However, our findings
were relatively construct specific; for example, Partner A's
self-
reported accommodation tended to exhibit specific links with
Partner B's perceived partner accommodation, not with other
model variables. Thus, a social constructionist would need to argue
that partners develop quite specific "stories" about their relation-
ships and that these stories align with those identified in our model.
In light of previous demonstrations that self-reported commitment,
accommodation, and sacrifice exhibit relatively good agreement
with direct behavioral measures (e.g., Drigotas & Rusbult, 1992;
Rusbult et al., 1991; Van Lange et al., 1997), we think the most
parsimonious interpretation of our findings is that interdependence
structure to some degree is "real," and that individuals are reason-
ably in touch with circumstances of interdependence in their
relationships. Interdependence structure and processes would
seem to constitute more than simple "in the head" perceptual
phenomena.
In accord with Hypothesis 4, we obtained moderately good
evidence that perceived partner accommodation and sacrifice are
associated with enhanced trust. These findings are compatible with
the assertion that when a partner is seen to place the welfare of the
relationship above his or her immediate self-interest, individuals
come to feel more trusting of the partner (cf. Holmes & Rempel,
1989).
According to this analysis, trust is a function of three
elements: the individual, the partner, and the situation. Diagnostic
situations provide individuals with opportunities to display their
pro-relationship motives and serve as tests by which partners may
form inferences regarding the individual's pro-relationship mo-
tives.
Thus, if a relationship provides for few (or no) diagnostic
situations in which individuals may display their pro-relationship
motives, trust cannot develop. If individuals "fail the test" when
confronted with diagnostic situations—that is, if individuals pur-
sue their immediate self-interest rather than exhibiting pro-
relationship motives and behavior—trust cannot develop. And if
partners fail to perceive the individual's pro-relationship motives
and behavior, trust cannot develop.
Is it possible that findings relevant to Hypothesis 4 reflect
"self-processes" rather than interdependence processes? For exam-
ple,
is it possible that trust results from projection, such that (a)
perceptions of the partner's behavior reflect perceptions of one's
own behavior, and (b) the association of perceived partner behav-
ior with trust is an artifact of the link between own behavior and
trust? Auxiliary analyses revealed that although both own accom-
modation and perceived partner accommodation exhibit simple
associations with trust, in simultaneous concurrent analyses (a)
perceived partner accommodation accounts for unique variance in
trust, whereas (b) own accommodation accounts for substantially
reduced variance in trust. (Parallel analyses performed for mea-
sures of own sacrifice and perceived partner sacrifice revealed
inconclusive results.) These findings help demonstrate the inde-
pendence of model variables, revealing a pattern of findings that is
compatible with an interdependence-based analysis regarding the
proximal and distal causes of trust.
Consistent with Hypothesis 5, we obtained good evidence that
trust is positively associated with dependence. Our findings re-
garding associations with trust are compatible with the notion that
strong trust frees individuals from anxiety regarding their relation-
ships,
allowing them to risk increased dependence on a partner (cf.
Holmes & Rempel, 1989). Indeed, it may be useful to conceptu-
alize both trust and commitment as aspects of "relationship regu-
lation." Our model suggests that as individuals become increas-
ingly dependent they develop strong commitment, which in turn
yields enhanced willingness to engage in pro-relationship behav-

962WIESELQUIST, RUSBULT, FOSTER, AND AGNEW
Table 8
Summary of Concurrent Mediation Analyses, Studies 1 and 2
p values
Simple associations Mediation analyses
AnalysesProximal Distal Proximal Distal
predictor predictor predictor predictor
Hypothesis 1:
A's dependence level -> A's commitment level
(supported in previous research)
Study 1
Study 2 .05
.05.05
.05.05
.05ns
.10
.05
.05
.05
.05
.05
.05
.05
.10
.10
ns
ns
ns
.05
.05
.05
.05
.05
.05
.05
.05
.05
ns
ns
ns
Hypothesis 2:
A's commitment level
—>
A's pro-relationship behavior
(supported in previous research)
Study 1 accommodation
Study 1 sacrifice
Study 2 accommodation
Hypothesis 3:
A's pro-relationship behavior
—»
B's perceived pro-relationship behavior
Study 1 accommodation
Study 1 sacrifice
Study 2 accommodation
Hypothesis 4:
B's perceived pro-relationship behavior
—>
B's trust level
Study 1 accommodation .05 .05 .05 ns
Study 1 sacrifice .05 ns .10 ns
Study 2 accommodation .05 .05 .05 ns
Hypothesis 5:
B's trust level -» B's dependence level
Study 1 accommodation
as distal predictor .05 .05 .05 ns
Study 1 sacrifice as
distal predictor Identical3 .05 .05 .10
Study 2 accommodation
as distal predictor .05 .05 .05 ns
3 The simple association is identical to the association presented in the
immediately preceding row.
iors.
When partners perceive such pro-relationship acts they de-
velop enhanced trust, which in turn leads them to become increas-
ingly dependent—increasingly satisfied, willing to drive away or
derogate alternatives, and willing to invest in a relationship in
material and nonmaterial ways. This brings us full circle, in that
enhanced dependence is argued to yield increased commitment.
Thus,
relationships to some degree may be internally regulated:
Via the process of adaptation to evolving patterns of interdepen-
dence, changes in each person's actions and motives trigger com-
plementary changes in the partner. Adaptations such as these
reside at the heart of an interdependence analysis (cf. Holmes,
1981;
Kelley, 1983b; Rusbult & Van Lange, 1996).
Compatible with our model of mutual cyclical growth, we
expected to observe positive associations of all model variables
with couple well-being. Concurrent analyses revealed associations
of all model variables with dyadic adjustment; residualized lagged
analyses revealed several simple associations with change in ad-
justment. These results are particularly meaningful in combination
with earlier-reported findings: The fact that Partner A's commit-
ment is associated with Partner B's trust is compatible with our
general model. The support for Hypotheses 1 through 5 that we
observed in concurrent and residualized lagged simple analyses
and mediation analyses provides further support for this model.
That model variables are also positively associated with couple
well-being represents yet another piece of evidence in a network of
findings that is congruent with our model. Although it is impos-
sible to obtain definitive evidence in studies using nonexperimen-
tal methods, these findings are compatible with our analysis of the
means by which close partners adapt to changing circumstances in
their relationship—a pattern whereby shifts in each person's ac-
tions and motives yield predictable shifts in the actions and mo-
tives of the partner, moving couples toward enhanced (or decayed)
couple functioning.
Table 9
Summary of Residualized Lagged Mediation Analyses,
Studies 1 and 2
p values
Simple associations Mediation analyses
AnalysesProximal Distal Proximal Distal
predictor predictor predictor predictor
Hypothesis 1:
A's dependence level
—»
A's commitment level
(supported in previous research)
Study 1
Study 2 .05
.05.05
.05ns
.10ns
ns
Hypothesis 2:
A's commitment level
—>
A's pro-relationship behavior
(supported in previous research)
Study 1 accommodation
Study 1 sacrifice
Study 2 accommodation
.05
.05
.05
.05
.05
.05
ns
ns
ns
ns
ns
ns
.05
.05
.05
.05
.10
.05
ns
ns
.10
ns
ns
ns
Hypothesis 3:
A's pro-relationship behavior
—»
B's perceived pro-relationship behavior
Study 1 accommodation .05 .05 ns ns
Study 1 sacrifice .05 .05 .10 ns
Study 2 accommodation .05 .05 .10 ns
Hypothesis 4:
B's perceived pro-relationship behavior
—>
B's trust level
Study 1 accommodation
Study 1 sacrifice
Study 2 accommodation
Hypothesis 5:
B's trust level
—»
B's dependence level
Study 1 accommodation
as distal predictor .05 .05 ns ns
Study 1 sacrifice as
distal predictor Identical3 .05 .10 ns
Study 2 accommodation
as distal predictor .05 .10 ns ns
aThe simple association is identical to the association presented in the
immediately preceding row.

COMMITMENT AND TRUST963
Attachment Style and Interdependence Structure
A priori, we anticipated that attachment style might play a role
in shaping trust and other model variables. In the introduction we
advanced an interdependence-based analysis of trust—an analysis
providing ample room for influence by personal dispositions.
Humans do not enter into relationships as tabula rasa; prior inter-
dependence history is assumed to play a role in shaping current
transformational tendencies. Moreover, the concept of felt security
resides at the heart of attachment theory, and all extant empirical
evidence points to the conclusion that mental models of attachment
should exert meaningful effects on the capacity to trust a close
partner (for reviews, see Hazan & Shaver, 1994; Reis & Patrick,
1996).
However, analyses of the Study 2 data revealed that com-
pared with interdependence structure and processes, attachment
style accounts for relatively little of the variance in trust and
related phenomena.
How should we interpret the relatively weak contributions of
attachment style compared with interdependence variables? First,
it is possible that our attachment measures were less reliable than
our measures of interdependence variables. Although many recent
studies examining attachment phenomena have used multiple-item
indexes, in the present research we measured secure, anxious-
ambivalent, and avoidant attachment with one item each. In con-
trast, we measured interdependence variables with multiple-item
instruments. Second, it is possible that the relatively weak results
for attachment variables result from the fact that our findings rest
on self-report. Perhaps it is easier to accurately describe features of
interdependence than it is to describe generalized attachment style.
Third, it is possible that well into adulthood, close partners have
the capacity to change one another's dispositions. Indeed, involve-
ment in a healthy close relationship may eliminate preexisting
differences in attachment style. Participants in Study 2 had been
married for 2 to 3 years. It is possible that experiences during the
early years of involvement "healed" anxious-ambivalent and
avoidant persons, reducing or eliminating individual differences in
attachment style. However, the sizeable test-retest correlations for
attachment style (rs ranged from .47 to .61) would seem to argue
against this line of reasoning, in that (a) attachment style exhibits
at least moderate stability over time, and (b) if sufficient variability
in attachment style existed to reveal significant test-retest corre-
lations, there would appear to be sufficient variability to reveal
significant correlations with other variables.
More generally, it is important to note that attachment style does
exhibit simple associations with some aspects of interdependence.
Secure attachment was significantly or marginally positively as-
sociated with three of five model variables, anxious-ambivalent
attachment was negatively associated with all five variables, and
avoidant attachment was negatively associated with two model
variables. Notably, anxious-ambivalent attachment was (margin-
ally) negatively associated with partner trust; that is, when one
person in a relationship is anxious-ambivalent, the partner finds it
difficult to experience strong trust. Moreover, there were instances
in which attachment variables accounted for unique variance be-
yond interdependence variables—secure attachment exhibited a
positive unique association with accommodation, and anxious-
ambivalent attachment exhibited negative unique associations with
accommodation and trust.
Thus,
our findings regarding attachment style are not notable for
the fact that they are null; our attachment findings are notable in
that they are weak in comparison to links among interdependence
variables. It is not that the emperor is wearing no clothes; he is
simply wearing fewer clothes than we might have imagined. Ac-
knowledging the caveats noted earlier, it seems suitable to (tenta-
tively) conclude that attachment style may play a less important
role than interdependence structure in shaping such phenomena as
the willingness to become dependent on and committed to a
relationship, the tendency to engage in pro-relationship acts, and
the emergence of trust. A relationship is constituted of more than
the personal histories of the two partners: What transpires in an
ongoing relationship has much to do with the interdependence
history the partners create together.
Broader Implications
Transformation of
motivation.
Before closing we should com-
ment on several broader implications of the present findings. One
implication concerns the concept of transformation of motivation.
Interdependence theory highlights the importance of actual and per-
ceived disparities between the given situation (preferences based on
immediate self-interest) and the effective situation (preferences that
directly guide behavior). A positive disparity between (a) given
pref-
erences and (b) effective preferences is argued to result from pro-
relationship transformation (Kelley & Thibaut, 1978). A positive
disparity between (a) what is perceived to be in a partner's immediate
self-interest and (b) the partner's actual behavior is argued to serve as
the basis for attributional activity, or the process by which individuals
infer the existence of others' pro-relationship motives and dispositions
(Kelley, 1979). Such positive disparities are argued to yield consid-
erable positive affect, in that the partner's behavior is not only re-
warding in itself but possesses the "added value" of providing a basis
for inferring a partner's pro-relationship intentions, motives, and
dispositions (Kelley, 1984).
In the context of the present work, commitment is argued to
promote pro-relationship disparities between given and effective
preferences (Rusbult et al., 1994), and trust is argued to reflect
perceived pro-relationship disparities between given and effective
partner preferences (Holmes & Rempel, 1989). The measures of
own pro-relationship behavior and perceived partner pro-
relationship behavior that we used in the present work (i.e., ac-
commodation and willingness to sacrifice) are assumed to reflect
pro-relationship transformation of motivation. But such measures
could also be argued to reflect the simple positivity of partners'
behavior during conflicted interaction. To the extent that accom-
modation and willingness to sacrifice are construed as simple
indexes of positivity, it could be argued that our findings have little
or no relevance for understanding transformation of motivation.
That is, it could be argued that strong commitment promotes
positive behavior during conflict, not departures from one's im-
mediate self-interest. And it could be argued that individuals come
to trust their partners simply because they observe their partners
behave in a positive manner, not because they perceive that their
partners have sacrificed personal well-being for the good of the
relationship. As noted earlier, previous studies have revealed find-
ings in support of the claim that accommodation does result from
pro-relationship transformation (e.g., Rusbult et al., 1991; Yove-
tich & Rusbult, 1994). Nevertheless, in future research it will be

964WIESELQUIST, RUSBULT, FOSTER, AND AGNEW
important to provide direct evidence that the association of com-
mitment with one's own pro-relationship acts reflects pro-
relationship transformation and that the association of perceived
partner behavior with trust reflects the inference of partner pro-
relationship transformation.
Reciprocity, equality, and mutuality. A second implication of
the present work concerns the extant literature regarding equity
and mutuality in ongoing relationships. Our model of mutual
cyclical growth represents an ideal process through which equality of
dependence, mutuality of commitment, and reciprocity of pro-
relationship acts may be sustained. Our model suggests that (a) each
person's dependence produces strengthened commitment, (b) each
person's commitment yields enhanced tendencies toward pro-
relationship acts, (c) each person's pro-relationship acts are perceived
by the partner, (d) the perception of a partner's pro-relationship acts
strengthens trust, and (e) each person's trust yields enhanced depen-
dence. Over time with a partner, this steady cyclical pattern will
function in such a manner as to yield equal vulnerability, mutual
commitment, and equivalent acts of good will.
As noted earlier, this mutual cyclical pattern could have consider-
able functional value in the context of a well-functioning relationship.
Indeed, we suspect that our model may explain general tendencies
toward reciprocity, equality, and mutuality better than equity models,
which tend to assume rather conscious attention to matters of fairness
and equality (e.g., Hatfield, Traupmann, Sprecher, Utne, & Hay,
1985;
Sprecher,
1986).
Moreover, the forms of growth observed in the
present work cannot be explained on the basis of equity per se, in that
according to equity theory, (a) a relationship in which partners exhibit
equally low levels of dependence, commitment, pro-relationship be-
havior, and trust should function as well as (b) a relationship in which
partners exhibit equally high levels of each construct. To use the
technical language of interdependence theory, there is an important
difference between MinDiff transformation (i.e., seeking to minimize
the discrepancy between partners' well-being) and MaxJoint transfor-
mation (i.e., seeking to maximize joint well-being; Kelley & Thibaut,
1978).
Relationship growth requires more than 50-50 reciprocity—
partners must sometimes be willing to give more than the other
partner seems to be giving, and they must sometimes take it on faith
that the partner enacts as many pro-relationship behaviors as they
themselves enact (cf. Drigotas, Whitney, & Rusbult, 1995). In future
research it would be interesting to explore whether the associations of
equity with couple well-being that have been observed in previous
research may be partially attributable to interdependence processes of
the sort identified in the present work.
Mutual cyclical growth as a closed system. A third implication
of the present work rests on the fact that our model is represented as
a closed system. So long as each person continues to move in positive
directions (e.g., so long as each person continues to behave in a
generous, giving, and forgiving manner) the system moves forward,
feeding
itself.
But what happens if the system becomes stalled or
moves toward entropy? What sort of infusion of energy is needed to
redirect the system toward positive functioning? One solution may
reside in a process we term the "Enchanted April phenomenon." In
the novel Enchanted April, Lotty and Mellersh initially are in entropy.
Mellersh "counts" daily activities—Lotty must keep a record of all
expenses and is criticized for frivolous expenditures. Importantly,
both Mellersh and Lotty "count" couple activities, keeping a close eye
on departures from self-interest: He will not accommodate unless she
has done so; she will not sacrifice unless she knows he will credit her
for doing so; both refuse to make themselves vulnerable unless they
are sure of the other's probable reciprocity. Aside from the fact that
such vigilant record keeping in itself may be problematic (e.g., record
keeping communicates distrust; cf. Clark & Mills, 1979; Holmes,
1981),
movement toward growth may require an act of faith on the
part of one or both persons. As is evident in the bargaining literature,
sometimes the cycle of mutual growth must be initiated or renewed by
a unilateral gesture of good will—by a no-strings-attached pro-
relationship act (cf. Osgood, 1962; Pruitt & Rubin, 1986). Lotty
makes such a gesture, inviting Mellersh to vacation with her in
Portofino. Mellersh responds favorably, and in short order the two are
bathing together, caring for one another, and behaving in a loving and
generous manner. In future research it would be interesting to directly
examine exceptional gestures of good will, studying departures from
self-interest that extend well beyond the "everyday" accommodations
and acts of sacrifice that were examined in the present work (e.g.,
willingness to forgive a partner's act of betrayal).
More generally, it is important to note that although we have
represented our model as a closed system, in point of fact, many of the
variables in our model may well be influenced by exogenous vari-
ables.
For example, factors other than trust presumably affect indi-
viduals' decisions to become increasingly dependent by investing
numerous or sizeable resources in their relationships (e.g., the fore-
casting of future benefits). Also, factors other than commitment may
induce individuals to engage in exceptional pro-relationship acts. For
example, pro-relationship acts such as forgiveness of betrayal may
rest on factors extending beyond that which is strictly dictated by
preceding events in the relationship (e.g., personal dispositions may
contribute to the motivation and ability to forgive the partner and
move on, relegating a betrayal to the past). Future research should
include detailed analyses of each of the specific links in our model,
seeking to identify relevant exogenous variables.
Limitations and Directions for Future Work
It is important to briefly comment on the most obvious limita-
tions of this work. First, our findings rest on self-report measures.
Prior research has revealed evidence for the validity of such
measures. For example, self-reported accommodation is associated
with (a) partner reports of individual accommodation, (b) measures
of accommodation obtained by coding audiotaped or videotaped
conversations, and (c) measures of accommodation obtained in
laboratory tasks assessing conciliatory behavior (e.g., behavior in
prisoner's dilemma-type games; Rusbult, Bissonnette, et al., 1998;
Rusbult et al., 1991). Self-reported sacrifice is associated with (a)
partner reports of individual sacrifice and (b) willingness to step up
and down a stairstep to earn money for the partner (Van Lange et
al.,
1997). And commitment level is associated with probability of
persisting in a relationship (Drigotas & Rusbult, 1992; Rusbult,
1983;
Rusbult, Martz, & Agnew, 1998). In the present research we
obtained good agreement between self-reports of our constructs
and partner reports of parallel constructs. As noted earlier, these
findings suggest that there is a reality to interdependence structure
that both partners can perceive and describe. Nevertheless, the fact
remains that the present findings rest on self-report. In future work
it would be interesting to (a) obtain parallel reports of model
variables from individuals, their partners, and from "observer
friends" (i.e., from outside observers who are in a position to report
on key model variables) and (b) obtain behavioral measures of

COMMITMENT
AND
TRUST965
model variables. Moreover,
in
future research
it
would
be
inter-
esting
to
determine whether there
is
meaningful variability
in
partners' perceptions
of
interdependence structure.
For
example,
is
across-partner agreement influenced
by
perspective taking,
and
does degree
of
across-partner agreement influence overall quality
of couple functioning
(cf.
Arriaga
&
Rusbult, 1998)?
A second limitation
of
the present work
is
that
our
conclusions
rest
on
correlational analyses.
The
significant findings observed
in
many
of the
residualized lagged analyses
are
congruent with
the
assumption that
the
observed associations among variables may
be
causal
in
nature. Also,
in
light
of the
character
of our
mutual
cyclical growth model,
we are not so
much concerned with direc-
tion
of
causation: Ultimately, "causes"
are
also "effects"
in
this
model.
But
even
if we
feel somewhat comfortable living with
ambiguity regarding direction
of
causation, third-variable prob-
lems plague correlational research.
The
findings observed
in me-
diation analyses help
to
allay such concerns,
but it is
nevertheless
the case that
our
findings
are
correlational.
In
future work
it
will
be important to test critical hypotheses using both nonexperimental
and experimental methods.
In
addition,
if
we
are to
generalize
the
present findings beyond the domain of close relationships
it
will
be
important to examine key processes
in
the context of both romantic
and nonromantic involvements.
At
present
we are
conducting
experimental research relevant
to
this model, examining stranger
interactions
in a
prisoner's dilemma paradigm
to
test
the
predic-
tions that
(a) an
extended history
of
partner pro-relationship
be-
havior yields enhanced trust
and (b)
trusting individuals
are
more
likely
to
react
in a
benevolent manner
to
later acts
of
partner
betrayal (Hannon, Childs,
&
Rusbult, 1999).
A third limitation
of the
present work
is
that although
we
obtained nearly perfect support
for
model predictions
in
concurrent
analyses
(see
Tables
7 and 8),
findings from residualized lagged
analyses were somewhat weaker
(see
Tables
7 and 9).
Many
residualized lagged simple associations were only marginally
sig-
nificant,
and in
residualized lagged mediation analyses, presumed
proximal predictors accounted
for
substantial unique variance
in
fewer than half
of
our analyses. (At the same time, presumed distal
predictors consistently accounted
for
substantially reduced vari-
ance.)
We
have suggested that
the
weak residualized lagged
me-
diation findings
may be
attributable
to
insufficient change over
time
in our
criteria. Although
it is not
possible
to
definitively
describe what constitutes "adequate change,"
we
think
it is
plau-
sible that many
of our
nonsignificant findings
may
have emerged
because
the
residualized lagged criterion
did not
exhibit adequate
variance. Indeed,
in the
residualized lagged analyses
in
which
associations with presumed proximal predictors were weak,
the
associations
of
earlier criteria with later criteria were substantial
(betas ranged from
.72 to .87). In
future longitudinal research,
researchers should take particular care
to
maximize
the
odds
of
attaining adequate variance
in
residualized lagged criteria (i.e.,
by
recruiting samples
in
such
a
manner that
the
probability
of
persis-
tence
vs.
breakup approaches 50-50,
or the
distribution
of
move-
ment toward vitality
vs.
deterioration
is
relatively normal).
Conclusion
The present research brought together
two
previously unrelated
literatures, combining concepts identified
by
Rusbult
and her col-
leagues
in
their analysis
of
commitment processes (Rusbult
et al.,
1994) with concepts identified
by
Holmes
and
Rempel
in
their anal-
ysis
of
trust (1989). Given that both
of
these theoretical analyses rest
on concepts from interdependence theory (Kelley
&
Thibaut,
1978;
Thibaut
&
Kelley,
1959) and
given that both analyses concern
the
implications
of
dependence
and the
origins
of
pro-relationship orien-
tation,
the
analyses
are
highly complementary.
The
results
of two
longitudinal studies provide good support
for our
model, which
sug-
gests that individuals come
to
trust their partners
as a
consequence
of
observing the partner depart from his
or
her direct self-interest
for
the
good
of the
relationship. Commitment-inspired acts such
as
accom-
modation
and
sacrifice
are
argued
to
provide particularly diagnostic
information regarding the strength
of
a partner's inclination to behave
in a generous and giving manner.
It
is
notable that the present analysis
is both
(a)
dyadic,
in
that
it
rests
on the
interaction between partners,
and
(b)
dynamic,
in
that
it
represents
a
process that unfolds over
the
course
of
extended interaction.
Our
model
of
mutual cyclical growth
received good support across
a
variety
of
analyses, suggesting that
such a model may account for continued health and vitality in ongoing
relationships. Thus, interdependence theory propositions regarding
interpersonal motivation
and
behavior provide
a
sound basis
for
understanding important processes
in
ongoing close relationships.
References
Agnew,
C. R., Van
Lange,
P. A. M.,
Rusbult,
C. E.,
& Langston,
C. A.
(1998).
Cognitive interdependence: Commitment and the mental repre-
sentation
of
close relationships. Journal
of
Personality
and
Social
Psy-
chology,
74,
939-954.
Aron,
A., &
Aron,
E. N.
(1997). Self-expansion motivation
and
including
other
in the self. In S.
Duck (Ed.), Handbook of personal relationships:
Theory, research,
and
interventions
(2nd ed., pp.
251—270). Chichester,
Great Britain: Wiley.
Arriaga,
X. B., &
Rusbult,
C. E.
(1998). Standing
in my
partner's shoes:
Partner perspective-taking
and
reactions
to
accommodative dilemmas.
Personality
and
Social Psychology Bulletin,
9,
927-948.
Axelrod,
R.
(1984). The evolution of cooperation. New York: Basic Books.
Baron,
R. M., &
Kenny,
D. A.
(1986).
The
moderator-mediator variable
distinction
in
social psychological research: Conceptual, strategic,
and
statistical considerations. Journal
of
Personality
and
Social Psychol-
ogy,
51,
1173-1182.
Bissonnette,
V. L.,
Rusbult,
C. E., &
Kilpatrick,
S. D.
(1997). Empathic
accuracy
and
marital conflict resolution.
In W.
Ickes (Ed.), Empathic
accuracy
(pp.
251-281).
New
York: Guilford.
Bui,
K. T.,
Peplau,
L. A., &
Hill,
C. T.
(1996). Testing
the
Rusbult model
of
relationship commitment
and
stability
in a
15-year study
of
heterosexual
couples. Personality and Social Psychology
Bulletin,
22,
1244-1257.
Clark,
M. S., &
Mills,
J.
(1979). Interpersonal attraction
in
exchange
and
communal relationships. Journal
of
Personality
and
Social Psychol-
ogy,
37,
12-24.
Drigotas,
S. M., &
Rusbult,
C. E.
(1992). Should
I
stay
or
should
I
go?:
A
dependence model
of
breakups. Journal
of
Personality
and
Social
Psy-
chology,
62,
62-87.
Drigotas,
S. M.,
Rusbult,
C. E., &
Verette,
J. (in
press). Level
of
commit-
ment, mutuality
of
commitment,
and
couple well-being. Personal Rela-
tionships.
Drigotas,
S. M.,
Rusbult,
C. E.,
Wieselquist,
J., &
Whitton,
S.
(1999).
Close partner as sculptor
of
the
ideal
self:
Behavioral affirmation
and the
Michelangelo phenomenon. Journal of Personality
and
Social Psychol-
ogy,
77,
293-323.
Drigotas,
S. M.,
Whitney,
G. A., &
Rusbult,
C. E.
(1995).
On the
pecu-
liarities
of
loyalty:
A
diary study
of
responses
to
dissatisfaction
in
everyday life. Personality and Social Psychology Bulletin, 21, 596-609.

966WIESELQUIST, RUSBULT, FOSTER, AND AGNEW
Felmlee, D., Sprecher, S., & Bassin, E. (1990). The dissolution of intimate
relationships: A hazard model. Social Psychology Quarterly, 53, 13-30.
Gaines, S. O., Jr., Reis, H. T., Summers, S., Rusbult, C. E., Cox, C. L.,
Wexler, M. O., Marelich, W. D., & Kurland, G. J. (1997). Impact of
attachment style on reactions to accommodative dilemmas in close
relationships. Personal Relationships, 4, 93-113.
Gergen, K. J. (1985). The social constructionist movement in modern
psychology. American Psychologist, 40, 266-275.
Hannon, P. A., Childs, N. M., & Rusbult, C. E. (1999). The impact of
trust
level on
willingness
to forgive partner acts of
betrayal.
Unpublished manu-
script, University of North Carolina at Chapel Hill, Chapel Hill, NC.
Hatfield, E., Traupmann, J., Sprecher, S., Utne, M., & Hay, J. (1985).
Equity and intimate relations: Recent research. In W. Ickes (Ed.), Com-
patible and incompatible relationships (pp. 91—117). New York:
Springer-Verlag.
Hazan, C, & Shaver, P. (1987). Romantic love conceptualized as an
attachment process. Journal of Personality and Social Psychology, 52,
511-524.
Hazan, C, & Shaver, P. R. (1994). Attachment as an organizational
framework for research on close relationships. Psychological Inquiry, 5,
1-22.
Hinde, R. A. (1979). Towards understanding relationships. London: Ac-
ademic Press.
Holmes, J. G. (1981). The exchange process in close relationships: Mi-
crobehavior and macromotives. In M. J. Lemer & S. C. Lerner (Eds.),
The justice motive in social behavior
(pp.
261-284). New York: Plenum.
Holmes, J. G. (1989). Trust and the appraisal process in close relationships.
In W. H. Jones & D. Perlman (Eds.), Advances
in
personal relationships
(Vol. 2, pp. 57-104). London: Jessica Kingsley.
Holmes, J. G., & Rempel, J. K. (1989). Trust in close relationships. In C.
Hendrick (Ed.), Review of personality and social psychology (Vol. 10,
pp.
187-220). London: Sage.
Johnson, D. J., & Rusbult, C. E. (1989). Resisting temptation: Devaluation of
alternative partners as a means of maintaining commitment in close rela-
tionships. Journal of Personality and Social
Psychology,
57, 967-980.
Johnson, M. P. (1991). Commitment to personal relationships. In W. H.
Jones & D. W. Perlman (Eds.), Advances in personal relationships
(Vol. 3, pp. 117-143). London: Jessica Kingsley.
Kelley, H. H. (1979). Personal relationships: Their structures and pro-
cesses. Hillsdale, NJ: Erlbaum.
Kelley, H. H. (1983a). Love and commitment. In H. H. Kelley, E. Ber-
scheid, A. Christensen, J. H. Harvey, T. L. Huston, G. Levinger, E.
McClintock, L. A. Peplau, & D. R. Peterson (Eds.), Close relationships
(pp.
265-314). New York: W. H. Freeman.
Kelley, H. H. (1983b). The situational origins of human tendencies: A
further reason for the formal analysis of structures. Personality and
Social Psychology Bulletin, 9, 8-30.
Kelley, H. H. (1984). Affect in interpersonal relations. In P. Shaver (Ed.),
Review of personality and social psychology (Vol. 5, pp. 89-115).
Newbury Park, CA: Sage.
Kelley, H. H., & Thibaut, J. W. (1978). Interpersonal relationships: A
theory of interdependence. New York: Wiley.
Levinger, G. (1979). A social exchange view on the dissolution of pair
relationships. In R. L. Burgess & T. L. Huston (Eds.), Social exchange
in developing relationships (pp. 169-193). New York: Academic Press.
McClintock, C. G. (1972). Social motivation—a set of propositions. Be-
havioral Science, 17, 438-454.
Miller, R. S. (1997). Inattentive and contented: Relationship commitment
and attention to alternatives. Journal of Personality and Social Psychol-
ogy, 73, 758-766.
Murray, S. L., Holmes, J. G., & Griffin, D. W. (1996). The benefits of
positive illusions: Idealization and the construction of satisfaction in
close relationships. Journal of Personality and Social Psychology, 70,
79-98.
Osgood, C. E. (1962). An alternative to war or surrender. Urbana, IL:
University of Illinois Press.
Pruitt, D. G., & Rubin, J. Z. (1986). Social conflict: Escalation, stalemate,
and settlement. New York: Random House.
Reis,
H. T., & Patrick, B. C. (1996). Attachment and intimacy: Component
processes. In E. T. Higgins & A. Kruglanski (Eds.), Social psychology:
Handbook of basic principles (pp. 523-563). New York: Guilford.
Rempel, J. K., Holmes, J. G., & Zanna, M. P. (1985). Trust in close
relationships. Journal of Personality and Social Psychology, 49, 95-112.
Rotter, J. B. (1980). Interpersonal trust, trustworthiness, and gullibility.
American Psychologist, 35, 1-7.
Rusbult, C. E. (1983). A longitudinal test of the investment model: The
development (and deterioration) of satisfaction and commitment in
heterosexual involvements. Journal of Personality and Social Psychol-
ogy, 45, 101-117.
Rusbult, C. E., Bissonnette, V. L., Arriaga, X. B., & Cox, C. L. (1998).
Accommodation processes during the early years of marriage. In T. N.
Bradbury (Ed.), The developmental course of marital dysfunction (pp.
74-113). New York: Cambridge.
Rusbult, C. E., Drigotas, S. M., & Verette, J. (1994). The investment
model: An interdependence analysis of commitment processes and re-
lationship maintenance phenomena. In D. Canary & L. Stafford (Eds.),
Communication and relational maintenance (pp. 115-139). New York:
Academic Press.
Rusbult, C. E., Martz, J. M., & Agnew, C. R. (1998). The investment
model scale: Measuring commitment level, satisfaction level, quality of
alternatives, and investment size. Personal Relationships, 5,
357-391.
Rusbult, C. E., & Van Lange, P. A. M. (1996). Interdependence processes.
In E. T. Higgins & A. Kruglanski (Eds.), Social psychology: Handbook
of basic principles (pp. 564-596). New York: Guilford.
Rusbult, C. E., Van Lange, P. A. M., Yovetich, N. A., Wildschut, T., &
Verette, J. (1999). A functional analysis of perceived superiority in close
relationships. Unpublished manuscript, University of North Carolina at
Chapel Hill, Chapel Hill, NC.
Rusbult, C. E., Verette, J., Whitney, G. A., Slovik, L. F., & Lipkus, I.
(1991).
Accommodation processes in close relationships: Theory and
preliminary empirical evidence. Journal of Personality and Social Psy-
chology, 60, 53-78.
Simpson, J. A. (1987). The dissolution of romantic relationships: Factors
involved in relationship stability and emotional distress. Journal of
Personality and Social Psychology, 53, 683-692.
Sorrentino, R. M., Holmes, J. G., Hanna, S. E., & Sharp, A. (1995).
Uncertainty orientation and trust in close relationships: Individual dif-
ferences in cognitive styles. Journal of Personality and Social Psychol-
ogy, 68, 314-327.
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.
Sprecher, S. (1986). The relation between inequity and emotions in close
relationships. Social Psychology Quarterly, 49,
309-321.
Thibaut, J. W., & Kelley, H. H. (1959). The social psychology of groups.
New York: Wiley.
Uleman, J. S., & Bargh, J. A. (Eds.). (1989). Unintended thought: Limits of
awareness, intention, and control. New York: Guilford.
Van Lange, P. A. M., Rusbult, C. E., Drigotas, S. M., Arriaga, X. B., Witcher,
B.
S., & Cox, C. L. (1997). Willingness to sacrifice in close relationships.
Journal of Personality and Social
Psychology,
72, 1373-1395.
Yovetich, N. A., & Rusbult, C. E. (1994). Accommodative behavior in
close relationships: Exploring transformation of motivation. Journal of
Experimental Social Psychology, 30, 138-164.
Received September 15, 1997
Revision received March 12, 1999
Accepted April 11, 1999 •