Rumination, Emotion, and Forgiveness: Three Longitudinal Studies

Abstract

In 3 studies, the authors investigated whether within-persons increases in rumination about an interpersonal transgression were associated with within-persons reductions in forgiveness. Results supported this hypothesis. The association of transient increases in rumination with transient reductions in forgiveness appeared to be mediated by anger, but not fear, toward the transgressor. The association of rumination and forgiveness was not confounded by daily fluctuations in positive affect and negative affect, and it was not moderated by trait levels of positive affectivity, negative affectivity, or perceived hurtfulness of the transgression. Cross-lagged associations of rumination and forgiveness in Study 3 more consistently supported the proposition that increased rumination precedes reductions in forgiveness than the proposition that increased forgiveness precedes reductions in rumination.

Full text

Rumination, Emotion, and Forgiveness: Three Longitudinal Studies
Michael E. McCullough, Giacomo Bono, and Lindsey M. Root
University of Miami
In 3 studies, the authors investigated whether within-persons increases in rumination about an interper-
sonal transgression were associated with within-persons reductions in forgiveness. Results supported this
hypothesis. The association of transient increases in rumination with transient reductions in forgiveness
appeared to be mediated by anger, but not fear, toward the transgressor. The association of rumination
and forgiveness was not confounded by daily fluctuations in positive affect and negative affect, and it was
not moderated by trait levels of positive affectivity, negative affectivity, or perceived hurtfulness of the
transgression. Cross-lagged associations of rumination and forgiveness in Study 3 more consistently
supported the proposition that increased rumination precedes reductions in forgiveness than the propo-
sition that increased forgiveness precedes reductions in rumination.
Keywords: forgiveness, rumination, emotion, personality, longitudinal
For some of us whose unforgiving emotions lurk beneath a barely
civil surface, allowing ourselves to recall our hurts and offenses can
plunge us straight into rumination and thus bitter unforgiveness. To
forgive, we must recall the hurt and offense, but we must do it
differently than we usually do. (Worthington, 2001, pp. 53–54).
In popular and professional writings on forgiveness, few pieces
of conventional wisdom are more widely accepted than the notion
that recalling the transgression and then deeply considering its
implications for oneself is a fundamental preliminary step toward
forgiveness (e.g., Enright, 2001; Worthington, 2001). Indeed,
many of the theoretical models that have stimulated basic research
and forgiveness interventions articulate this same point (McCul-
lough & Worthington, 1994). But as the opening quotation sug-
gests, it is unlikely that all forms of attention to a transgression are
adaptive. Rumination about the transgression may be particularly
counterproductive. This may especially be the case insofar as
rumination leads to transient increases in negative emotion regard-
ing the transgression.
When personality, social, and clinical psychologists read the
word rumination, most likely they think of “engaging in a passive
focus on one’s symptoms of distress and on the possible causes
and consequences of these symptoms” (Nolen-Hoeksema & Jack-
son, 2001, p. 37), which has been a common way to think about
rumination since Nolen-Hoeksema (1987) proposed that gender
differences in this form of rumination accounted for the gender
difference in depressive symptoms. However, Skinner, Edge, Alt-
man, and Sherwood (2003) defined rumination more generally as
a coping strategy characterized by a “passive and repetitive focus
on the negative and damaging features of a stressful transaction”
(p. 242). Aggression researchers in particular have discussed ru-
mination about previous provocations for over 30 years (Konecni,
1974), and a self-report measure of individual differences in this
tendency is widely used (Caprara, 1986).
But whether one talks about rumination regarding one’s own
negative moods or rumination regarding negative life events, such
as interpersonal provocations, few recent ideas in personality-
social psychology have drawn as much consensus as the idea that
rumination is counterproductive for psychosocial adjustment and
interpersonal functioning. For example, ruminating about the
causes of one’s depressive symptoms appears to prolong depressed
moods and negative affect (Mor & Winquist, 2002), ruminating
about the causes of one’s anger appears to prolong anger (Rusting
& Nolen-Hoeksema, 1998), and ruminating about an insult while
hitting a punching bag appears to increase one’s likelihood of
aggressing against the insulter (Bushman, 2002). Furthermore,
rumination is an important factor in the phenomenon of triggered
displaced aggression (Miller, Pedersen, Earleywine, & Pollock,
2003).
Rumination and Forgiveness: Predictions From a
Cognitive Neoassociation (CNA) Model
In light of the consistently deleterious effects of rumination on
psychological well-being and interpersonal behavior, it seems
likely that rumination is also negatively associated with forgive-
ness. Indeed, self-report measures of forgiveness as a personality
trait are negatively associated with rumination as a personality trait
(Berry, Worthington, O’Connor, Parrott, & Wade, 2005; Berry,
Worthington, Parrott, O’Connor, & Wade, 2001). Moreover, two
studies have demonstrated that rumination about a transgression is
related cross-sectionally to higher scores on measures of revenge
and/or avoidance motivation for the transgressor (McCullough,
Bellah, Kilpatrick, & Johnson, 2001; McCullough et al., 1998).
McCullough et al. (2001) also found in a two-wave longitudinal
Michael E. McCullough, Giacomo Bono, and Lindsey M. Root, Depart-
ment of Psychology, University of Miami.
This research was generously supported by a grant from the Campaign
for Forgiveness Research and National Institute of Mental Health Grant
MH071258. We are grateful to Jo-Ann Tsang, Anna Brandon, Sharon
Brion, and the many undergraduate research assistants who assisted us with
data collection.
Correspondence concerning this article should be addressed to Michael
E. McCullough, Department of Psychology, University of Miami, P.O.
Box 248185, Coral Gables, FL 33124-0751. E-mail: mikem@miami.edu
Journal of Personality and Social Psychology Copyright 2007 by the American Psychological Association
2007, Vol. 92, No. 3, 490 –505 0022-3514/07/$12.00 DOI: 10.1037/0022-3514.92.3.490
490

study that changes in rumination were positively correlated with
changes in avoidance and revenge motivation regarding the trans-
gressor.
The CNA model of aggression (Berkowitz, 1990) and Miller et
al.’s (2003) recent elaboration of this model help to explain why
rumination may be negatively associated with forgiveness. Miller
et al. (2003) proposed that transgressions elicit negative emotions
in the victim that, in turn, activate other cognitive and motivational
structures (including thoughts, feelings, perceptual biases, motiva-
tions, and even programs for motoric behavior) that are linked in
an associative network. What the nodes in this network have in
common is that they all pertain to perceiving and reacting to
interpersonal harms. For example, anger caused by a transgression
may trigger memories of previous transgressions and thus elicit
physiological preparedness for fight-or-flight responding to the
transgression. The strength of the associations among these nodes
also influences the likelihood that they will be activated when
future transgressions occur.
According to Miller et al. (2003), rumination about a transgres-
sion in the hours, days, or weeks after a transgression reactivates
the associative network that is typically activated when one expe-
riences a transgression de novo. By rehearsing memories or images
of the transgression and considering its implications for the self,
some of the negative affect that the transgression initially elicited
is reelicited, and this negative affect spreads again to influence
cognition, motivation, and physiological preparedness for fight-or-
flight responding. Therefore, rumination may cause a reexperienc-
ing of the cognitive, affective, motivational, and physiological
consequences of the transgression as if it were occurring once
again, although probably at a lesser magnitude. More to the point
of the present article, whenever people ruminate about a transgres-
sion, they should experience a heightened readiness to avoid
and/or seek revenge against a transgressor. In other words, rumi-
nation about an unresolved transgression may reenergize people’s
efforts to avoid and/or seek revenge against the person who
harmed them. Reductions in these negative interpersonal motiva-
tions are how we have come to define interpersonal forgiveness
(McCullough, Fincham, & Tsang, 2003; McCullough et al., 1998;
McCullough, Worthington, & Rachal, 1997), so stated another
way, our main hypothesis is that rumination is negatively associ-
ated with forgiveness.
Defining and Measuring Forgiveness
Forgiveness is a suite of prosocial changes in one’s motivations
toward an interpersonal transgressor such that one becomes less
avoidant of and less vengeful toward the transgressor (and, per-
haps, more benevolent as well; McCullough et al., 1997). Forgive-
ness is an unusual construct in psychology because it is not a state,
but a process of change per se. Specifically, forgiveness is the
change process by which an individual becomes more positively
disposed and less negatively disposed toward an individual who
has harmed him or her at some point in the past. Although nearly
every forgiveness theorist posits that forgiveness involves changes
in one’s emotions, motivations, or behaviors regarding an inter-
personal transgressor such that the person thinks, feels, or behaves
more positively and less negatively toward the transgressor
(Baumeister, Exline, & Sommer, 1998; Enright & Coyle, 1998),
few researchers have directly measured forgiveness as temporal
change.
Modeling Forgiveness as Change
McCullough et al. (2003) proposed that forgiveness can be
operationalized as change in two ways using repeated measures of
people’s transgression-related interpersonal motivations, or
TRIMs. Specifically, they proposed decomposing repeated mea-
sures of people’s TRIMs regarding a transgression according to the
following statistical model:
TRIM
ij
⫽ ␤
0j
⫹ ␤
1j
(Time)
ij
⫹ r
ij
. (1)
In this model, Person j’s score on one of the TRIMs (e.g., the
amount of revenge he or she has for a transgressor) at Time i is
modeled as a linear function of two higher level parameters—an
initial status estimate of how much revenge motivation the person
would have had just after the transgression, and an estimate of the
rate of subsequent linear change in revenge. The latter parameter
representing the degree of linear change in one’s TRIMs indicates
how much Person j has forgiven his or her transgressor (with
declines in avoidance and revenge indicating more forgiveness), or
what McCullough et al. (2003) called “trend forgiveness”. Con-
sider the hypothetical longitudinal trajectories for Alan and Bill in
Figure 1 (we focus on changes in revenge for illustrative purposes,
but we could also have focused on changes in avoidance). On Day
0 (i.e., immediately after the transgression), Alan had a high level
of revenge motivation that declined steadily until Day 35. How-
ever, on Day 0, Bill had less revenge motivation, which did not
decline over the 35 days. Thus, Alan was more vengeful initially
but had more trend forgiveness, whereas Bill was less vengeful at
the outset but also less forgiving.
2
2.5
3
3.5
4
4.5
5
Day 0 Day 35
Alan
Bill
r
ij
Revenge Score
Time
Figure 1. Hypothetical longitudinal trajectories of revenge motivation for
two individuals—Alan and Bill—during the 35 days following a trans-
gression. Alan had more revenge motivation initially following the trans-
gression (i.e., on Day 0) than did Bill. Alan’s revenge motivations declined
over the 35-day period, so we can say that he has experienced a degree of
trend forgiveness. Bill’s revenge motivations did not decline, so he does
not exhibit evidence of trend forgiveness. The residual r
ij
represents the
deviation of Alan’s (i.e., person j’s) revenge score on Day i from what
would have been expected for him on the basis of his initial level of
revenge motivation and his degree of trend forgiveness. On Day i, Bill has
become more vengeful than is to be expected on the basis of his initial
status and degree of trend forgiveness, and therefore, temporarily less
forgiving.
491
RUMINATION, EMOTION, AND FORGIVENESS

Trend Forgiveness, Temporary Forgiveness, and the
Epistemic Value of Within-Subjects Research
Discovering that people’s rates of change in rumination are
associated with their rates of change in avoidance and revenge
motivation, or trend forgiveness, would essentially be a replication
of McCullough et al.’s (2001) previous findings, and, thus, would
be unsurprising. But note in Figure 1 that neither Alan’s nor Bill’s
revenge scores fall perfectly on their linear regression lines: On
some days, Alan and Bill had more or less revenge motivation
regarding their transgressors than would be expected given their
initial levels of those motivations and the rates of change in their
motivations over the 35 days (i.e., trend forgiveness). McCullough
et al. (2003) named such temporary deviations in people’s TRIMs
“temporary forgiveness,” with people being temporarily more for-
giving when their avoidance and revenge motivations were lower
than would be expected on the basis of their initial status and trend
forgiveness estimates. Of course, these residuals also contain mea-
surement error, but insofar as the residuals are correlated with
intraindividual differences in other psychological processes, the
residuals themselves may reveal important insights about why
people are more forgiving toward a transgressor on some occa-
sions than they might be on others (McCullough et al., 2003). In
light of McCullough et al.’s (2001) findings that people whose
rumination decreases over time are the people who experience
reductions in avoidance and revenge over time, it is important to
discover whether people become temporarily more forgiving (i.e.,
temporarily less avoidant and less vengeful than would be ex-
pected on the basis of their initial status and trend forgiveness
estimates) on occasions when they are also less ruminative about
the transgression than is typical for them.
The difference between these two questions is subtle but not
trivial. A conclusion generated from a between-persons research
study (e.g., the conclusion that people who experience a reduction
in rumination over a finite period of time also tend to experience
reductions in revenge and avoidance motivation) cannot be gen-
eralized to within-persons relationships. That is, we cannot extrap-
olate to the conclusion that when a specific individual becomes
less ruminative than is typical for him or her, that individual will
also tend to become less avoidant and vengeful than is typical for
him or her (Borsboom, Mellenbergh, & van Heerden, 2003). This
is because the psychological processes of individuals that create
between-persons covariation in two constructs (e.g., the finding
that people who stop ruminating over time about a transgression
are more likely to forgive the transgression than are people who do
not stop ruminating over time) may not be the same psychological
processes that generate within-persons covariation in the two con-
structs.
Fortunately, it is possible to evaluate whether a between-persons
relationship between two variables obtains at the within-persons
level, but to do so, one must use a research design that explicitly
represents within-persons changes (Borsboom et al., 2003). An
additional advantage to hypothesis tests that explicitly incorporate
within-persons changes is that they permit investigators to evaluate
how well a theoretically expected association (e.g., a negative
association between rumination and forgiveness) generalizes
across persons. In other words, it becomes possible to estimate the
percentage of individuals within a population for whom an ex-
pected association obtains. Our longitudinal model of forgiveness
(McCullough et al., 2003), which decomposes forgiveness into
between-persons components and within-persons components, is
perfectly suited to addressing such issues. Thus, a major goal of
the present article was to determine whether the previously estab-
lished between-persons association of rumination and forgiveness
obtains within persons also.
Anger and Fear: Two Possible Mediators of the
Rumination–Forgiveness Relationship
Assuming that a rumination–forgiveness association did exist at
the within-persons level, it would still be unclear what psycholog-
ical mechanisms were responsible for this association. We think
that rumination’s ability to elicit negative emotions is one of the
strongest candidates for a link by which rumination might be
associated with forgiveness. Our thinking in this regard is influ-
enced heavily by the CNA model linking negative affect and
aggression (Berkowitz, 1990). The CNA model gives pride of
place to negative emotion as a mediator of the link between
rumination and aggressive responses (Berkowitz, 1990; Miller et
al., 2003). Specifically, the CNA model posits that people rumi-
nate about a transgression, which leads to negative affects like
anger or fear that, in turn, activate other nodes in an aggression-
relevant cognitive network. This proposition implies that negative
emotions, such as anger and fear, may mediate the relationship
between rumination and forgiveness.
However, forgiveness is often conceptualized not only as a
cessation of vindictive, vengeful motivation (i.e., the sort that
would seemingly be motivated by anger) but also as a reduction in
the motivation to avoid contact with the offender (McCullough et
al., 1998; 1997). In previous work, we assumed that a vengeful
stance is associated with an angry, approach-related desire to harm
one’s transgressor in kind, whereas an avoidant stance is associ-
ated with a fearful desire to maintain a safe distance from the
transgressor (McCullough et al., 1998, 1997). This latter motiva-
tion would seem less strongly linked to anger toward a transgressor
and more strongly linked to fear of the transgressor.
If this is the case, then one should expect (a) that transient
increases in rumination would be associated with increases in
avoidance motivation only insofar as rumination led to transient
increases in fear of the transgressor and (b) that transient increases
in rumination would be associated with increases in revenge mo-
tivation only insofar as rumination led to transient increases in
anger toward the transgressor (McCullough et al., 1998, 1997).
Other investigators recently proposed a similar “matching hypoth-
eses” regarding the relations among rumination, forgiveness, and
affect (Berry, Worthington, O’Connor, et al., 2005), and it com-
ports well with other evidence that anger is associated with ap-
proach motivation (Harmon-Jones, Vaughn-Scott, Mohr,
Sigelman, & Harmon-Jones, 2004) and optimistic appraisals of
risk, whereas fear leads to pessimistic appraisals of risk (Lerner &
Keltner, 2000, 2001).
The Present Investigation
We conducted three studies to examine whether ruminating about
a real-life transgression is associated with reduced forgiveness, op-
erationalized as transient fluctuations in avoidance and revenge mo-
tivation (McCullough et al., 2003). To do so, we used multilevel
492
MCCULLOUGH, BONO, AND ROOT

random coefficient models (Bryk & Raudenbush, 1992; Hedeker,
2004) to represent within-persons changes in people’s TRIMs during
the weeks following the interpersonal transgressions they incurred.
We also examined whether within-persons fluctuations in fear and
anger toward the transgressor could be considered mediators of the
associations between rumination and forgiveness. In addition, we
investigated the possible confounding effects of fluctuations in posi-
tive and negative affect and the potential moderating effects of trait
positive and negative affectivity.
Study 1
Method
Participants
Participants were 89 students in undergraduate psychology
courses (69 women, 20 men; M age ⫽ 20.44, SD ⫽ 3.09) at
Southern Methodist University. All participants, who had incurred
a transgression in the last 7 days (M ⫽ 4.66 days, SD ⫽ 1.86),
received extra course credit for participating. Students who com-
pleted all five repeated assessments received $10. Analyses of
different aspects of this data set were reported in McCullough et al.
(2003).
Measures
TRIMs. On up to five occasions, we measured participants’
avoidance and revenge motivations toward their transgressors with
McCullough et al.’s (1998) TRIM Inventory. For this measure,
participants completed 5-point Likert-type scales to indicate how
much they agreed or disagreed with items related to their current
motivation to avoid and to seek revenge against their transgressors
(e.g., “I avoid him/her”; “I’ll make him/her pay”). These subscales
had adequate internal consistency estimates (i.e., ␣sⱖ .85 for all
measurement occasions) and test–retest stability estimates (i.e., rs
ranging from .41 to .90 across all measurement occasions). The
TRIM Inventory has shown good convergent and discriminant
validity (McCullough et al., 2001; McCullough et al., 2003; Mc-
Cullough & Hoyt, 2002; McCullough et al., 1998).
Rumination. We measured rumination about the transgression
with an eight-item scale that was inspired by the Intrusiveness
subscale of the Impact of Event Scale (Horowitz, Wilner, &
Alvarez, 1979). Participants rated on a 6-point scale ranging from
0(not at all true of me)to5(extremely true of me) how much they
had the following experiences in the last 24 hours: “I couldn’t stop
thinking about what he/she did to me”; “Thoughts and feelings
about how he/she hurt me kept running through my head”; “Strong
feelings about what this person did to me kept bubbling up”;
“Images of the offense kept coming back to me”; “I brooded about
how he/she hurt me”; “I found it difficult not to think about the
hurt that he/she caused me”; “I found myself playing the offense
over and over in my mind”; “Even when I was engaged in other
tasks, I thought about how he/she hurt me.” The linear combina-
tion of these eight items had high internal consistency (␣s ⱖ .94
for all measurement occasions) and moderate test–retest stability
(i.e., rs ranging from .24 to .82 across all five measurement
occasions).
1
Anger and fear regarding the transgressor. Participants rated
how much they felt angry, mad, enraged, afraid, and fearful
toward their transgressors on a 6-point scale ranging from 0 (not at
all)to5(extremely). The means of these adjectives were used to
measure anger and fear, respectively, toward transgressors. Inter-
nal consistency reliabilities ranged from ␣⫽.85 to .95. Test–retest
correlations ranged from .32 to .82 across the five measurement
occasions.
Trait positive and negative affect. The Positive and Negative
Affect Schedule (Watson, Clark, & Tellegen, 1988) was adminis-
tered in the initial survey to measure positive and negative trait
affectivity. The measure consists of 20 emotion words (e.g., inter-
ested, distressed) that participants rated on 5-point scales to indi-
cate the degree to which they “generally feel this way; that is, how
you feel on average” for each emotion. Internal consistency esti-
mates for the Positive and Negative subscales were high (␣s ⫽ .85
and .87, respectively).
Painfulness of the transgression. To indicate how painful they
experienced the transgression to be upon enrollment, participants
completed an item that read “How painful is the offense to you
right now?” to which they responded using a 7-point Likert-type
scale ranging from 0 (not very painful at all)to6(worst pain I ever
felt).
Procedure
We announced in several undergraduate psychology courses our
interest in surveying people who had incurred a serious interper-
sonal hurt within the previous 7 days. Throughout the semester, we
revisited these courses to enroll qualified participants and supply
them with initial packets that included the measures described
above. After they completed the initial survey, we tried to collect
follow-up data on four other occasions by contacting them in class
via announcements at the end of professors’ lectures. These
follow-up sessions were spaced roughly 2 weeks apart. Thus, for
every participant, we aimed to collect data 1, 3, 5, 7, and 9 weeks
after the transgression. Of our 89 participants, 43 provided data for
five time points, 8 for four time points, 11 for three time points, 14
for two time points, and 13 for only one time point. Thus, about
28% of the data were missing. This rate of missingness is not
considered too problematic because maximum likelihood estima-
tion is used in hierarchical linear modeling (HLM) to yield fixed
and random effects that remain unbiased by missing data if the
1
To confirm that the Rumination scale was distinct from the measures of
avoidance and revenge motivation, we conducted a maximum likelihood
factor analysis on the 12 TRIM items and the eight rumination items using data
from the first assessment for participants in Study 1 and Study 2. We used
oblimin rotation (⌬⫽ 0). Three clear factors emerged after five iterations. The
first factor, accounting for 30.6% of total item variance, consisted of the seven
TRIM-Avoidance subscale items with pattern loadings ⬎ .5. The second
factor, accounting for 21.7% of the total item variance, consisted of the eight
Rumination scale items with pattern loadings ⬎ .69. The third factor, account-
ing for 9.6% of the variance, consisted of the five TRIM-Revenge subscale
items with pattern loadings ⬎ .5. None of the items loaded on their nontarget
factors with pattern loadings of .3 or greater. The TRIM-Avoidance factor was
correlated with the TRIM-Revenge and Rumination factors at r(187) ⫽ .43
and .16, respectively. The TRIM-Revenge factor was correlated with the
Rumination factor at r(N ⫽ 189) ⫽ .12.
493
RUMINATION, EMOTION, AND FORGIVENESS

data can be assumed to be missing at random (Raudenbush &
Bryk, 2002; Schafer & Graham, 2002).
2
Statistical Models and Analyses
Within-persons models. Because our data conformed to a two-
level hierarchical structure (repeated measures nested within indi-
viduals), we ran two-level models with the HLM 6.2 statistical
software package (Raudenbush, Bryk, Cheong, & Congdon, 2000).
HLM is especially useful for this application because it accom-
modates repeated measures designs with missing data. This pro-
gram enabled us to conduct longitudinal analyses that simulta-
neously fit within-person models and between-person models that
can account for individual differences in the parameters of the
within-person models. The within-persons (or Level 1) models
were of the form:
TRIM
ij
⫽ ␤
0j
⫹ ␤
1j
(Time)
ij
⫹ ␤
2j
(Rumination)
ij
⫹ r
ij
, (2)
where TRIM
ij
⫽ Person j’s score on one of the TRIMs (i.e.,
avoidance or revenge motivation) at Time i; ␤
0j
⫽ Person j’s initial
status on the TRIM when the time scale is set to zero; ␤
1j
(Time) ⫽
the rate of change in the TRIM scores as a linear function of time
(i.e., number of weeks to the nearest 1/7th of a week since each
individual’s transgression occurred); and ␤
2j
(Rumination) ⫽ the
extent to which fluctuations above and below the regression line in
Person j’s TRIMs— based on their initial status and degree of
linear change (trend forgiveness)—are associated with within-
person fluctuations in Person j’s rumination about the transgres-
sion.
The r
ij
represents the residual in the observed TRIM score that
cannot be accounted for by the initial status, linear change, and
rumination parameters. In other words, r
ij
in Equation 2 represents
the residual in TRIM
ij
after controlling for intraindividual variabil
-
ity in rumination, whereas the r
ij
in Equation 1 represents the
residual in TRIM
ij
before controlling for intraindividual variability
in rumination. By decomposing individuals’ instantaneous TRIM
scores into components representing initial status and linear
change, the coefficient for rumination (with scores centered around
each person’s mean) represents the association between fluctua-
tions in people’s TRIMs (above and below the values we would
expect on the basis of their initial status and linear change param-
eters) and fluctuations in how much they ruminated about the
transgression. In other words, by interpreting the coefficients for
rumination derived from these within-person models, one can
evaluate whether people tend to be more avoidant or vengeful,
respectively, toward their transgressors on occasions when they are
more ruminative toward their transgressors than would be ex-
pected on the basis of (a) how avoidant or vengeful, respectively,
they were immediately after the transgression occurred and (b)
how much their avoidance (or revenge) motivation changed as a
linear function of time during the measured interval. Controlling
for initial levels and rates of linear change in this fashion enabled
us to control for similarities in initial status and rates of linear
change in the predictor and outcome variables when we evaluated
the within-persons hypotheses (McCullough et al., 2003).
3
Between-persons models. The between-persons (or Level 2)
models accounted for individual differences in the Level 1 ␤
coefficients (i.e., initial status, linear change, and the within-person
association of rumination and the outcome variable). To control
individual differences in perceived transgression painfulness and
trait positive and negative affectivity, the between-persons equa-
tions that modeled these interindividual differences included co-
variates for these variables (centered around their grand means).
The between-person equations were of the form:
␤
0j
⫽ ␥
00
⫹ ␥
01
(Painful)
j
⫹ ␥
02
(Trait Neg Affect)
j
⫹ ␥
03
(Trait Pos Affect)
j
⫹ u
0j
(3)
␤
1j
⫽ ␥
10
⫹ ␥
11
(Painful)
j
⫹ ␥
12
(Trait Neg Affect)
j
⫹ ␥
13
(Trait Pos Affect)
j
⫹ u
1j
(4)
␤
2j
⫽ ␥
20
⫹ ␥
21
(Painful)
j
⫹ ␥
22
(Trait Neg Affect)
j
⫹ ␥
23
(Trait Pos Affect)
j
⫹ u
2j
. (5)
By evaluating the significance of the effects of trait negative affect
and trait positive affect upon the ␤
2j
parameters, we were able to
evaluate whether the strength of the within-persons association
between rumination and the two TRIM variables differed as a
function of trait positive and trait negative affect.
Mediational analyses. According to Baron and Kenny (1986),
to examine whether a variable, such as anger or fear toward a
transgressor, mediates the rumination–forgiveness association,
several conditions must be evaluated. First, the predictor variable
(in this case, rumination) must be associated with both (a) the
outcome variable (i.e., avoidance or revenge motivation) and (b)
the presumed mediator variable (i.e., fear of or anger toward the
transgression) in bivariate regressions. Second, the presumed me-
diator must maintain its significant association with the outcome
variable when the outcome variable is regressed simultaneously on
the predictor and the presumed mediator. These conditions can be
evaluated jointly with Sobel’s (1982) test for mediation.
Krull and MacKinnon (2001) showed that these criteria can be
used for testing mediation in multilevel models as well. The
2
“Missing at random” is a surprisingly complex concept. Within the
context of this study, missingness at random occurs when missingness for
a given variable y on a given measurement occasion i is not uniquely
related to the unobserved measurements of y on occasion i, even though
missingness can be related to other variables that are included in the model.
In other words, as long as the missing values of y on occasion i are not
unique predictors of missingness for y on occasion i after controlling for
observed values of y from previous time points or other predictors in the
model (e.g., other within-subjects or between-subjects covariates), then the
missingness is ignorable under maximum likelihood estimation, which will
yield unbiased parameter estimates and standard errors (Schafer & Gra-
ham, 2002). Moreover, Collins, Schafer, and Kam (2001) recently dem-
onstrated that in a variety of realistic missingness scenarios, the persistence
of a unique relationship between scores on y and missingness on y can have
a rather minor impact on parameter estimates and their standard errors.
3
We controlled for initial status and linear change in these models and
equivalent models in Studies 2 and 3 because multivariate models that we
ran with these data (Raudenbush, Brennan, & Barnett, 1995) indicated that
linear changes in rumination were correlated with linear changes in avoid-
ance and revenge (rs[N ⫽ 89] ⫽ .65 and .19, respectively, in Study 1; rs
[N ⫽ 115] ⫽ .87 and .87, respectively, in Study 2; and rs[N ⫽ 162] ⫽ .39
and .59, respectively, in Study 3). These findings show that people who
experienced reductions in rumination tended also to experience reductions
in avoidance and revenge motivation, or trend forgiveness. This replicates
previous findings from McCullough et al., 2001.
494
MCCULLOUGH, BONO, AND ROOT

within-person equations for evaluating simultaneously the associ-
ation of anger or fear (the presumed mediators) and rumination
(the predictor variable) with the TRIM variables (the outcomes)
were of the form:
TRIM
ij
⫽ ␤
0j
⫹ ␤
1j
(Time)
1j
⫹ ␤
2j
(Rumination)
ij
⫹ ␤
3j
(Anger)
ij
⫹ r
ij
, (6)
and we again included between-persons equations as in Equations
3–5 above.
Results
Descriptive Statistics
As described in McCullough et al. (2003; Study 2), the types of
relationship partners who had committed transgressions against
our participants were diverse. Most transgressions were committed
by girlfriends or boyfriends (42%), friends of the same gender
(23%), and friends of the other gender (15%). Smaller numbers
reported on transgressions by relatives (10%), husbands and wives
(3%), and “others” (8%). Participants described several types of
transgressions, including betrayals of a confidence or insults by a
friend (36%); neglect by a romantic partner, spouse, or ex-
romantic partner (25%); infidelity by a romantic partner or spouse
(13%); rejection, neglect, or insult by a family member (10%);
termination of romantic relationship (7%); insults by people other
than family or friends (3%); and rejection or abandonment by
friend or prospective relationship partner (3%). Two participants
declined to describe the specific transgression. Mean level of pain
reported by participants was 3.81 (SD ⫽ 1.39). Means and stan-
dard deviations for the major variables in Study 1 appear in
Table 1.
Fluctuations in Rumination as Predictors of Temporary
Forgiveness
We first examined whether people are temporarily more forgiv-
ing (i.e., whether they have less avoidance and revenge motivation
than would be expected on the basis of their initial status and linear
change estimates) when they are less ruminative about the trans-
gression than is typical for them, per Equation 2. Table 2 lists
statistics associated with the parameter estimates of initial status
and linear change (trend forgiveness) in avoidance and revenge as
well as the average within-persons association of rumination and
each of the TRIMs.
Table 2 shows that the average participant had an initial status
of 2.46 and a linear change of (essentially) zero units per week on
the avoidance metric, and an initial status of 1.79 and a linear
decrease of .02 units per week on the revenge metric. For the
average person, every one-unit increase in rumination (above his
or her mean) was associated with a .16-unit increase in avoidance.
Similarly, every one-unit increase in rumination above the average
person’s typical degree of rumination was associated with a .07
increase in revenge. Therefore, rumination covaried considerably
and uniquely with people’s TRIMs, effect size rs(85) ⫽ .39 and
.29 for the avoidance and revenge coefficients, respectively
(Hunter & Schmidt, 1990). On occasions when people reported
more rumination about their transgression than was typical for
them, they also reported higher levels of avoidance and revenge
motivations than would have been expected on the basis of their
initial status and linear change (trend forgiveness) estimates. In
other words, on days when people ruminated more about the
transgression than was typical for them, they also tended to be less
forgiving.
4
The standard deviations for the within-persons associations of
rumination with avoidance and revenge motivation, respectively,
were 0.24 and 0.02, respectively. By dividing the coefficients for
the associations of rumination with avoidance and revenge moti-
vation by their standard deviations, we can determine the percent-
age of individuals in the population who would be expected to
have positive associations of rumination with avoidance and re-
venge motivation. In the case of the rumination–avoidance asso-
ciation, people who scored more than .16/.24 ⫽ .67 standard
deviations below the mean would be expected to have rumination–
avoidance associations that were not positive. In other words,
about 75% of the population would be expected to have positive
within-persons associations of rumination and avoidance. In the
case of the rumination–revenge association, people who scored
4
Gender differences in the within-persons associations of rumination
with avoidance and revenge motivation were consistently nonsignificant.
Table 1
Means and Standard Deviations for Major Variables in Studies 1–2
Measure
Time 1 Time 2 Time 3 Time 4 Time 5
MSDMSDMSDMSDMSD
Study 1
Avoidance 2.57 1.10 2.46 1.09 2.39 1.05 2.32 1.12 2.46 1.12
Revenge 1.84 0.85 1.71 0.91 1.65 0.75 1.57 0.69 1.55 0.73
Rumination 3.18 1.14 1.71 1.57 1.19 1.39 0.97 1.29 0.80 1.25
Study 2
Avoidance 3.11 1.08 2.96 1.10 2.82 1.13 2.74 1.10 2.70 1.15
Revenge 1.83 0.83 1.61 0.80 1.53 0.77 1.52 0.79 1.48 0.87
Rumination 3.55 0.96 2.01 1.21 1.24 1.13 0.90 1.00 0.80 1.05
495
RUMINATION, EMOTION, AND FORGIVENESS

more than .07/.02 ⫽ 3.5 standard deviations below the mean would
be expected to have rumination–avoidance associations that were
not positive. In other words, more than 99% of the population
would be expected to have positive within-persons associations of
rumination and revenge.
The values in the “%VAF” (% variance accounted for) column
of Table 2 indicate that the initial status, trend forgiveness, and
rumination parameters did a good job of accounting for variance in
people’s avoidance and revenge scores (i.e., 85.31% and 79.88%
of the variance, respectively).
Is the Within-Persons Association of Rumination and
Temporary Forgiveness Moderated by Trait Positive or
Negative Affectivity?
We proceeded to examine whether individual differences in the
association of rumination with temporary forgiveness were mod-
erated by individual differences in positive and negative affectiv-
ity, per Equation 5. Neither trait positive affectivity nor trait
negative affectivity appeared to moderate the associations of ru-
mination with revenge and avoidance ( ps ⬎ .10).
Is the Association of Rumination and Forgiveness
Mediated by Anger or Fear Toward the Transgressor?
Because fluctuations in rumination were associated with tem-
porary forgiveness, we then examined whether people also expe-
rienced changes in anger and/or fear vis a` vis the transgressor
when they ruminated about the transgression more than was typ-
ical for them and, if so, whether those emotional changes appeared
to mediate the associations of rumination and temporary forgive-
ness.
The first two columns of data in Table 3, labeled ␤(yx) and
␤(mx), contain the coefficients resulting from regressing the pre-
sumed outcomes y (i.e., avoidance or revenge motivation) and the
presumed mediators m (i.e., fear of and anger toward the trans-
gressor) upon the predictor variable x (i.e., rumination). These
coefficients show that rumination was significantly related to the
presumed outcome variables (avoidance and revenge) as well as
the presumed mediators (fear of and anger toward the transgres-
sor). The third column of data, labeled ␤(ym.x) demonstrates that
when anger was used as a potential mediator, it maintained a
significant association with the TRIM variables, with rumination
simultaneously controlled. The same was not the case when fear
was used as a potential mediator. The coefficients in the column of
data labeled ␤(yx.m), which displays the associations of the pre-
dictor x (rumination) with the presumed outcomes y (i.e., avoid-
ance or revenge motivation) with the presumed mediator m (fear or
anger) simultaneously controlled, were nonsignificant when anger
was examined as a potential mediator. These results are consistent
with the hypothesis that anger (but not fear) toward the transgres-
sor functioned as a mediator of the associations of rumination with
avoidance and revenge. The Sobel tests, whose t values appear in
Table 3, also support this interpretation.
The traditional Sobel test for mediation ignores the so-called ab
covariance (Kenny, Korchmaros, & Bolger, 2003), which is co-
variation between the random effects for the association of the
predictor with the mediator (the a path) and the random effects for
the unique association of the mediator with the outcome while
controlling for the direct effect of the predictor (the b path). Failure
to consider this ab covariance when testing mediational hypothe-
ses in multilevel models can lead to underestimates or overesti-
mates of mediated effects. Using Korchmaros and Kenny’s (2003)
Table 2
Parameter Estimates for Linear Models of Longitudinal Change in TRIMs in Studies 1–2
Measure (Parameter) MSDSE t % VAF
a
␹
2
(1, N ⫽ 50)
Study 1
Avoidance
Initial status 2.46 1.03 .12 80.95 275.83
***
Linear change (forgiveness) 0.00 0.10 .02 0.03 0.02 128.18
***
Rumination 0.16 0.24 .04 3.98
**
4.34 99.37
**
Revenge
Initial status 1.79 0.80 .09 79.82 238.47
***
Linear change (forgiveness) ⫺0.02 0.07 .01 ⫺2.08
*
0.01 107.18
***
Rumination 0.07 0.02 .03 2.78
**
0.05 57.89
Study 2
Avoidance
Initial status 2.96 1.09 .12 82.64 360.21
***
Linear change (forgiveness) ⫺0.02 0.11 .02 ⫺0.97 0.90 186.27
***
Rumination 0.12 0.26 .04 2.84
**
4.72 177.49
***
Revenge
Initial status 1.61 0.65 .07 79.30 292.04
***
Linear change (forgiveness) 0.00 0.06 .01 0.20 0.67 156.46
**
Rumination 0.14 0.13 .02 5.44
***
3.19 131.00
*
Note. All models included Level 2 covariates representing perceived transgression painfulness, trait positive affect, and trait negative affect, all centered
around their grand means (coefficients not shown). TRIMs ⫽ transgression-related interpersonal motivations; VAF ⫽ variance accounted for.
a
% VAF ⫽ percentage of variance accounted for by the initial status, linear change, and rumination parameters.
*
p ⬍ .05.
**
p ⬍ .01.
***
p ⬍ .001.
496
MCCULLOUGH, BONO, AND ROOT

approach for incorporating the ab covariance, we calculated the
percentage of the association of rumination on avoidance that
appeared to be mediated by anger (100%) and fear (2.71%),
respectively, and the percentage of the association of rumination
and revenge that appeared to be mediated by anger (100%) and
fear (14.70%), respectively. (Because of sampling error, estimates
can exceed 100% or fall below 0%. Here and elsewhere in the
present article, we constrain these estimates to their upper bounds
of 100% and lower bounds of 0%.)
Summary of Study 1
Study 1 revealed that rumination was negatively associated with
temporary forgiveness (transient fluctuations in people’s avoid-
ance and revenge motivations). When people were more rumina-
tive than was typical for them, they also tended to be more
avoidant and vengeful toward their transgressors than was typical
for them. These associations obtained even though we controlled
for initial perceptions of transgression painfulness as well as trait
positive and negative affectivity, and there was no evidence that
these person-level variables moderated the within-persons associ-
ations of rumination with avoidance and revenge motivation. Me-
diational analyses were consistent with the idea that the associa-
tions of rumination with avoidance and revenge motivation were
mediated by anger toward the transgressor.
To examine the rumination–forgiveness association with greater
statistical power and control, we conducted a second study. In Study
2, we examined whether the associations of transient fluctuations in
rumination were associated with temporary forgiveness while simul-
taneously controlling for transient fluctuations in positive and nega-
tive mood. Rumination and negative affect are dependably related
(Mor & Winquist, 2002), so people may engage in more rumination
when they are in negative moods. Insofar as negative mood states are
associated with more rumination and less forgiveness, any apparent
correlation between rumination and forgiveness might be spurious.
Therefore, it is prudent to rule out the possibility that the rumination–
forgiveness relationship is a byproduct of intraindividual fluctuations
in positive and negative mood. Moreover, Berkowitz (1990) sug-
gested that it was negative affect generally, rather than any specific
negative affect such as anger or fear, that elicits aggressive respond-
ing, so in the context of forgiveness, it is useful to examine its
associations with discrete affects such as anger or fear while control-
ling for negative affect more generally.
Study 2
Method
Participants and Procedure
Participants were 115 students in undergraduate psychology
courses (91 women, 24 men; M age ⫽ 19.76, SD ⫽ 2.61) at
Table 3
Mediational Analyses for Rumination–TRIM Associations in Studies 1–3
a
Variable ␤(yx) ␤(mx) ␤(ym.x) ␤(yx.m) Sobel’s t
Study 1
x-m-y
b
Rum-Anger-Avo 0.16
**
0.47
***
0.33
***
⫺0.00 5.22
***
Rum-Fear-Avo 0.16
**
0.20
**
0.02 0.14
**
0.43
Rum-Anger-Rev 0.07
*
0.47
***
0.20
***
⫺0.02 4.06
***
Rum-Fear-Rev 0.07
*
0.20
**
0.06 0.07
*
1.66
Study 2
x-m-y
Rum-Anger-Avo 0.12
**
0.46
***
0.24
***
0.01 5.41
***
Rum-Fear-Avo 0.12
**
0.23
***
⫺0.00 0.12
*
0.07
Rum-Anger-Rev 0.14
***
0.46
***
0.11
***
0.08
**
3.93
***
Rum-Fear-Rev 0.14
***
0.23
***
⫺0.02 0.12
***
0.49
Study 3
c
x-m-y
Rum-Anger-Avo 0.03
*
0.13
**
0.13
***
0.02 3.99
***
Rum-Fear-Avo
d
0.03
*
0.01
Rum-Anger-Rev 0.03
*
0.13
***
0.12
***
0.01 3.89
***
Rum-Fear-Rev
d
0.03
*
0.02
Note. In the regressions for Study 3, scores on Day i ⫺ 1 for anger, fear, and the TRIM (trangression-related
interpersonal motivation) being modeled as a dependent variable are statistically controlled. Rum ⫽ Rumination;
Avo ⫽ Avoidance; Rev ⫽ Revenge.
a
All models included Level 2 covariates representing perceived transgression painfulness, trait positive affec
-
tivity, and trait negative affectivity, all centered around the grand means (coefficients not shown).
b
x ⫽
presumed predictor variable; m ⫽ presumed mediator variable; y ⫽ presumed outcome variable.
c
Study 3
coefficients reflect mediational analyses involving rumination on Day i ⫺ 1 as the x variable and anger, fear,
avoidance motivation, and revenge motivation on Day i as the presumed mediator and outcome variables.
d
Missing cells indicate that mediational analyses were not performed because the preliminary conditions for
assessing mediation were not fulfilled.
*
p ⬍ .05.
**
p ⬍ .01.
***
p ⬍ .001.
497
RUMINATION, EMOTION, AND FORGIVENESS

Southern Methodist University. As in Study 1, all participants had
incurred a transgression within the last 7 days (M ⫽ 4.04 days,
SD ⫽ 1.82); they received extra course credit for participating and
$10 for completing all five assessments. Other analyses based on
this data set are reported in McCullough, Orsulak, Brandon, and
Akers (2007). We used the same procedure as in Study 1, except
that the five measurement occasions were spaced approximately 2
weeks apart (on the same day of the week and time of day), and
they occurred in Michael E. McCullough’s laboratory. Of the 115
participants, 96 completed all five visits, 5 completed 4 visits, 5
completed 3 visits, 5 completed 2 visits, and 4 completed only one
visit. With only 8% missing data, attrition was much lower than in
Study 1.
Measures
We used the same instruments as in Study 1 to measure people’s
TRIMs, rumination, fear and anger toward the transgressor, trait
positive and negative affect, and painfulness of the transgression.
All of the measures had comparable internal consistency and
test–retest reliability estimates as in Study 1. In addition, we
measured the moodlike manifestations of positive and negative
affect by having participants also complete the PANAS on re-
peated occasions. On this form of the PANAS, participants were
instructed to indicate how much they experienced 20 different
emotions “in the last two weeks.” Internal consistency and test–
retest stability estimates across all measurement occasions were
acceptable (i.e., ␣s ranged from .85 to .95, and rs ranged from .54
to .93).
Statistical Models and Data Analyses
We used the same statistical modeling approach as in Study 1.
That is, we decomposed people’s five repeated TRIM scores into
a within-person model, with parameters representing initial levels
of the variable, trend forgiveness, rumination (with values centered
around each person’s mean), and residual variance, per Equation 2.
We simultaneously included between-persons covariates repre-
senting perceived transgression painfulness and trait positive and
negative affectivity, per Equations 3–5.
To determine whether temporal fluctuations in positive and
negative mood could account for the association of fluctuations in
rumination with temporary forgiveness, we ran HLMs, with pos-
itive and negative state affect included as time-varying predictors
of the TRIMs. Because we only had a maximum of five observa-
tions per individual, degrees of freedom within persons were
inadequate for introducing positive mood and negative mood as
simultaneous predictors, so we had to evaluate their associations
with the outcome variables in separate models. This resulted in two
new Level 1 equations:
TRIM
ij
⫽ ␤
0j
⫹ ␤
1j
(Time)
ij
⫹ ␤
2j
(Positive mood)
ij
⫹ ␤
3j
(Rumination)
ij
⫹ r
ij
(7)
TRIM
ij
⫽ ␤
0j
⫹ ␤
1j
(Time)
ij
⫹ ␤
2j
(Negative mood)
ij
⫹ ␤
3j
(Rumination)
ij
⫹ r
ij
. (8)
As in Study 1, we controlled for perceived transgression painful-
ness and positive and negative trait affectivity by including these
variables in the between-persons equations, per Equations 3–5.
To evaluate whether anger or fear toward the transgressor me-
diated the links between rumination and temporary forgiveness, we
conducted mediational analyses, as in Study 1. To do so, we tested
within-person models in the form of Equation 6.
Results
Descriptive Statistics
As in Study 1, most participants described transgressions com-
mitted by girlfriends or boyfriends (59%), friends of the same
gender (19%), or friends of the other gender (11%). A few partic-
ipants reported transgressions by relatives (10%), husbands or
wives (3%), and “others” (9%). One person did not report the type
of relationship involved. Participants experienced insults by a
friend or betrayals of a confidence (28%); neglect by a romantic
partner, spouse, or ex-romantic partner (22%); infidelity by a
romantic partner or spouse (19%); rejection, neglect, or insult by a
family member (10%); termination of romantic relationship
(11%); insults by people other than family or friends (3%); and
rejection or abandonment by a friend or prospective relationship
partner (2%). Five participants did not describe the transgression.
Participants in Study 2 reported that their transgressions were more
painful (M ⫽ 4.78, SD ⫽ 0.81) compared with participants in
Study 1 (M ⫽ 3.81, SD ⫽ 1.39). Table 1 displays the means and
standard deviations for the major variables in Study 2.
Fluctuations in Rumination as Determinants of
Temporary Forgiveness
To replicate Study 1⬘s finding that people are temporarily more
forgiving when they ruminate about the transgression less than is
typical for them, we examined the within-persons associations of
rumination with TRIM-avoidance and TRIM-revenge scores. Ta-
ble 2 displays the statistics associated with the parameter estimates
for initial status and linear change in the TRIMs and for time-
varying fluctuations in rumination.
As in Study 1, time-varying fluctuations in rumination covaried
significantly and uniquely with avoidance and revenge motivation.
On occasions when people ruminated more about the transgression
than was typical for them, they also tended to have higher levels of
avoidance and revenge motivation than would have been expected
on the basis of their initial status and linear change (trend forgive-
ness) parameter estimates over the measured interval, effect size
rs(111) ⫽ .26 and .46 for the avoidance and revenge coefficients,
respectively.
The standard deviations for the within-persons associations of
rumination with avoidance and revenge motivation, respectively,
were 0.26 and 0.13, respectively. By dividing the coefficients by
their standard deviations, we can determine the percentage of
individuals in the population who would be expected to have
positive associations of rumination with avoidance and revenge
motivation. In the case of the rumination–avoidance association,
people who scored more than .12/.26 ⫽ .46 standard deviations
below the mean would be expected to have rumination–avoidance
associations that were not positive. In other words, about 68% of
the population would be expected to have positive within-persons
associations of rumination and avoidance. In the case of the
rumination–revenge association, people who scored more than
498
MCCULLOUGH, BONO, AND ROOT

.14/.13 ⫽ 1.08 standard deviations below the mean would be
expected to have rumination–avoidance associations that were not
positive. In other words, about 86% of the population would be
expected to have positive within-persons associations of rumina-
tion and revenge.
The values in the “%VAF” column of Table 2 indicate that the
initial status, trend forgiveness, and rumination parameters ac-
counted for most of the variance in people’s avoidance and re-
venge scores (i.e., 88.26% and 83.16%).
Is the Association of Rumination With Temporary
Forgiveness Moderated by Trait Affectivity?
As in Study 1, neither trait positive affect nor trait negative
affect significantly moderated the within-persons associations of
rumination and TRIM-avoidance and TRIM-revenge scores in
Study 2 ( ps ⬎ .15).
Is the Association of Rumination With Temporary
Forgiveness Confounded by Mood?
To examine whether the associations of day-to-day fluctuations
in rumination with temporary forgiveness could be attributed to
fluctuations in positive and negative mood (i.e., whether rumina-
tion’s negative association with temporary forgiveness could be
explained by the fact that people who were in bad moods were
both more ruminative and less forgiving), we examined the within-
persons associations of rumination with avoidance and revenge
motivation while controlling for within-persons fluctuations in
positive and negative mood, per Equations 7– 8.
Fluctuations in positive mood and negative mood were not
significantly associated with fluctuations in avoidance and revenge
motivation. As expected, therefore, the within-persons associations
of rumination with avoidance and revenge motivation, respec-
tively, while controlling for positive mood were 0.13, t(111) ⫽
2.85, p ⬍ .01 (effect size r ⫽ .26), and 0.14, t(111) ⫽ 5.72, p ⬍
.001 (effect size r ⫽ .47). Thus, controlling for state positive affect
did not change the within-persons associations of rumination with
avoidance and revenge motivation.
When simultaneously controlling for negative mood, rumination
maintained its unique significant associations with avoidance and
revenge motivation. The coefficients for the associations of rumi-
nation with avoidance and revenge, respectively, while controlling
for negative mood were 0.13, t(111) ⫽ 2.96, p ⬍ .01 (effect size
r ⫽ .27), and 0.11, t(111) ⫽ 4.70, p ⬍ .001 (effect size r ⫽ .40).
Therefore, the within-persons association of rumination with
avoidance and revenge motivation could not be attributed to the
influence of negative mood.
Is the Association of Rumination and Forgiveness
Mediated by Anger or Fear Toward the Transgressor?
As in Study 1, we proceeded to examine whether people also
experienced changes in anger and/or fear vis a` vis the transgressor
on occasions when they ruminated about the transgression more
than was typical for them and, if so, whether those emotional
changes appeared to mediate the associations between rumination
and temporary forgiveness.
The first two columns of data in the second part of Table 3,
labeled ␤(yx) and ␤(mx), contain the coefficients that resulted from
regressing the outcomes y (i.e., avoidance and revenge motivation)
and presumed mediators m (i.e., fear and anger) upon the predictor
variable x (i.e., rumination). These coefficients demonstrate that
rumination was significantly related to the presumed outcome
variables (avoidance and revenge) as well as the presumed medi-
ators (fear of and anger toward the transgressor). The third column
of coefficients, labeled ␤(ym.x), shows that anger toward the
transgressor (m) maintained a significant association with the
TRIM variables (y), with rumination (x) simultaneously controlled.
This is consistent with the proposition that anger mediated the
association of rumination with avoidance and revenge motivation.
The same was not true when we examined fear as a potential
mediator of the rumination–TRIM relationships: Because the co-
efficients in the third column were not significant when fear was
the presumed mediator, we can conclude that fear did not mediate
the associations of rumination with avoidance and revenge moti-
vation. The coefficients in the column of data labeled ␤(yx.m),
demonstrating the unique relationships between the outcomes y
(avoidance and revenge motivation) and the predictor x (rumina-
tion) when controlling for the presumed mediators m (i.e., avoid-
ance and fear motivation), were nonsignificant when anger was the
mediator. The Sobel tests, whose t values appear in Table 3, also
support the conclusion that anger (but not fear) toward the trans-
gressor mediated the associations of rumination with avoidance
and revenge.
Kenny et al. (2003) wrote that incorporating the ab covariance
for multilevel mediational models is only necessary when the
random effects variance for the a path (i.e., the coefficient repre-
senting the regression of the putative mediator on the putative x
variable) and the random effects variance for the b path (i.e., the
coefficient representing the association of the outcome variable y
with the putative mediator variable when then putative x variable
is simultaneously controlled) are significantly different from zero.
These conditions were not fulfilled in our data—that is, for all
mediational models, the a path, the b path, or both lacked signif-
icant random effects variance—so Kenny et al.’s calculations were
not necessary to accurately estimate mediation. Using the tradi-
tional approach to estimating the percentage of an association of a
relation between two variables that can be explained by a putative
mediator (i.e., multiplying the a path and the b path, and dividing
their product by the total association of the predictor and the
outcome variable), we calculated the percentage of the association
of rumination on avoidance that appeared to be mediated by anger
(90.0%) and fear (0%), respectively, and the percentage of the
association of rumination and revenge that appeared to be medi-
ated by anger (37.4%) and fear (0%), respectively.
Summary of Study 2
Study 2 replicated the main findings from Study 1 with better
statistical power and control. On occasions when people ruminated
to a greater extent than was typical for them about the transgres-
sions they had incurred, they also tended to have higher levels of
avoidance and revenge motivation than was typical for them.
These associations between temporal fluctuations in rumination
and “temporary forgiveness” were not moderated by perceived
painfulness of the transgression, trait positive affect, or trait neg-
499
RUMINATION, EMOTION, AND FORGIVENESS

ative affect. These associations were also independent of positive
and negative mood. In addition, we found that when people rumi-
nated more about a transgression than was typical for them, they
also tended to experience transient increases in anger toward their
transgressors, and these increases in anger appeared to mediate the
within-persons associations of rumination with avoidance and re-
venge motivation.
The results of Studies 1 and 2 merely document that rumination
is correlated within measurement occasions with higher levels of
anger, avoidance motivation, and revenge motivation. Causal con-
clusions are certainly not justifiable on the basis of such findings.
However, when the number of observations per person provides
adequate degrees of freedom, it becomes possible to specify more
rigorous models that enable one to evaluate whether rumination on
a given day is associated with avoidance and revenge motivation
on the next day, even after controlling for previous levels of
avoidance and revenge motivation. Establishing that changes in a
putative x variable (e.g., rumination) precede changes in a putative
y variable (i.e., forgiveness) would provide stronger (although by
no means definitive) support for the proposition that the
rumination–forgiveness association is a causal one (Finkel, 1995;
West & Hepworth, 1991). Therefore, we conducted a third study
using data from a 21-day diary study that allowed us to evaluate
the temporal nature of the relations among our variables.
Study 3
Method
Participants and Procedure
Participants were 163 students in undergraduate psychology
courses (112 women, 51 men) at the University of Miami. As in
Studies 1 and 2, all participants had incurred a transgression within
the last 7 days (M ⫽ 4.37 days, SD ⫽ 1.85) at the time of
enrollment. They received extra course credit for participating and,
if they completed the tasks described here and a separate labora-
tory session not described here, $20.
As in Studies 1 and 2, participants were recruited via their
undergraduate psychology courses. Throughout the semester, we
visited these courses, enrolled qualified participants in the study,
and provided them with initial screening packets. After they com-
pleted the initial survey, we gave them a packet of 21 question-
naires; participants were to complete one questionnaire per day.
We instructed participants to complete each entry honestly and
advised them that it was better to leave a day’s entry blank than to
falsify their responses. In addition, we contacted participants pe-
riodically during the 21-day diary period to encourage compliance
and timely completion of their diary entries. Participants dated
each entry and returned the packets to Michael E. McCullough’s
laboratory approximately 21 days after beginning the study. As the
histogram in Figure 2 shows, completion rates were very high,
with 109 participants (66.9%) completing 21 diary entries, and
only 9 participants (5.5%) failing to complete at least 10 diary
entries.
5
Measures
We used the same instruments as in Studies 1 and 2 to measure
people’s TRIMs, rumination, fear and anger toward their trans-
gressions, trait positive and negative affect, and painfulness of the
transgression (the only difference being that participants were
asked to report on their daily experience for all measures because
the measures were administered daily). All of the measures’ reli-
abilities were comparable to those in Studies 1 and 2.
Statistical Models and Data Analyses
The goal of Study 3 was to evaluate whether ruminating about
a transgression on a given day predicts one’s avoidance and
revenge motivation on the successive day. In other words, we
wished to determine whether the extent to which a person was
ruminating about a transgression on any given day (while control-
ling for their revenge or avoidance motivations on that day, as well
as their fear and anger toward the transgressor on that day) was
associated with their revenge and avoidance motivations toward
the transgressor on the next day. Insofar as this turned out to be the
case, we wanted to determine whether the prospective association
of rumination with the TRIM scores could be explained by the fact
that ruminating on a given day, even after controlling for anger
toward or fear of the transgressor on that day, was associated with
participants’ anger toward or fear of their transgressor on the next
day (Cole & Maxwell, 2003; West & Hepworth, 1991). Therefore,
we estimated two sets (one set for avoidance motivation and one
for revenge motivation) of two-level regression models of the
form:
TRIM
ij
⫽ ␤
0j
⫹ ␤
1j
(Time)
ij
⫹ ␤
2j
(TRIM)
i⫺1,j
⫹ ␤
3j
(Rumination)
i⫺1,j
⫹ ␤
4j
(Anger)
i⫺1,j
⫹ ␤
5j
(Fear)
i⫺1,j
⫹ r
ij
. (9)
In Equation 9, Person j’s TRIM score on Day i is modeled as a
function of Person j’s intercept (i.e., expected TRIM score when
other predictors obtain a value of 0), a linear function of time for
Person j, and Person j’s TRIM rumination, anger, and fear scores
from the previous day (thus the use of the subscript
i⫺1, j
). To
evaluate the hypothesis that increases in rumination precede re-
ductions in forgiveness, we evaluated the statistical significance of
the ␤
3j
coefficient for rumination. To examine whether anger
toward or fear of the transgressor on a given day was associated
uniquely with rumination on the previous day, we estimated mod-
els in the form of Equation 9, but which used anger and fear vis a`
vis the transgressor on Day i for Person j as the criterion variables.
In two final models, we evaluated whether the association of the
previous day’s rumination with the TRIM variables on any given
day could be explained by the effects of the previous day’s
rumination on the given day’s levels of anger or fear toward the
5
A reviewer pointed out Stone, Shiffman, Schwartz, Broderick, and
Hufford’s (2003) findings that medical patients often complete their dairy
entries in an untimely fashion (e.g., completing entries hours or days later
than they were supposed to). We had no way of checking whether partic-
ipants did so in Study 3, but because we made it clear to our participants
that we preferred for them to omit missed entries rather than to fake them
(and because course credit and financial incentives were not contingent
upon completion of all entries), it seems rather unlikely that such confab-
ulation was very prevalent. Nevertheless, we acknowledge that other steps
could have been taken to maximize compliance and timely completion of
diary entries, particularly in Study 3 (Green, Rafaeli, Bolger, Shrout, &
Reis, 2006).
500
MCCULLOUGH, BONO, AND ROOT

transgressor. To do so, we estimated models in the form of Equa-
tion 9 with one more predictor variable, which was Person j’s
anger toward or fear of the transgressor, respectively, on Day i.
These analyses allowed us to evaluate whether (a) rumination was
associated with reduced forgiveness on the next day, (b) rumina-
tion was associated with increased anger or fear of the transgressor
on the next day, and (c) insofar as (a) and (b) were true, whether
the lagged association of rumination and forgiveness could be
explained in terms of an intermediate association of rumination
with anger or fear regarding the transgressor. As in Studies 1 and
2, each of the Level 1 ␤ coefficients was modeled at Level 2 as a
function of a grand mean, a person-specific residual, trait positive
affectivity, trait negative affectivity, and perceived transgression
painfulness so that we could examine the latter as possible mod-
erator variables.
Through a series of analogous models, we also evaluated
whether avoidance and revenge on a given day were uniquely
associated with rumination on the next day and, if so, whether
those associations appeared to be mediated by the associations of
avoidance and revenge on a given day with anger and/or fear of the
transgressor on the next day.
Results
Descriptive Statistics
As in Studies 1 and 2, the types of relationship partners who had
committed transgressions against our participants were diverse.
Most incurred transgressions committed by girlfriends or boy-
friends (50%), friends of the same gender (19%), or relatives
(13%). A smaller number of participants reported transgressions
by friends of the other gender (9%), husbands or wives (1%), and
“others” (8%). Participants described several types of transgres-
sions, including infidelity by a romantic partner or spouse (29%);
insults by a friend or betrayals of a confidence (20%); rejection,
neglect, or insult by a family member (13%); termination of a
romantic relationship (13%); neglect by a romantic partner,
spouse, or ex-romantic partner (10%); rejection or abandonment
by a friend or prospective relationship partner (10%); and insults
by people other than family or friends (5%). At the beginning of
the study, participants in Study 3 reported transgressions that were
about as painful (M ⫽ 4.84, SD ⫽ 0.88) as did participants in
Study 2 (M ⫽ 4.78, SD ⫽ 0.81).
Fluctuations in Rumination as Determinants of
Temporary Forgiveness
Multilevel models revealed that on any given Day i ⫺ 1, if
people had relatively high levels of rumination, then they were
expected to have relatively high levels of avoidance motivation
(␤⫽0.035, SE ⫽ 0.015), t(160) ⫽ 2.31, p ⫽ .022, effect size r ⫽
.18, on the following Day i, even after controlling for the covaria-
tion among rumination, avoidance motivation, anger toward the
transgressor, and fear of the transgressor on Day i ⫺ 1. The
standard deviation for this association was .094, implying that
people who scored more than .035/.094 ⫽.37 SDs below the mean
had rumination–avoidance associations that were zero or less. In
other words, about 64% of participants in this sample would be
expected to have positive within-persons associations between
rumination on Day i ⫺ 1 and avoidance motivation on Day i.
Similarly, on any given Day i ⫺ 1, if people had relatively high
levels of rumination, then they were expected to have relatively
high levels of revenge motivation (␤⫽0.031, SE ⫽ 0.012),
t(160) ⫽ 2.69, p ⫽ .008, effect size r ⫽ .21, on the following Day
i, even after controlling for the covariation among rumination,
revenge motivation, anger toward the transgressor, and fear of the
transgressor on Day i ⫺ 1.
The standard deviation for this association was .056, implying
that people who scored more than .031/.056 ⫽.55 SDs below the
mean had rumination–avoidance associations that were not posi-
tive. In other words, about 71% of participants in this sample
would be expected to have positive within-persons associations
between rumination on Day i ⫺ 1 and revenge motivation on
Day i.
The associations between rumination on Day i ⫺ 1 and the
TRIMs on Day i were not moderated by trait negative affect, or
perceived transgression painfulness, although the association of
rumination on Day i ⫺ 1 and revenge on Day i was moderated by
trait positive affect (␤⫽⫺0.033, SE ⫽ 0.013), t(160) ⫽⫺2.54,
p ⫽ .012. For every unit increase in trait positive affect above the
mean, the association between rumination on Day i ⫺1 and re-
venge motivation on Day i went down by .033 units. In other
words, people with high levels of trait positive affect could appar-
ently ruminate about a transgression that had occurred to them
without experiencing a large increase in revenge motivation the
following day.
Is the Association of Rumination and Forgiveness
Mediated by Anger or Fear Toward the Transgressor?
Because increases in rumination on any given day appeared to
be associated with increases in avoidance and revenge motivation
on the next day, we proceeded to examine whether rumination on
any given Day i ⫺ 1 predicted people’s levels of anger toward, or
fear of, the transgressor on the next Day i and, if so, whether those
lagged associations of rumination with fear and anger mediated the
associations of rumination on Day i ⫺ 1 and the TRIMs on Day i.
The first two columns of data in the bottom third of Table 3,
labeled ␤(yx) and ␤(mx), contain the coefficients that resulted from
regressing the outcomes y (i.e., avoidance and revenge motivation
on any given Day i) and presumed mediators m (i.e., fear or anger
on Day i) upon the predictor variable x (i.e., rumination on Day i ⫺
1, controlling for levels of the outcome, fear, and anger on Day i ⫺
0
20
40
60
80
100
120
21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1
Number of Entries Completed
Number of Participants
Figure 2. Daily questionnaire completion rates for participants in Study
3(N ⫽ 163).
501
RUMINATION, EMOTION, AND FORGIVENESS

1). These coefficients show that rumination on Day i ⫺ 1 was
associated with avoidance motivation, revenge motivation, and
anger toward the transgressor (but not fear toward the transgressor)
on Day i. The fourth column of coefficients, labeled ␤(yx.m),
demonstrate that the unique associations of the outcomes y (avoid-
ance and revenge motivation on Day i) with the predictor x
(rumination on Day i ⫺ 1) when controlling for the presumed
mediator m were nonsignificant when anger was the presumed
mediator variable. Because fear of the transgressor was not asso-
ciated with rumination, we did not estimate equations in which fear
of the transgressor was a presumed mediator. The Sobel tests,
whose t values appear in the last column, are also consistent with
the proposition that rumination on Day i ⫺ 1 preceded an increase
in avoidance and revenge motivation the next Day i and that these
increases in avoidance and revenge motivation may have been
because rumination on Day i ⫺ 1 preceded increased anger toward
the transgressor, which may have then led to increases in avoid-
ance and revenge motivation.
Using Kenny et al.’s (2003) approach to estimating the percent-
age of an association between two variables that appears to be
mediated by a third variable, we estimated that 41.09% of the
association between rumination on Day i ⫺ 1 and avoidance
motivation on Day i appeared to be mediated by anger toward the
transgressor. Likewise, 56% of the association between rumination
on Day i ⫺ 1 and revenge motivation on Day i appeared to be
mediated by anger toward the transgressor.
Alternative Models
Through a series of analogous models to those above, we
evaluated whether avoidance and revenge on a given day were
uniquely associated with rumination on the next day and, if so,
whether those associations appeared to be mediated by the asso-
ciations of avoidance and revenge on a given day with anger
and/or fear of the transgressor on the next day. While controlling
for rumination, fear, and anger on Day i ⫺ 1, avoidance motivation
on Day i ⫺ 1 was positively associated with rumination on Day i
(␤⫽0.09, SE ⫽ 0.03), t(160) ⫽ 2.65, p ⬍ .01. While controlling
for rumination, fear, and anger on Day i ⫺ 1, revenge motivation
on Day i ⫺ 1 was not significantly associated with rumination on
Day i (␤⫽ 0.04, SE ⫽ 0.04), t(160) ⫽ 1.10, p ⫽ .27. Thus, it
appeared that avoidance motivation, but not revenge motivation,
was uniquely associated with lagged changes in rumination.
To explore whether avoidance motivation on Day i ⫺ 1 obtained
its association with rumination on Day i by predicting intermediate
increases in anger or fear of the transgressor, we conducted me-
diational analyses. We found evidence that the association of
avoidance motivation on Day i ⫺1 with rumination on Day i was
mediated by anger on Day i (Sobel t ⫽ 2.58, p ⬍ .01). Avoidance
motivation on Day i ⫺1 was not associated with fear on Day i,so
we did not pursue further the possibility that avoidance on Day i ⫺
1 obtains its association with rumination on Day i via its interme-
diate association with fear on Day i. Thus, we did find some
evidence that avoidance motivation (but not revenge motivation)
on a given day was associated with rumination on the next day and
that this association was mediated by the association of avoidance
motivation on a given day with anger toward the transgressor on
the next day.
Summary of Study 3
In Study 3, we replicated the main findings of Studies 1 and 2
by using data from a 21-day diary study that allowed for an
examination of the cross-lagged associations of rumination and
forgiveness (Cole & Maxwell, 2003; Finkel, 1995; West & Hep-
worth, 1991). With these data, we were able to evaluate the
possibility that rumination on any given day predicted how much
avoidance and revenge motivation one would experience on the
following day. Results were largely consistent with the hypothesis
that ruminating about a transgression on any given day preceded
increases in how motivated people felt to avoid and to seek
revenge against their transgressors on the next day. Much of this
lagged association between rumination and forgiveness could be
attributed to the fact that rumination about the transgression on any
given day preceded increases in how angry participants felt toward
their transgressors on the following day, which may, in turn, have
predicted increases in avoidance and revenge motivation.
In contrast, the evidence that forgiveness precedes changes in
rumination was more mixed: We found evidence that increased
avoidance motivation on a given day preceded increased rumina-
tion on the following day (and that this association was mediated
by association of avoidance motivation with anger toward the
transgressor on the next day), but revenge motivation was not
related to increased rumination on the next day. Taken together,
the results of Study 3 provide slightly better support to the notion
that rumination precedes forgiveness than to the notion that for-
giveness precedes rumination, although support for both hypoth-
eses was uncovered.
General Discussion
Experimental research shows that rumination—whether about
negative interpersonal events or one’s own affective state—
prolongs and exacerbates psychological and interpersonal distress.
When people ruminate about their depressive symptoms or anger,
they get more depressed or angry and stay that way longer (Mor &
Winquist, 2002; Rusting & Nolen-Hoeksema, 1998). Moreover,
when people ruminate about someone who has harmed them, they
become more aggressive than when they distract themselves
(Bushman, 2002).
Although nonexperimental studies can never prove definitively
that the relationship between two variables is a causal one (Cole &
Maxwell, 2003; West & Hepworth, 1991); the present results
suggest that forgiveness can be added to the list of adaptive
psychological processes that are negatively associated with rumi-
nation. Using short-term longitudinal methods and a statistical
model for depicting forgiveness as a process of psychological
change (McCullough et al., 2003), we found that rumination was
related inversely to forgiveness in two ways. First, researchers in
previous work (McCullough et al., 2001) have found that the
extent to which people made progress in reducing their rumination
over the course of days, weeks, and months after a transgression
was strongly related to the extent to which they made progress in
forgiving over time (see Footnote 2). In McCullough et al.’s
(2003) longitudinal conceptualization of forgiveness, this means
that reducing one’s rumination about a transgression was related to
trend forgiveness, or linear reductions in avoidance and revenge,
over time.
502
MCCULLOUGH, BONO, AND ROOT

We also found that on occasions when individuals ruminated
about a transgression to a greater extent than was typical for them,
they also became temporarily more avoidant and more vengeful
toward their transgressors than was typical for them. In addition,
the results of Study 3 suggest that changes in rumination precede
the changes in forgiveness with which they are linked. As Bors-
boom et al. (2003) explained, between-subjects associations are
distinct from within-subjects associations, and without evidence of
within-subjects relations between variables, it is unwarranted to
conceptualize their within-persons relationships between two vari-
ables as causal in nature. Therefore, the fact that we have demon-
strated that the typical individual experiences increases in avoid-
ance and revenge motivation following occasions when he or she
has ruminated more about a transgression than is typical for him or
her provides credence to the possibility that the rumination–
forgiveness relationship is a causal one. In Study 3, evidence for
this causal account was more consistent than for the reverse causal
account—that is, forgiveness precedes reductions in rumination.
Nevertheless, because third variables can always be invoked to
explain associations derived from nonexperimental research, ex-
perimental studies of the rumination–forgiveness relationship will
be indispensable for shedding any definitive light on the causal
status of the rumination–forgiveness association.
The Role of Emotion in the Rumination–Forgiveness
Association
Berkowitz’s (1990) CNA model of aggression and Miller et al.’s
(2003) extension of this model to triggered displaced aggression
suggest that rumination about a transgression reactivates affects
that organize people’s thoughts, motivations, physiology, and mo-
toric behavior so that they can coordinate effective fight-or-flight
responses to provocation. As expected, we found that temporary
increases in rumination were related to temporary changes in
affect—anger most notably. Moreover, we found that these affec-
tive changes could be conceptualized as a psychological mediator
through which rumination is associated with reduced forgiveness.
This account for our findings also fits well with an understand-
ing of forgiveness as a motivational phenomenon (McCullough et
al., 2003, 1998, 1997). Some researchers have proposed that ru-
mination occurs when people perceive an important goal to be
unfulfilled. In such cases, it is argued that ruminative thoughts
serve as cues to redouble one’s efforts to pursue the goal. In this
light, when one ruminates about a transgression, the ruminative
cognitions may signify that important psychological states (viz.,
safety and status) that were threatened by the transgression have
not been fully restored to one’s ideals. Ruminative cognitions
therefore may emerge to energize people in their pursuits of safety
and status, and it may be avoidance and revenge motivations
specifically—two very old adaptations to group living (de Waal,
2000)—that rumination reenergizes to help people in their search
for these important social-psychological goods.
It is also of interest that anger, but not fear, appeared to operate
as a mediator of rumination’s associations with avoidance and
revenge motivation. This is inconsistent with what we presumed
about the emotional specificity of avoidance and revenge motiva-
tions as well as with evidence that fear leads to risk aversion,
whereas anger leads to risk seeking (Lerner & Keltner, 2000,
2001). In previous work, we have assumed that a vengeful stance
is associated with an angry, approach-related (i.e., risk seeking)
desire to harm one’s transgressor in kind, whereas an avoidant
stance was associated with a fearful (i.e., risk averse) desire to
maintain a safe distance from one’s transgressor (McCullough et
al., 1998, 1997). However, the present findings cast doubt on this
interpretation: In all three studies, anger appeared to mediate the
associations of rumination with revenge and avoidance motivation
alike. The utility of anger, but not fear, as a potential mediator of
the rumination–forgiveness association is also inconsistent with
Berkowitz’s (1990) neoassociationist model of the relations be-
tween affect and aggression, which proposes that negative affect of
any sort should increase the accessibility of aggression-related
constructs.
It may simply be the case that rumination does more to make
people angry toward their transgressors than it does to make them
fearful of their transgressors— especially when the ruminative
thoughts are about the types of transgressions that our participants
experienced (mostly violations of trust or faithfulness in close
relationships). Thus, different mediational results may have oc-
curred if we had examined the role of anger and fear as mediators
in the context of more severe transgressions, in which participants
had legitimate concerns for their safety or well-being that may
have caused them to fear their transgressors (e.g., intimate partner
violence, imbalanced power relationships, and the like) when they
ruminated. Another possibility is that other negative emotions
associated with the avoidance of aversive stimuli (disgust seems
like a promising candidate) that we did not explore here would
have been better mediators of the rumination–avoidance relation-
ship than the rumination–revenge relationship. Yet a third possi-
bility is that the emotions that elicit avoidance motivation and
revenge motivation are in fact more similar than different (cf.
McCullough et al., 2003; McCullough & Hoyt, 2002; McCullough
et al., 1998, 1997).
Limitations and Directions for Future Work
Because the data we examined herein were nonexperimental, the
causal status of the relationships among rumination, affect, and
forgiveness cannot be established definitively without experimen-
tal research. This is especially noteworthy because earlier levels of
avoidance motivation predicted later levels of rumination, suggest-
ing that the relations between rumination and forgiveness may be
reciprocal or even the result of unmeasured third variables. Future
research on this topic could be improved also by directly measur-
ing the supposed behavioral sequelae of rumination and forgive-
ness (viz., avoidance and revenge behavior) and by studying the
possible motivational effects of revenge and avoidance through
other channels (e.g., facial expressions, physiological functioning,
and actual behavior). Finally, it would be a great boon to forgive-
ness theory to arrive at a more formal understanding of how
specific emotions influence forgiveness. The present results sug-
gest that anger precedes reduced forgiveness, and other research
suggests that empathy facilitates forgiveness (McCullough et al.,
1998, 1997). Evolutionary models of emotion and social behavior
may be particularly useful in generating predictions regarding the
role of specific emotions in forgiveness (Keltner & Haidt, 1999).
Studies incorporating considerations such as these that we have
listed could help to clarify how rumination obtains its negative
503
RUMINATION, EMOTION, AND FORGIVENESS

association with forgiveness, how emotion may be implicated, and
the possible consequences of rumination and forgiveness for per-
sonal and interpersonal functioning.
References
Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator distinction
in social psychological research: Conceptual, strategic, and statistical
considerations. Journal of Personality and Social Psychology, 51, 1173–
1182.
Baumeister, R. F., Exline, J. J., & Sommer, K. L. (1998). The victim role,
grudge theory, and two dimensions of forgiveness. In E. L. Worthington
(Ed.), Dimensions of forgiveness: Psychological research and theolog-
ical perspectives (pp. 79 –104). Radnor, PA: Templeton Foundation
Press.
Berkowitz, L. (1990). On the formation and regulation of anger and
aggression: A cognitive-neoassociationistic analysis. American Psychol-
ogist, 45, 494 –503.
Berry, J. W., Worthington, E. L., O’Connor, L. E., Parrott, L., III., &
Wade, N. G. (2005). Forgivingness, vengeful rumination, and affective
traits. Journal of Personality, 73, 183–225.
Berry, J. W., Worthington, E. L., Parrott, L., III., O’Connor, L. E., &
Wade, N. G. (2001). Dispositional forgivingness: Development and
construct validity of the Transgression Narrative Test of Forgivingness
(TNTF). Personality and Social Psychology Bulletin, 27, 1277–1290.
Borsboom, D., Mellenbergh, G. J., & van Heerden, J. (2003). The theo-
retical status of latent variables. Psychological Review, 110, 203–219.
Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchical linear models.
Newbury Park, CA: Sage.
Bushman, B. J. (2002). Does venting anger feed or extinguish the flame?
Catharsis, rumination, distraction, anger and aggressive responding.
Personality and Social Psychology Bulletin, 28, 724 –731.
Caprara, G. V. (1986). Indicators of aggression: The Dissipation-
Rumination Scale. Personality and Individual Differences, 7, 763–769.
Cole, D. A., & Maxwell, S. E. (2003). Testing mediational models with
longitudinal data: Questions and tips in the use of structural equation
modeling. Journal of Abnormal Psychology, 112, 558 –577.
Collins, L. M., Schafer, J. L., & Kam, C. M. (2001). A comparison of
inclusive and restrictive strategies in modern missing-data procedures.
Psychological Methods, 6, 330 –351.
de Waal, F. B. M. (2000, July 28). Primates–A natural heritage of conflict
resolution. Science, 289, 586 –590.
Enright, R. D. (2001). Forgiveness is a choice. Washington, DC: American
Psychological Association.
Enright, R. D., & Coyle, C. T. (1998). Researching the process model of
forgiveness within psychological interventions. In E. L. Worthington
(Ed.), Dimensions of forgiveness: Psychological research and theolog-
ical perspectives (pp. 139 –161). Philadelphia: Templeton Foundation
Press.
Finkel, S. E. (1995). Causal analysis with panel data. Thousand Oaks, CA:
Sage.
Green, A. S., Rafaeli, E., Bolger, N., Shrout, P. E., & Reis, H. T. (2006).
Paper or plastic? Data equivalence in paper and electronic diaries.
Psychological Methods, 11, 87–105.
Harmon-Jones, E., Vaughn-Scott, K., Mohr, S., Sigelman, J., & Harmon-
Jones, C. (2004). The effect of manipulated sympathy and anger on left
and right frontal cortical activity. Emotion, 95–101.
Hedeker, D. (2004). An introduction to growth modeling. In D. Kaplan
(Ed.), The Sage handbook of quantitative methodology for the social
sciences (pp. 215–234). Thousand Oaks, CA: Sage.
Horowitz, M. J., Wilner, N., & Alvarez, W. (1979). Impact of Event Scale:
A measure of subjective stress. Psychosomatic Medicine, 41, 209 –218.
Hunter, J. E., & Schmidt, F. L. (1990). Methods of meta-analysis: Cor-
recting error and bias in research findings. Newbury Park, CA: Sage.
Keltner, D., & Haidt, J. (1999). Social functions of emotions at four levels
of analysis. Cognition & Emotion, 13, 505–521.
Kenny, D. A., Korchmaros, J. D., & Bolger, N. (2003). Lower level
mediation in multilevel models. Psychological Methods, 8, 115–128.
Konecni, V. J. (1974). Self-arousal, dissipation of anger, and aggression.
Personality and Social Psychology Bulletin, 1, 192–194.
Korchmaros, J. D., & Kenny, D. A. (2003). Step by step procedure for
estimating lower-level mediation in random-effects multilevel models
using HLM5. Unpublished manuscript.
Krull, J. L., & MacKinnon, D. P. (2001). Multilevel modeling of individual
and group level mediated effects. Multivariate Behavioral Research, 36,
249 –277.
Lerner, J. S., & Keltner, D. (2000). Beyond valence: Toward a model of
emotion-specific influences on judgement and choice. Cognition &
Emotion, 14, 473– 493.
Lerner, J. S., & Keltner, D. (2001). Fear, anger, and risk. Journal of
Personality and Social Psychology, 81, 146 –159.
McCullough, M. E., Bellah, C. G., Kilpatrick, S. D., & Johnson, J. L.
(2001). Vengefulness: Relationships with forgiveness, rumination, well-
being, and the Big Five. Personality and Social Psychology Bulletin, 27,
601– 610.
McCullough, M. E., Fincham, F. D., & Tsang, J. (2003). Forgiveness,
forbearance, and time: The temporal unfolding of transgression-related
interpersonal motivations. Journal of Personality and Social Psychol-
ogy, 84, 540 –557.
McCullough, M. E., & Hoyt, W. T. (2002). Transgression-related motiva-
tional dispositions: Personality substrates of forgiveness and their links
to the Big Five. Personality and Social Psychology Bulletin, 28, 1556 –
1573.
McCullough, M. E., Orsulak, P., Brandon, A., & Akers, L. (2007). Rumi-
nation, fear, and cortisol: An in vivo study of interpersonal transgres-
sions. Health Psychology, 26, 126 –132.
McCullough, M. E., Rachal, K. C., Sandage, S. J., Worthington, E. L.,
Brown, S. W., & Hight, T. L. (1998). Interpersonal forgiving in close
relationships. II: Theoretical elaboration and measurement. Journal of
Personality and Social Psychology, 75, 1586 –1603.
McCullough, M. E., & Worthington, E. L. (1994). Models of interpersonal
forgiveness and their applications to counseling: Review and critique.
Counseling and Values, 39, 2–14.
McCullough, M. E., Worthington, E. L., & Rachal, K. C. (1997). Inter-
personal forgiving in close relationships. Journal of Personality and
Social Psychology, 73, 321–336.
Miller, N., Pedersen, W. C., Earleywine, M., & Pollock, V. E. (2003). A
theoretical model of triggered displaced aggression. Personality and
Social Psychology Review, 7, 75–97.
Mor, N., & Winquist, J. (2002). Self-focused attention and negative affect:
A meta-analysis. Psychological Bulletin, 128, 638 – 662.
Nolen-Hoeksema, S. (1987). Sex differences in unipolar depression: Evi-
dence and theory. Psychological Bulletin, 101, 259 –282.
Nolen-Hoeksema, S., & Jackson, B. (2001). Mediators of the gender
difference in rumination. Psychology of Women Quarterly, 25, 37– 47.
Raudenbush, S. W., Brennan, R. T., & Barnett, R. C. (1995). A multivar-
iate hierarchical model for studying psychological change within cou-
ples. Journal of Family Psychology, 9, 161–174.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models:
Applications and data analysis methods (2nd ed.). Thousand Oaks, CA:
Sage.
Raudenbush, S. W., Bryk, A., Cheong, Y. F., & Congdon, R. (2000). HLM
5: Hierarchical linear and nonlinear modeling. Lincolnwood, IL: Sci-
entific Software International.
Rusting, C. L., & Nolen-Hoeksema, S. (1998). Regulating responses to
anger: Effects of rumination and distraction on angry mood. Journal of
Personality and Social Psychology, 74, 790 – 803.
504
MCCULLOUGH, BONO, AND ROOT

Schafer, J. L., & Graham, J. W. (2002). Missing data: Our view of the state
of the art. Psychological Methods, 7, 147–177.
Skinner, E. A., Edge, K., Altman, J., & Sherwood, H. (2003). Searching for
the structure of coping: A review and critique of category systems for
classifying ways of coping. Psychological Bulletin, 129, 216 –269.
Sobel, M. E. (1982). Asymptotic intervals for indirect effects in structural
equations models. In S. Leinhart (Ed.), Sociological methodology 1982
(pp. 290 –312). San Francisco: Jossey-Bass.
Stone, A. A., Shiffman, S., Schwartz, J. E., Broderick, J. E., & Hufford,
M. R. (2003). Patient compliance with paper and electronic diaries.
Controlled Clinical Trials, 24, 182–199.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and vali-
dation of brief measures of positive and negative affect: The PANAS
scales. Journal of Personality and Social Psychology, 54, 1063–1070.
West, S. G., & Hepworth, J. T. (1991). Statistical issues in the study of
temporal data: Daily experiences. Journal of Personality, 59, 609 – 662.
Worthington, E. L. (2001). Five steps to forgiveness: The art and science
of forgiving. New York: Crown House Publishing.
Received October 14, 2005
Revision received August 3, 2006
Accepted August 14, 2006 䡲
505
RUMINATION, EMOTION, AND FORGIVENESS