Content uploaded by John G Holmes
Author content
All content in this area was uploaded by John G Holmes on Jan 08, 2014
Content may be subject to copyright.
More Than Words: Reframing Compliments From Romantic Partners
Fosters Security in Low Self-Esteem Individuals
Denise C. Marigold, John G. Holmes, and Michael Ross
University of Waterloo
Although people with low self-esteem (LSEs) doubt their value to their romantic partners, they tend to
resist positive feedback from their partners. This resistance undermines their relationships and has been
difficult to overcome in past research. The authors investigated whether LSEs could be induced to take
their partners’ kind words to heart by manipulating how abstractly they described a recent compliment.
In 3 studies, LSEs felt more positively about the compliments, about themselves, and about their
relationships—as positively as people with high self-esteem (HSEs) felt—when they were encouraged to
describe the meaning and significance of the compliments. The effects of this abstract meaning
manipulation were still evident 2 weeks later. Thus, when prompted, LSEs can reframe affirmations from
their partners to be as meaningful as HSEs generally believe them to be and, consequently, can feel just
as secure and satisfied with their romantic relationships.
Keywords: romantic relationships, relationship security, self-esteem, positive feedback, memory
“Does he love me? I want to know
How can I tell, if he loves me so?”
—Betty Everett, The Shoop Shoop Song, 1964
Romantic relationships have enormous potential to be self-
affirming. Whether relationships fulfill this potential, however,
depends in part on people’s confidence that their partners truly
love and value them. One might intuit that people who are rela-
tively uncertain that they are loved would be particularly eager to
embrace positive feedback from their partners. Considerable re-
search suggests the opposite: Although people with low self-
esteem (LSEs) have more doubts about their partners’ love than do
people with high self-esteem (HSEs), they are less likely to benefit
from their partners’ praise. In this article, we review previous
research demonstrating and explaining this paradox and then
present three experiments showing how LSEs can be persuaded to
take their partners’ kind words to heart.
Self-Esteem (SE), Unwarranted Insecurities, and
Dependence Regulation
LSEs tend to report less satisfaction than do HSEs in both
marital and dating relationships (Fincham & Bradbury, 1993;
Murray, Holmes, & Griffin, 1996). There is no evidence, however,
to suggest that they attract less desirable partners (Murray et al.,
1996; Murray, Holmes, & Griffin, 2000). Instead, LSEs’ relatively
negative evaluations of their partners seem to be self-protective
responses to their beliefs that their partners think relatively nega-
tively of them (Murray et al., 2000; Murray, Holmes, Griffin,
Bellavia, & Rose, 2001). However, LSEs underestimate how much
they are actually loved by their partners, and, in reality, they are
loved just as much as are HSEs (Murray et al., 2000, 2001).
The consequences of these unwarranted insecurities for relation-
ship well-being are well documented. Doubtful LSEs search for
information about their partners’ caring in a biased manner, being
too quick to perceive and generalize from signs of possible rejec-
tion (Murray, Rose, Bellavia, Holmes, & Kusche, 2002). LSEs
react to various potential threats (e.g., guilt about a transgression,
a conflict with their partner) with reduced security in their part-
ners’ acceptance. When feeling insecure, they regulate their de-
pendence on the relationship by devaluing it and by distancing
themselves from their partner. On the other hand, HSEs tend to
respond to threats by embellishing their partners’ acceptance and
drawing closer to the relationship (Murray, Bellavia, Rose, &
Griffin, 2003; Murray, Holmes, MacDonald, & Ellsworth, 1998;
Murray et al., 2002). Ironically, LSEs are initially accepted and
valued by their romantic partners as much as are HSEs, but their
oversensitivity to rejection manifests in defensive behaviors that
over time, tarnish their partners’ rosy views and, ultimately, un-
dermine the well-being of the relationship (Murray, Bellavia, et al.,
2003; Murray, Griffin, Rose, & Bellavia, 2003).
SE and Interpreting Positive Behaviors
LSEs report wanting their romantic partners to see them much
more positively than they see themselves (Murray et al., 2000). Yet
their consistent underestimation of how much their partners actu-
ally love and value them is hard to overcome because they fail to
seize opportunities to enhance their feelings of acceptance. Just as
Denise C. Marigold, John G. Holmes, and Michael Ross, Department of
Psychology, University of Waterloo, Waterloo, Ontario, Canada.
This research was supported by Social Sciences and Humanities Re-
search Council grants awarded to John G. Holmes and Michael Ross. We
thank Anna Blackwell, Jennifer Cunningham, Danielle Gaucher, Manika
Khanna, and Lisa-Dawn Wismer for their assistance in conducting this
research. We also thank Ian McGregor and Joanne Wood for their very
helpful comments on a draft of this article.
Correspondence concerning this article should be addressed to Denise
C. Marigold, Department of Psychology, University of Waterloo, 200
University Avenue West, Waterloo, Ontario, N2L 3G1 Canada. E-mail:
dcmarigo@watarts.uwaterloo.ca
Journal of Personality and Social Psychology Copyright 2007 by the American Psychological Association
2007, Vol. 92, No. 2, 232–248 0022-3514/07/$12.00 DOI: 10.1037/0022-3514.92.2.232
232
LSEs overgeneralize the implications of failure, but not success,
for their overall self-evaluation (Brown & Dutton, 1995; Kernis,
Brockner, & Frankel, 1989), they readily make generalizations
about the (in)security of their partners’ love only after negative and
not positive events. LSEs react to threats to SE by becoming less
certain about their partners’ acceptance, but they do not report
feeling more accepted after boosts to SE (Murray, Griffin, et al.,
2003).
Several researchers have reported findings that suggest that
LSEs are too hesitant to read abstract, generalized meaning into
their partners’ positive behaviors. They have used moderating
variables that have been shown to be reasonable proxies for LSE,
such as attachment anxiety (Brennan & Morris, 1997; Collins &
Read, 1990) and dissatisfaction (Fincham & Bradbury, 1993;
Murray et al., 1996). In one study, anxiously attached individuals
rated their partners’ behavior as less supportive than did secure
participants in a stressful situation, but observers rated the partners
as equally supportive (Collins & Feeney, 2004). In another study,
LSEs reported seeing significantly fewer acceptance cues than did
HSEs when they believed that an interaction partner’s behavior
was directed at them. When they thought the same behavior was
directed at someone else, however, LSEs saw as many acceptance
cues as did HSEs (Cameron, Anthony, Gaetz, & Balchen, 2006).
When asked directly to make attributions for positive partner
behaviors, dissatisfied individuals were less likely to report that
such behaviors were global in their implications or likely to recur
(Camper, Jacobson, Holtzworth-Munroe, & Schamling, 1988),
even though they reported that these behaviors had a positive
impact on them (Holtzworth-Munroe & Jacobson, 1985). Relative
to securely attached individuals, anxiously attached individuals
also have been shown to make more relationship-threatening at-
tributions for their partners’ positive behaviors, for example, that
their partner was motivated by selfish concerns rather than genuine
love and affection. However, anxious individuals were no less
likely than secures to report being happy about these behaviors
(Collins, Ford, Guichard, & Allard, 2006). People who are dis-
tressed about their romantic relationships do seem to appreciate
positive feedback from their partners. However, their unwilling-
ness to make meaningful generalizations from positive feedback
interferes with their ability to fully benefit from it.
Why are the people most in need of acceptance, such as LSEs
(Leary, Tambor, Terdal, & Downs, 1995; Rudich & Vallacher,
1999), the least likely to find it? Drawing from Murray, Holmes,
and Collins’s (2006) risk regulation model, we suggest that LSEs
tend to minimize the meaning of positive events in their relation-
ship to avoid being hurt if it turns out that they were not really
loved after all. LSEs’ uncertainty about their worth leads them to
err on the side of caution, taking a risk-averse orientation to
appraising the meaning of events in their relationships. That is,
they are more ready to generalize from signs of rejection than to
trust signs of acceptance (Murray et al., 2006; see also L. Camp-
bell, Simpson, Boldry, & Kashy, 2005, for a similar conclusion
regarding anxiously attached individuals). One problem with this
self-protective orientation
1
is that the strongest, most satisfying
relationships are those in which partners can set aside self-
protection goals and risk behaving in relationship-promotive ways
that increase closeness and dependence (Murray et al., 2006). The
research of Murray and her colleagues has focused on LSEs’
self-protective tendency to distance themselves from their partners
in response to perceived signs of rejection. In the present research,
we examined LSEs’ hesitancy to trust signs of acceptance, and
how overcoming this hesitancy could facilitate relationship-
promotive behaviors.
SE and Reactions to Positive Feedback
How might LSEs be helped to recognize that they are valued by
their partners? The most direct approach would be simply to tell
them that they possess desirable attributes. Having less positive
and certain self-conceptions than HSEs (J. D. Campbell et al.,
1996), LSEs might be expected to embrace such feedback. And
there is no evidence to suggest that they are any less likely than
HSEs to possess desirable attributes, as objective measures show
they are equally well-liked and intelligent (Baumeister, Campbell,
Krueger, & Vohs, 2003). But telling people how well-liked or
intelligent they are evokes quite different reactions from LSEs and
from HSEs.
For LSEs, positive feedback has no impact, or it backfires and
activates relationship-damaging dependence regulation processes.
After success on a bogus intelligence test, compared with neutral
or no feedback, LSEs felt more anxious about being accepted by
their family, friends, and romantic partners (Logel, Spencer,
Wood, Holmes, & Zanna, 2006; Murray et al., 1998). Positive
feedback on a relationship trait had similar effects: LSEs felt more
insecure about their relationship and evaluated it more negatively
after being told that they were high in “considerateness” (Murray
et al., 1998). Similar results have been found with attachment
anxiety as a moderating variable: Compared with a no feedback
condition, anxious participants paradoxically viewed their relation-
ships more pessimistically after having been told that their partners
viewed them as exceptionally warm, attractive, and intelligent
(Peach & MacDonald, 2004). In contrast, in all of these studies,
positive feedback benefited secure, HSE participants.
Why do LSEs react unfavorably when they are directly told that
they are worthy and desirable individuals? Consistent with other
researchers’ speculations (Wood, Anthony, & Foddis, 2006), we
suggest that positive feedback highlights self-discrepant standards
and activates self-evaluation concerns for LSEs. Because LSEs
doubt their possession of positive attributes, they have difficulty
believing that others see such positive attributes in them. They may
worry that success will raise others’ expectations of them, which
they are not confident about meeting in the future (Logel et al.,
2006; Murray et al., 1998). They may question what they must be
lacking if, despite possessing such positive qualities, their partners
still do not value them. Further, simply focusing on strengths may
remind LSEs of their faults (Showers, 1992) and highlight that
acceptance from their partners is conditional on their maintenance
of virtues (Baldwin & Sinclair, 1996). Therefore, LSEs become
anxious that they cannot live up to such a positive self-image and
imagine disappointing and being rejected by close others when the
truth is revealed. This interpretation is consistent with self-
1
The tendency of LSEs to engage in self-protection has also been
documented in a number of domains outside of romantic relationships.
They would like to be recognized as having desirable traits in general, not
just those desired by romantic partners, but they are cautious to claim them
in order to minimize the chance of being humiliated when they fail to
demonstrate them consistently (Baumeister, Tice, & Hutton, 1989).
233
REFRAMING COMPLIMENTS FROM ROMANTIC PARTNERS
verification theory’s notion of pragmatic concerns, which suggests
that people with negative self-views expect to have more difficult
interactions with relationship partners who see them overly posi-
tively and hence fail to recognize their limitations (Swann, Rent-
frow, & Guinn, 2003).
It may be simply too much of a risk, then, for LSEs to conclude
that their partners’ positive regard and love are genuine, secure,
and enduring when they can easily think of personal weaknesses
that will most assuredly come to light and jeopardize their accep-
tance (cf. Murray et al., 2006). The only past attempts that suc-
ceeded in making LSEs feel better about their relationships seem
to have bypassed participants’ self-evaluation concerns in one way
or another (e.g., describing a value they shared with their partner,
Lomore, Spencer, & Holmes, in press; focusing on a fault in their
partner; Murray et al., 2005). The goal of the current research is to
develop a cognitive reframing technique that LSEs could use to
generalize from partner affirmations without activating potentially
destructive self-evaluative processes.
In all studies, participants recalled a compliment that they had
received from their romantic partner. We believed that LSEs
would normally view compliments as relatively concrete, isolated,
past events and would not make meaningful generalizations about
how much they are valued. In the critical condition in each study,
we induced participants to describe the compliment in an abstract
fashion: They explained why their partner admired them, what the
compliment meant to them, and what significance it had for their
relationship. We expected the abstract manipulation to induce
LSEs to form a more global interpretation of the compliment
without engaging their relationship-damaging dependence regula-
tion tactics. The novel aspect of our approach is that it helps LSEs
to help themselves. It may be empowering for them to learn how
to meaningfully reframe their partners’ affirmations in order to
assuage their doubts about their partners’ love for them.
Overview of Experiments
In three experiments, participants first recalled a compliment
that their romantic partner had paid them (which pilot testing
suggested was not more difficult for LSEs than for HSEs). We then
manipulated how they described the compliment. In Study 1, we
tested whether an abstract description would indeed be more
effective than a concrete description in making LSEs feel good
about the positive feedback. In Study 2 we added a no-instructions
control condition to clarify the direction of the abstraction effects
and to examine SE differences in spontaneous descriptions of
compliments. We also extended the dependent variables to include
perceptions of relationship quality. In Study 3, we included a
condition that encouraged participants to question whether they
could rightfully apply an abstract generalization to the specific
compliment they recalled, in contrast to instructing them to make
the generalization, to investigate whether the mere suggestion of
abstract meaning was sufficient to boost LSEs.
Study 1
In Study 1, we examined whether being instructed to describe a
past compliment from their romantic partner in an abstract (vs.
concrete) manner would make LSEs feel more positively about the
compliment and consequently more valuable generally. We also
had participants think about the compliment again 2 weeks after
the initial session to determine whether the effects of the cognitive
reframing persisted. We did not expect the reframing manipulation
to affect HSEs as much as LSEs. HSEs should be sufficiently
confident in their partners’ positive regard to generalize from signs
of acceptance without experimental encouragement.
Method
Participants and Procedure
Fifty-three undergraduate students
2
(17 men and 36 women) in
romantic relationships participated in a study of “Relationship
Events” in exchange for course credit. Mean age was 19 years, and
mean relationship length was 20 months. The majority of participants
(42) indicated that they were in an “exclusive dating” relationship. Six
individuals reported their relationship status as “casual dating,” 2 were
“living together,” 1 was “engaged,” and 2 were “married.”
Between 1 and 4 participants reported for each session and were
randomly assigned to a condition upon arrival. First, they completed
an SE scale and some filler questionnaires. Next they were asked to
think of a time their partner said something nice to them and to report
the compliment with a few cue words. The manipulation instructions
appeared on the next page; following the instructions, participants
described the compliment in detail. They then answered several
questions about the compliment itself and completed a measure of
state SE. Finally, they answered some demographic questions.
At the end of the study, participants wrote down their e-mail
address if they were interested in completing a brief follow-up
questionnaire 2 weeks later for a chance to win a cash prize.
Forty-nine agreed to be contacted and 33 returned the question-
naire via e-mail. Four were excluded from follow-up analyses
because they had broken up with their partner since completing
Part 1. Thus, the Part 2 sample included 29 participants (8 men and
21 women), or 55% of the original sample. The two samples did
not differ on SE or on any of the dependent variables.
Part 1 Materials
Trait SE. Participants responded to the 10 Rosenberg (1965)
SE items on a scale ranging from 1 (strongly disagree)to7
(strongly agree). Cronbach’s alpha was .93.
3
2
The initial sample included 70 participants, 17 of whom identified
themselves as Asian when asked for their ethnicity. We decided to elimi-
nate Asian participants from our analyses after discovering that they had
significantly lower SE. We were concerned that the mean differences did
not reflect more negative self-views but rather a different interpretation of
the items on the scales (Heine, Lehman, Markus, & Kitayama, 1999).
Asians were also excluded from analyses in Study 2 and were not selected
for Study 3. Results were weaker, but generally in the same direction, if
Asians were included in the analyses.
3
Participants also completed two attachment style questionnaires (an
adapted version of Brennan, Clark, & Shaver’s, 1998, Experience in Close
Relationship Scale and Bartholomew & Horowitz’s, 1991, four-category
measure of attachment style) and a measure of perceived regard (Interper-
sonal Qualities Scale; Murray et al., 1996). These measures were included
both for exploratory purposes and to prevent participants from focusing on
the SE measure as our primary moderating variable. The attachment style
questionnaires were also included in Studies 2 and 3. These scales did not
yield strong or consistent results and so are not discussed further.
234
MARIGOLD, HOLMES, AND ROSS
Compliment manipulation. All participants received the fol-
lowing instructions: “Think of a time when your partner told you
how much he/she liked something about you. For example, a
personal quality or ability you have that he/she thinks very highly
of, or something you did that really impressed him/her.” They
were asked to write down a few cue words that would identify that
memory to them, note how long ago it occurred, then turn to the
next page to describe the compliment more fully.
In the concrete (N ⫽ 25) condition, participants were asked to
“Describe exactly what your partner said to you. Include any
details you can recall about where you two were at the time, what
you were doing, what you were both wearing, etc.” In the abstract
condition (N ⫽ 28), participants were asked to “Explain why your
partner admired you. Describe what it meant to you and its
significance for your relationship.” We portrayed the partners’
positive behavior in terms of “said” (in the concrete condition) vs.
“admired” (in the abstract condition) on the basis of Semin and
Fiedler’s (1988) linguistic category model (see also Semin & De
Poot, 1997). In contrast to an action verb like “said,” a state verb
like “admired” implies that the behavior has lasted for a longer
period and is more likely to recur in the future. We hoped that the
use of a state verb would lead LSEs to perceive the compliments
to be more broad and global in their implications about their value
to their partners.
Compliment questions. Participants were asked how happy
they felt when recalling the compliment, how far away the com-
pliment felt, how significant it seemed now, and how much they
attributed the compliment to something about their partner, and to
the situation, all on 7-point scales.
Positive expectations. Three items assessed how likely partic-
ipants thought it was that their partner would behave in various
positive ways toward them in the future (e.g., “How likely is it that
your partner would make an extra effort to spend time with you
when his/her schedule was especially busy?”). The items were
rated on a scale ranging from 1 (extremely unlikely)to7(extremely
likely) and were averaged to form one index with ␣⫽.52.
State SE. Participants indicated how they felt about them-
selves “right now, at this moment” on ten 7-point, bipolar adjective
scales (e.g., accepted–rejected, unimportant–important). Items
were reverse scored when appropriate and were averaged to create
a measure of state SE (␣⫽.94; adapted from McFarland & Ross,
1982). State SE captures temporary fluctuations in self-evaluations
caused by receipt of positive or negative information about the
self, whereas trait SE, as measured by the Rosenberg Self-Esteem
Scale in this study, captures average or typical self-evaluations
across time (Heatherton & Polivy, 1991).
Part 2 Materials
All participants, regardless of the condition to which they had
been assigned in Part 1, received the same instructions and ques-
tions 2 weeks later in Part 2. First, they were reminded that in the
previous questionnaire they were asked to “Think of a time when
your partner told you how much he/she liked something about
you.” Participants’ cue words were inserted into their version of
the questionnaire. Recall that these cue words were recorded prior
to the manipulation, so there was no cue to elicit abstract meaning
or concrete details presented here. They were then asked to
“Briefly describe the event as you recall it now.” Next they
answered four items about how they were currently feeling about
the compliment. In addition to how happy the compliment made
them feel, which they rated in Part 1, they also rated how secure,
valuable, and proud the compliment made them feel in Part 2.
Cronbach’s alpha for the 4-item scale was .63. The correlations
between all measured variables are presented in Table 1.
Part 1 and Part 2 Coding
Two independent coders, who were unaware of the conditions of
the study, rated each compliment narrative according to how
abstractly the participant described the compliment on a scale
ranging from 1 (only concrete details)to5(abstract meaning for
relationship). The coders’ ratings were correlated at .72 for Part 1
and at .57 for Part 2 and were averaged to create a measure of
abstraction for each narrative.
The Linguistic Inquiry and Word Count (LIWC) program (Pen-
nebaker, Francis, & Booth, 2001) was used to assess the extent to
which participants’ descriptions of their partners’ compliments
included past and present tense verbs. The LIWC is a computer-
ized text analysis program that yields percentages of total words
falling under a particular category. We were especially interested
in whether participants in the abstract condition would use more
present tense and less past tense. This would suggest that they were
seeing the compliment as an indication of their partner’s continu-
ing view of them as opposed to an isolated remark in the past. We
recalculated the values yielded by the program to reflect percent-
age of total verbs, rather than total words, which were past or
present tense.
Results
Part 1
For the analyses, all dependent variables were regressed on
condition (abstract vs. concrete), SE, and the Condition ⫻ SE
interaction. Following Aiken and West (1996), condition was
effect-coded and SE was centered to make the mean equal to zero.
Condition was entered on the first step, and SE and the interaction
term were entered on the second step. Simple effects were calcu-
lated at one standard deviation below the mean for LSE and one
standard deviation above the mean for HSE. Temporal distance of
the compliment (M ⫽ 4.97 months) was centered and included as
a covariate.
4
Coding of abstractness. There was a main effect of condition
on abstraction (⫽.68), t(50) ⫽ 6.50, p ⬍ .01. Coders rated the
narratives in the abstract condition as more abstract (M ⫽ 3.46)
than the narratives in the concrete condition (M ⫽ 1.92). The
Condition ⫻ SE interaction was not significant ( p ⬎ .55). Both
LSEs and HSEs heeded our instructions for describing the com-
pliment.
Compliment questions. There was a main effect of condition,
⫽.27, t(50) ⫽ 2.06, p ⬍ .05, on happiness about the compli-
ment. Participants in the abstract condition reported being happier
about the compliment than did participants in the concrete condi-
4
There were no significant main effects of gender or any interactions
between condition and gender on any of the dependent variables in any of
the three studies. Thus, gender is not discussed further.
235
REFRAMING COMPLIMENTS FROM ROMANTIC PARTNERS
tion. Although the Condition ⫻ SE interaction did not reach
significance, ⫽⫺.19, t(48) ⫽⫺1.28, p ⫽ .20, we explored
simple effects pertinent to our investigation. The results are pre-
sented in Table 2. LSEs reported significantly greater happiness
about the compliment in the abstract condition than in the concrete
condition, ⫽.43, t(48) ⫽ 2.04, p ⬍ .05, whereas HSEs were
equally happy in both conditions. The simple slope of SE was
significant in the concrete condition, ⫽.37, t(48) ⫽ 2.49, p ⬍
.05, but not in the abstract condition (⫽⫺.02, ns). As expected,
LSEs felt happier when describing the compliment abstractly than
when describing it concretely, as happy as HSEs felt in both
conditions. The reliability of this effect may be questionable be-
cause of the nonsignificant overall interaction, but it is reassessed
in the next two studies with a multiitem measure. There were no
effects of condition or interactions on any of the other items
assessing reactions to the compliment or on the measure of general
positive expectations.
State SE. Would LSEs’ increased happiness about their part-
ners’ admiration of a specific attribute translate to increased feel-
ings of worth more generally? As with happiness about the com-
pliment, for state SE there was a significant main effect of
condition, ⫽.31, t(50) ⫽ 2.35, p ⬍ .05, as well as a main effect
of SE, ⫽.67, t(48) ⫽ 6.74, p ⬍ .01, which were qualified by a
marginal Condition ⫻ SE interaction, ⫽⫺.17, t(48) ⫽⫺1.73,
p ⬍ .10 (see Table 2 for predicted values). LSEs reported higher
state SE in the abstract condition than in the concrete condition,
⫽.35, t(48) ⫽ 2.56, p ⬍ .05, but HSEs did not differ between
conditions. The simple slope of SE was significant in both the
abstract condition, ⫽.49, t(48) ⫽ 2.87, p ⬍ .01, and the concrete
condition, ⫽.84, t(48) ⫽ 8.60, p ⬍ .01. Thus, LSEs felt better
about themselves more generally after describing a specific com-
pliment from their romantic partner in an abstract, as opposed to
concrete manner.
Verb tense. There were significant main effects of condition
on proportion of verbs in present tense, ⫽.45, t(50) ⫽ 3.58, p ⬍
.01, and past tense, ⫽⫺.47, t(50) ⫽⫺3.86, p ⬍ .01. Consistent
with our predictions, participants used more present tense in the
abstract condition than in the concrete condition (M ⫽ 44.02% vs.
M ⫽ 24.38%) and less past tense in the abstract condition than in
the concrete condition (M ⫽ 52.67% vs. M ⫽ 73.89%).
Part 2
The Part 2 sample included 15 participants from the abstract
condition and 14 participants from the concrete condition. Both
temporal distance of the compliment and time between Part 1 and
Part 2 (range 13–28 days; M ⫽ 16.21) were controlled for in the
analyses by centering them and including them as covariates.
Positive feelings. When reminded of the compliment they
wrote about 2 weeks earlier and by using only the cue words they
came up with prior to the manipulation, would we find any
evidence that participants had reframed the compliments in their
memories in accordance with the condition they had been assigned
to in Part 1? Results revealed no main or interaction effects on
participants’ positive feelings about the event. However, internal
analyses showed a marginally significant positive correlation be-
tween SE and positive feelings in the concrete condition (r ⫽ .48,
p ⬍ .09) but not in the abstract condition (r ⫽ .08, ns), which is
consistent with the simple effects we reported for happiness in Part
1; an SE difference existed in the concrete condition, which we
would expect, but was eliminated by the abstract condition.
Coding. Participants who had been in the abstract condition in
Part 1 continued to use more present tense, ⫽.44, t(21) ⫽ 2.40,
p ⬍ .05 (M ⫽ 36.18%), and less past tense, ⫽⫺.42, t(21) ⫽
⫺2.34, p ⬍ .05 (M ⫽ 59.00%), in their narratives than did
participants who had been in the concrete condition (M ⫽ 11.56%
and M ⫽ 85.41%, respectively). These findings suggest that the
reframing of the compliments in Part 1 continued to influence
participants’ construal of the compliments 2 weeks later. There
were no main or interaction effects on coders’ ratings of abstract-
ness for the Part 2 narratives.
Discussion
In Study 1, LSEs felt just as happy as HSEs about compliments
from their romantic partners when they explained the meaning and
significance of the compliments, rather than when they described
the concrete details of exactly what was said and where. This had
implications for how valuable they felt more generally; LSEs
reported higher state SE in the abstract condition than in the
concrete condition. The verb tense findings suggest one reason the
manipulation was effective: Thinking of a partner’s compliment as
meaningful and significant may convey that it was the result of an
ongoing experience as opposed to a one-time exclamation of
praise. We were encouraged to discover that some of the differ-
ences between the concrete and abstract narratives were still evi-
dent when participants were reminded of the compliment 2 weeks
Table 1
Correlations Among Measured Variables: Study 1
Variable 1 2 3 4
Part 1
1. Trait self-esteem —
2. Happiness .33
*
—
3. State self-esteem .79
**
.58
**
—
Part 2
4. Positive feelings .36
*
.50
**
.42
*
—
*
p ⬍ .05.
**
p ⬍ .01.
Table 2
Predicted Values for Concrete and Abstract Conditions at Low
and High Self-Esteem (SE): Study 1
Dependent variable
Low SE High SE
Concrete Abstract Concrete Abstract
Part 1
Happiness 5.45
a
6.34
b
6.23
b
6.30
b
State self-esteem 4.83
a
5.50
b
6.43
c
6.44
c
Note. Simple effects test were conducted comparing participants with
low self-esteem (LSEs) in the concrete condition with LSEs in the abstract
condition, participants with high self-esteem (HSEs) in the concrete con-
dition with HSEs in the abstract condition, and LSEs with HSEs within
each condition for each dependent variable. In each row, predicted scores
for the cells that were compared that do not share subscripts differ at p ⬍
.05. Low and high SE were calculated at ⫾1 SD.
236
MARIGOLD, HOLMES, AND ROSS
later and asked to describe it as they recalled it at that time. We
reminded them by using their own cue words, which they gener-
ated prior to the manipulation in Part 1. Thus, the lasting differ-
ences in narrative structure provide preliminary support that we
were able to reframe the meaning of the compliments in people’s
memories. Although some of the findings in Study 1 were weak,
we were optimistic that recruiting larger samples and using more
elaborate measures in Studies 2 and 3 would garner stronger
support for our predictions.
Study 2
Study 2 included a no-instructions control condition that served two
functions. The first was to assess the direction of the effects. On the
basis of the results of Study 1, it is unclear whether the instructions to
describe a romantic partner’s compliment in an abstract manner made
LSEs feel better about the compliment and themselves as we hypoth-
esized or whether the concrete instructions made them feel worse. The
second goal is to show that LSEs would typically (i.e., after no
specific instructions) rate their partners’ positive feedback less posi-
tively than would HSEs, consistent with our argument that they
self-protectively avoid generalizing from signs of acceptance.
Study 2 also included questions assessing feelings about the
relationships more generally. Our main hypothesis is that LSEs
would rate their relationships more negatively than would HSEs
(Murray et al. 1996, 2000), unless inspired to optimism by abstract
instructions for describing partners’ compliments. We did not
expect HSEs’ ratings to differ between conditions, given their
tendency to spontaneously extract affirming meanings from posi-
tive feedback (Murray et al., 1998).
Method
Participants and Procedure
One hundred twenty-three introductory psychology students in
romantic relationships participated in a two-part study of “Rela-
tionship Perceptions” in exchange for course credit. Four partici-
pants were excluded for not fully completing the materials. The
remaining sample of 119 consisted of 19 men and 100 women.
Mean age was 19 years, and mean relationship length was 17
months. The majority of participants (100) indicated that they were
in an “exclusive dating” relationship. Six individuals reported their
relationship status as “casual dating,” 8 were “living together,” 4
were “engaged,” and 1 was “married.”
The entire study was completed online. The manipulation of com-
pliment abstraction was the same as in Study 1, except we added a
control condition that received no specific instructions as to how to
describe the compliment. Participants then answered several questions
about the compliment and completed a measure of state SE. In this
study we also included several questions about the relationship in
general. The control condition made their relationship ratings prior to
the compliment questionnaire to establish baseline measures of rela-
tionship quality, whereas the remaining participants (abstract and
concrete conditions) made their relationship ratings after the compli-
ment manipulation and questions. In Part 2, participants reported their
thoughts and feelings about their relationship prior to being reminded
of the compliment with their cue words from Part 1 (in Study 1, the
cue words were presented immediately upon beginning the Part 2
questionnaire). We examined whether any increases in perceptions of
relationship quality would endure for the abstract condition when the
compliment was not salient.
All participants were sent the link to the Part 2 questionnaire
approximately 2 weeks after they completed Part 1. The cue words
they wrote to identify the compliment prior to the manipulation
instructions in Part 1 were attached to their participant ID number
and thus automatically inserted into their questionnaire when they
logged on to Part 2. One hundred fifteen participants completed
Part 2. Six were excluded from Part 2 analyses because they had
broken up with their partner since completing Part 1. Thus, the Part
2 sample included 109 participants (15 men and 94 women), or
92% of the original sample.
5
The two samples did not differ in SE.
The Part 2 sample was higher initially in compliment positivity,
felt security, and relationship valuing. However, these differences
are completely accounted for by the participants who broke up
with their partners since Part 1; they do not hold for participants
who simply chose not to complete Part 2.
Part 1 Materials
Trait SE. SE was measured by the Rosenberg (1965) Self-
Esteem Scale (␣⫽.92).
Compliment manipulation. As in Study 1, all participants were
instructed to think of a time when their partner told them some-
thing they liked about them. They received different instructions
about how to describe this compliment according to the condition
to which they were randomly assigned. The concrete condition
(N ⫽ 35) and the abstract condition (N ⫽ 35) were the same as in
Study 1. The control condition (N ⫽ 49) was simply instructed to
“Describe the event in the space below.”
6
Compliment questions. Participants responded to four items
about how positively the compliment made them feel (happy,
secure, valuable, and accepted) on scales ranging from 1 (not at
all)to7(extremely). Two items assessed how abstractly they
perceived the compliment (“How meaningful was this event to
you?” “How significant was this event to your relationship?”) on
the same 7-point scales. Two items asked about participants’
attributions for the compliment, that is, how deserving of it they
felt (1 ⫽ my partner just wanted to be nice to me,7⫽ I was truly
deserving) and how sincere they thought their partner was (1 ⫽ not
at all,7⫽ extremely). Another two items assessed the frequency
with which participants believed they received compliments
(“How often does your partner say things like this?” 1 ⫽ never,
7 ⫽ very often) and the frequency with which they expected to
receive compliments (“How likely is it that your partner would say
something like this again in the future?” 1 ⫽ extremely unlikely,
5
In Study 1, one participation credit was awarded for completing Part 1,
and participants were given the option of completing Part 2 for a chance to
win a cash prize in a draw. Because of the low rate of return of Part 2 in
Study 1 (62% of the sample), in Studies 2 and 3 we made the completion
of both parts necessary to receive one credit.
6
Students were invited to participate in the study via e-mail. The
unequal sample sizes across conditions reflect the differing response rates
of students who had been randomly assigned to each condition (via the
unique participant ID number included in their e-mail). They were not
aware of which condition they would be in when they decided whether to
participate.
237
REFRAMING COMPLIMENTS FROM ROMANTIC PARTNERS
7 ⫽ extremely likely). Cronbach’s alpha for these four indices of
compliment positivity were as follows: .92 for positive feelings,
.64 for abstraction, .57 for attributions, and .75 for frequency.
Memory accessibility. This three-item scale (␣⫽.85) asked
participants to rate how easily and quickly the compliment they
wrote about came to mind when they were first asked to think of
a specific example of a compliment (1 ⫽ not at all,7⫽ extremely)
and how detailed was their memory for it (1 ⫽ very vague,7⫽
very detailed).
State SE. The same state SE measure was used as in Study 1
(␣⫽.93).
Relationship quality. Participants were instructed to consider
“how you feel about your relationship right now.” They responded
to 19 statements on a scale of 1 (not at all true)to7(completely
true), adapted from Murray et al. (2000, 2002). Responses to 12
statements were averaged to create a measure of felt security (␣⫽
.89; e.g., “I am confident that my partner will always want to look
beyond my faults and see the best in me,” “Though times may
change and the future is uncertain, I know my partner will always
be ready and willing to offer me strength and support”). Responses
to four statements were averaged to create a measure of partici-
pants’ satisfaction with the relationship (␣⫽.89; e.g., “I am
extremely happy with my current romantic relationship,” “I have a
very strong relationship with my partner”). Responses to three
statements tapping commitment to the relationship were also av-
eraged (␣⫽.86; e.g., “I am very committed to my relationship,”
“I see my relationship as a burden” [reverse scored]). Finally,
participants indicated how confident they were that they would
still be in a romantic relationship with their current partner at each
of five specified time periods (6 months, 1 year, 2 years, 5 years,
a lifetime). They rated each item on a scale from 1 (extremely
uncertain)to7(extremely certain). The alpha for this five-item
future optimism scale was .94.
Part 2 Materials
All participants, regardless of the condition they had been as-
signed to in Part 1, received the same instructions and questions in
Part 2. They responded to questions about their relationship in
general before being reminded of the compliment to determine
whether the manipulation in Part 1 might cause lasting changes in
their perceptions of relationship quality.
Frequency of positive partner behavior. Participants were in-
structed to “think about what has happened in your relationship in
the last 2 weeks (since you completed the first part of this study).”
Using a 5-point scale (1 ⫽ not at all,5⫽ many times), they rated
the frequency of four positive partner behaviors (e.g., “told you
how much he/she cares about you,” “supported or encouraged
you”) and four negative behaviors (e.g., “criticized you,” “acted
inconsiderately toward you”). We reverse scored the negative
items and averaged all eight responses to form a measure of
positive partner behavior (␣⫽.81).
Relationship quality. Part 2 included seven of the felt security
statements from Part 1 (␣⫽.89) and two each from the commit-
ment and satisfaction scales (␣⫽.75 and ␣⫽.93, respectively).
Compliment questions. After being reminded of the compli-
ment by using their cue words and briefly describing the compli-
ment as they now recalled it, participants rated several items
according to how they now felt about the compliment. We used the
same four-item positive feelings index as in Part 1 (␣⫽.87). The
abstraction index included the meaning and significance items
from Part 1 and one additional item: “How broad was your
partner’s praise for you?” This was rated on a scale ranging from
1(about my behavior only)to7(about me as a person). The alpha
for this abstraction scale was .72.
Part 1 and Part 2 Coding
As in Study 1, trained coders who were unaware of condition
rated how abstract the narratives seemed to them. The coders’
ratings were correlated at .51 for Part 1 and .66 for Part 2 and were
averaged to create a measure of abstraction for each narrative. We
also assessed the percentage of verbs in each narrative that were
present tense and past tense by using the LIWC program (Penne-
baker et al., 2001).
Results
Part 1
We created two composite variables to simplify reporting of the
numerous dependent variables. Preliminary analyses showed sim-
ilar patterns of effects on each subscale described under compli-
ment questions and under relationship quality. An overall “com-
pliment positivity” composite (␣⫽.84) was created by averaging
standardized scores for each of the four indices of compliment
questions (positive feelings, abstraction, attributions, and fre-
quency). A “relationship valuing” composite (␣⫽.93) was cre-
ated by averaging the standardized scores for commitment, satis-
faction, and future optimism. We kept the felt security index of
relationship quality separate because it is a theoretically distinct
construct (Murray et al., 2006). We expected that feeling more
positively about the compliment would lead to increased felt
security, which would then increase relationship valuing for LSEs.
As in Study 1, all dependent variables were regressed on effect-
coded condition (abstract, concrete, or control), SE, and the Con-
dition ⫻ SE interaction. Temporal distance of the compliment
(M ⫽ 3.31 months) was included as a covariate. On the basis of the
findings of Study 1 and other research described earlier, we hy-
pothesized that LSEs would be more negative than HSEs on all
dependent variables in both the control and concrete conditions but
not in the abstract condition. Furthermore, we expected that LSEs
in the abstract condition would be significantly more positive than
LSEs in either the control condition or the concrete condition.
Unless indicated, there were no differences between conditions
among HSEs on any of the dependent variables examined in this
study.
The correlations between measured variables are presented in
Table 3, and the predicted values for the various analyses are
reported in Table 4. A graph is presented (Figure 1) only for felt
security, which we believe to be the most important finding in this
study. However, the pattern of means in Figure 1 is quite similar
to the pattern of means for the other dependent variables and thus
may be viewed as a prototypical finding to aid in understanding the
information in Table 4.
Coding of abstractness. As in Study 1, there was a main effect
of condition on coders’ ratings of abstraction, ⫽.56, t(115) ⫽
5.61, p ⬍ .01, and ⫽⫺.46, t(115) ⫽⫺4.59, p ⬍ .01, but this
238
MARIGOLD, HOLMES, AND ROSS
was qualified by a Condition ⫻ SE interaction, ⫽⫺.28,
t(112) ⫽⫺2.99, p ⬍ .01. LSEs wrote more abstract narratives in
the abstract condition than in either the control condition, ⫽.54,
t(112) ⫽ 5.11, p ⬍ .01, or the concrete condition, ⫽.78,
t(112) ⫽ 5.71, p ⬍ .01, and even more abstract narratives than
HSEs in the abstract condition, ⫽⫺.41, t(112) ⫽ 2.87, p ⬍ .01.
HSEs’ narratives were significantly less abstract in the concrete
condition than in either the control condition, ⫽⫺.27, t(112) ⫽
⫺2.14, p ⬍ .05, or the abstract condition, ⫽⫺.38, t(112) ⫽
2.95, p ⬍ .01, which did not differ from each other.
Compliment positivity. A significant main effect of condition,
⫽.31, t(115) ⫽ 2.86, p ⬍ .01, was qualified by a Condition ⫻
SE interaction, ⫽⫺.32, t(112) ⫽⫺3.22, p ⬍ .01, on compli-
ment positivity. In the control condition, when given no specific
instructions about how to think about and describe the compliment,
LSEs reported feeling significantly worse about the compliment
than did HSEs, ⫽.50, t(112) ⫽ 3.60, p ⬍ .01. LSEs in the
abstract condition were significantly more positive than were LSEs
in either the control condition, ⫽.64, t(112) ⫽ 4.82, p ⬍ .01, or
the concrete condition, ⫽.41, t(112) ⫽ 2.84, p ⬍ .01, which did
not differ from each other. In fact, in the abstract condition, LSEs
viewed the compliments as positively as did HSEs (⫽⫺.25, ns).
Thus, consistent with our hypotheses, LSEs were typically less
inclined than were HSEs to make much of their partners’ compli-
ments, but they viewed the compliments as positively as did HSEs
after describing their abstract meaning and significance.
Memory accessibility. A marginal main effect of SE, ⫽.17,
t(112) ⫽ 1.91, p ⬍ .06, was qualified by a significant Condition ⫻
SE interaction, ⫽⫺.27, t(112) ⫽⫺2.65, p ⬍ .01, on memory
accessibility. LSEs reported lower accessibility than did HSEs in
both the control condition, ⫽.32, t(112) ⫽ 2.22, p ⬍ .05, and
the concrete condition, ⫽.35, t(112) ⫽ 2.22, p ⬍ .05. LSEs
reported higher accessibility in the abstract condition then in either
the control condition, ⫽.45, t(112) ⫽ 3.26, p ⬍ .01, or the
concrete condition, ⫽.28, t(112) ⫽ 1.91, p ⬍ .06. In the abstract
condition, LSEs did not differ from HSEs in reported ease of
retrieving the memory (⫽⫺.17, ns). Note that all participants
were asked to come up with a specific compliment from their
partner prior to the manipulation. LSEs tended to recall that
process as more difficult than did HSEs unless they had been
Table 3
Correlations Among Measured Variables: Study 2
Variable 12345678910
Part 1
1. Trait self-esteem —
2. Compliment positivity .20
*
—
3. Memory accessibility .19
*
.43
**
—
4. State self-esteem .63
**
.43
**
.33
**
—
5. Felt security .28
**
.64
**
.42
**
.48
**
—
6. Relationship valuing .11 .61
**
.35
**
.40
**
.71
**
—
Part 2
7. Compliment positivity .07 .63
**
.42
**
.34
**
.43
**
.47
**
—
8. Frequency of positive partner behavior .25
**
.42
**
.48
**
.46
**
.51
**
.55
**
.52
**
—
9. Felt security .37
**
.50
**
.43
**
.59
**
.70
**
.56
**
.63
**
.72
**
—
10. Relationship valuing .11 .45
**
.42
**
.43
**
.48
**
.73
**
.57
**
.71
**
.69
**
—
*
p ⬍ .05.
**
p ⬍ .01.
Table 4
Predicted Values for Control, Concrete, and Abstract Conditions at Low and High Self-Esteem (SE): Study 2
Dependent variable
Low SE High SE
Control Concrete Abstract Control Concrete Abstract
Part 1
Compliment positivity ⫺.47
a
⫺.12
a
.49
b
.23
b
.08
ab
.15
b
Memory accessibility 4.78
a
5.25
a
6.08
b
5.64
b
6.15
b
5.64
b
State self-esteem 4.54
a
4.84
ab
5.20
b
6.14
c
6.30
c
6.10
c
Felt security 5.24
a
5.04
a
6.23
b
6.18
b
5.94
b
5.93
b
Relationship valuing ⫺.24
a
⫺.43
a
.53
b
.10
a
.03
a
.13
ab
Abstraction (coded) 1.86
ac
1.49
a
3.49
b
2.28
c
1.59
a
2.55
c
Past tense 67.27
a
65.66
a
37.44
b
51.87
a
61.40
a
58.90
a
Present tense 32.11
a
32.49
a
58.12
b
42.83
a
36.35
a
38.94
a
Part 2
Frequency of positive partner behavior 3.79
a
3.66
a
4.32
b
4.25
b
4.36
b
4.28
b
Felt security 5.49
a
5.16
a
6.34
b
6.49
b
6.42
b
6.46
b
Note. Simple effects test were conducted comparing differences between conditions at low self-esteem (LSE), differences between conditions at high
self-esteem (HSE), and LSEs with HSEs within each condition, for each dependent variable. In each row, predicted scores for the cells that were compared
that do not share subscripts differ at p ⱕ .05. Low and high SE were calculated at ⫾1 SD.
239
REFRAMING COMPLIMENTS FROM ROMANTIC PARTNERS
subsequently instructed to describe the meaning and significance
of the compliment.
State SE. As in Study 1, the significant main effect relation
between trait SE and state SE, ⫽.62, t(112) ⫽ 8.70, p ⬍ .01,
was qualified by a marginally significant Condition ⫻ SE inter-
action, ⫽⫺.16, t(112) ⫽⫺1.89, p ⫽ .06. It is not surprising that
LSEs reported lower state SE than did HSEs in all conditions: for
the control condition, ⫽.75, t(112) ⫽ 6.33, p ⬍ .01; for the
concrete condition, ⫽.69, t(112) ⫽ 5.43, p ⬍ .01; for the
abstract condition, ⫽.43, t(112) ⫽ 3.39, p ⬍ .01. Most pertinent
to our hypotheses, however, LSEs reported significantly higher
state SE in the abstract condition than in the control condition, ⫽
.28, t(112) ⫽ 2.51, p ⬍ .05. The concrete condition fell in between
and was not significantly different from the other two conditions
( ps ⬎ .20, ns). Thus, thinking about a specific compliment from
their partner in an abstract manner did have implications for how
valuable LSEs felt more generally.
Felt security. Could LSEs take this abstract compliment fur-
ther? In addition to making them feel better about the compliment
and about themselves, would reframing compliments abstractly
make LSEs feel more secure in their relationships? Indeed, the
main effects of condition, ⫽.27, t(115) ⫽ 2.42, p ⬍ .05, and of
SE, ⫽.27, t(112) ⫽ 3.17, p ⬍ .01, were qualified by a
Condition ⫻ SE interaction, ⫽⫺.34, t(112) ⫽⫺3.48, p ⬍ .01
(Figure 1). As usual, LSEs reported feeling less secure about their
relationship than did HSEs in both the control condition, ⫽.48,
t(112) ⫽ 3.48, p ⬍ .01, and the concrete condition, ⫽.47,
t(112) ⫽ 3.14, p ⬍ .01. In the abstract condition, however, they
felt just as secure as did HSEs (⫽⫺.15, ns), and more secure
than LSEs in either of the other two conditions: for control com-
parison, ⫽.46, t(112) ⫽ 3.52, p ⬍ .01; for concrete comparison,
⫽.56, t(112) ⫽ 3.87, p ⬍ .01. Recall that participants in the
control condition completed the relationship ratings prior to de-
scribing a compliment. So, these results demonstrate that for LSEs,
simply thinking about positive feedback from their romantic part-
ner (in a concrete way) is not sufficient to raise their felt security
from baseline. They need to be encouraged to think of the feedback
in a more abstract, meaningful way.
Relationship valuing. Feeling insecure in their romantic rela-
tionship typically leads LSEs to devalue the relationship (Murray
et al., 2000). After boosting their felt security, then, would we see
a corresponding increase in how positively they perceived their
relationship? As with felt security, on the relationship valuing
composite there was a main effect of condition, ⫽.29, t(115) ⫽
2.58, p ⫽ .01, qualified by a Condition ⫻ SE interaction, ⫽
⫺.25, t(112) ⫽⫺2.39, p ⬍ .05. LSEs were more positive about
their relationship in the abstract condition than in either the control
condition, ⫽.40, t(112) ⫽ 2.87, p ⬍ .01, or the concrete
condition, ⫽.51, t(112) ⫽ 3.31, p ⬍ .01. No other simple effects
reached significance. Thinking about a partner’s compliment ab-
stractly led LSEs to not only feel more secure, but also to value the
relationship more highly.
Mediation. We conducted three mediational analyses pertinent
to our theoretical model to investigate how the Condition ⫻ SE
interaction affected the dependent variables (following Sobel,
1982). Consistent with expectations, we found that the interaction
effect (a) on state SE was mediated by compliment positivity (z ⫽
⫺2.58, p ⬍ .01), (b) on felt security was also mediated by
compliment positivity (z ⫽⫺2.95, p ⬍ .01), and (c) on relation-
ship valuing was mediated by felt security (z ⫽⫺3.29, p ⬍ .01).
7
These analyses indicate that for LSEs, describing positive feed-
back from their romantic partner in an abstract manner led them to
perceive the feedback more positively, which raised their state SE
and made them feel more secure in the relationship. This increased
sense of security, in turn, allowed them to be more generous in
their conclusions about their relationship and evaluate it more
positively.
Verb tense. We were particularly interested in determining
whether in the control condition, LSEs would spontaneously de-
scribe past compliments from their romantic partners as “more
past”—that is, using more past tense and less present tense—than
would HSEs. As expected, analyses revealed main effects of
condition on both past tense verbs, ⫽⫺.26, t(114) ⫽⫺2.32,
p ⬍ .05, and present tense verbs, ⫽.25, t(115) ⫽ 2.26, p ⬍ .05,
which were both qualified by Condition ⫻ SE interactions: for
past, ⫽.31, t(111) ⫽ 3.05, p ⬍ .01; for present, ⫽⫺.28,
t(111) ⫽⫺2.65, p ⬍ .01. When LSEs were given no specific
instructions on how to describe a compliment from their romantic
partner (control condition), they used marginally more past tense
verbs than did HSEs, ⫽⫺.28, t(111) ⫽⫺1.90, p ⫽ .06. They
also tended to use fewer present tense verbs, though not signifi-
cantly so ( p ⫽ .18). LSEs used fewer past tense verbs and more
present tense verbs in the abstract condition than in the control
condition (past: ⫽⫺.50, t[111] ⫽⫺3.59, p ⬍ .01; present, ⫽
.46, t[111] ⫽ 3.22, p ⬍ .01) or the concrete condition (past: ⫽
⫺.48, t[111] ⫽⫺3.15, p ⬍ .01; present: ⫽.45, t[111] ⫽ 2.94,
p ⬍ .01). In fact, even compared with HSEs in the abstract
condition, LSEs used fewer past tense verbs, ⫽.39, t(111) ⫽
2.54, p ⬍ .05, and more present tense verbs, ⫽⫺.37, t(111) ⫽
⫺2.34, p ⬍ .05. Thus, with no specific instructions for describing
positive feedback from their partners, LSEs tended to dismiss the
feedback as being more in the past than did HSEs. The compli-
7
Baron and Kenny’s (1986) four requirements for mediation were also
met for each mediation analysis in both Study 2 and Study 3: There was a
significant relationship between the independent variable (the Condition ⫻
Self-Esteem interaction term) and the dependent variable, between the
interaction term and the mediator, and between the mediator and the
dependent variable (controlling for the interaction term). The effect of the
interaction term on the dependent variable dropped to nonsignificance
when the mediator was included in the model.
5.0
5.5
6.0
6.5
Low High
Self-Esteem
Felt Security
Abstract
Control
Concrete
Figure 1. Felt security as a function of condition and self-esteem (⫾1
SD): Study 2
240
MARIGOLD, HOLMES, AND ROSS
ments LSEs nominated were not actually any farther in the past,
however—the correlation between SE and temporal distance of the
compliment was ⫺.01. When instructed to describe the more
abstract meaning and significance of the compliments, LSEs wrote
about them more presently—more so even than did HSEs.
Part 2
The Part 2 sample included 33 participants from the abstract
condition, 32 from the concrete condition, and 44 from the control
condition. As in Part 1, we created composite measures for com-
pliment positivity (positive feelings and abstraction; ␣⫽.87) and
relationship valuing (satisfaction and commitment; ␣⫽.92). Both
the temporal distance of the compliment and days between Part 1
and Part 2 (M ⫽ 16.40 days) were controlled for in the regression
analyses.
Felt security. By placing the general relationship questions at
the beginning of the Part 2 questionnaire, we tested whether the
increase in felt security for LSEs in the abstract condition would
remain over time when the specific compliment that caused the
increase was not immediately salient. Indeed, the pattern of results
was identical to that for felt security in Part 1. There were main
effects of condition, ⫽.28, t(104) ⫽ 2.39, p ⬍ .05, and SE, ⫽
.41, t(101) ⫽ 4.73, p ⬍ .01, which were qualified by a Condi-
tion ⫻ SE interaction, ⫽⫺.30, t(101) ⫽⫺2.98, p ⬍ .01. LSEs
who had been in the abstract condition in Part 1 were still feeling
as secure as were HSEs (⫽.05, ns) and more secure than LSEs
who had been in the control condition, ⫽.41, t(101) ⫽ 3.11, p ⬍
.01, or the concrete condition, ⫽.56, t(101) ⫽ 3.82, p ⬍ .01. The
simple effect of SE was significant in both of these conditions:
control condition, ⫽.52, t(101) ⫽ 3.70, p ⬍ .01; concrete
condition, ⫽.65, t(101) ⫽ 4.15, p ⬍ .01. There were no
significant effects on the relationship valuing composite in Part 2.
Frequency of positive partner behavior. If LSEs came away
from the first part of this study feeling more highly valued by their
romantic partners, would they then allow themselves to be more
attentive to other positive behaviors and less attentive to negative
behaviors from their partners? Indeed this is what is suggested by
our findings on the measure of frequency of positive partner
behavior. There were main effects of condition, ⫽.23, t(104) ⫽
1.95, p ⫽ .05, and SE, ⫽.29, t(101) ⫽ 3.19, p ⬍ .01, qualified
by a Condition ⫻ SE interaction, ⫽⫺.26, t(101) ⫽⫺2.46, p ⬍
.05. LSEs reported lower frequencies of positive behavior from
their partners than did HSEs when they had been in either the
control condition, ⫽.36, t(101) ⫽ 2.42, p ⬍ .05, or the concrete
condition, ⫽.53, t(101) ⫽ 3.21, p ⬍ .01, in Part 1. LSEs
reported just as high frequencies as did HSEs when they had been
in the abstract condition (⫽⫺.02, ns) and significantly higher
frequencies than LSEs who had been in either of the other two
conditions: control condition, ⫽.38, t(101) ⫽ 2.65, p ⬍ .01;
concrete condition, ⫽.46, t(101) ⫽ 2.91, p ⬍ .01.
Compliment positivity. There was only a main effect of con-
dition on the compliment positivity composite in Part 2, ⫽.37,
t(101) ⫽ 3.27, p ⬍ .05. Participants who had described a compli-
ment abstractly in Part 1 still felt more positively about it (M ⫽
.36) than those who had described it concretely, M ⫽⫺.14, ⫽
.29, t(101) ⫽ 2.61, p ⫽ .01, or those who were in the control
condition, M ⫽⫺.18, ⫽.32, t(101) ⫽ 3.12, p ⬍ .01.
Mediation. The Condition ⫻ SE interaction on positive part-
ner behavior was mediated by felt security (z ⫽⫺2.82, p ⬍ .01).
For LSEs, then, the increase in felt security accrued from describ-
ing compliments abstractly led them to perceive greater frequency
of other types of positive behaviors (and lower frequency of
negative behaviors) in the 2–3 weeks since that abstract descrip-
tion. Feeling more secure about their partners’ regard for them,
LSEs were more willing to generalize signs of acceptance and
minimize signs of rejection.
Coding. There were no significant effects on coders’ ratings of
how abstract were the narratives or on the use of present and past
tense verbs. One reason the finding on verb tense in Part 2 of Study
1 may not have replicated here is that in Study 1, writing about the
compliment was the first thing participants did in Part 2. In Study
2, we asked them to answer a number of questions about their
relationship prior to being reminded of the compliment, which may
have washed out any further effects of the manipulation.
Discussion
Study 2 replicated and extended the findings of Study 1 in
several ways. The main findings of Study 1 were that among LSEs,
abstract compliment framing increased happiness and state SE as
compared with concrete compliment framing. In Study 2, an added
control condition showed that LSEs were typically less enthusias-
tic about compliments from their romantic partners, reported more
difficulty in remembering the compliments, and described them by
using more past tense verbs than did HSEs. When given instruc-
tions to describe the concrete details of the compliment, LSEs did
not budge from their typical stance. When instructed to describe
the feedback more abstractly, in terms of its meaning and signif-
icance, LSEs reported increased positive feelings, reported greater
ease in recalling the feedback, and used present tense more often.
In the abstract condition, LSEs did not differ significantly from the
usually more buoyant HSEs on these measures.
It is important that remembering one specific compliment more
abstractly had far-reaching consequences for LSEs. It increased
their state SE and sense of relationship security, which encouraged
them to offer more positive evaluations of the relationship in
general. It is impressive that some of these changes persisted over
2 weeks. LSEs who had been in the abstract condition in Part 1
continued to report increased felt security relative to other LSEs in
Part 2. They also recalled more frequent positive and less frequent
negative behaviors from their romantic partners during the inter-
vening time period. The short and simple abstraction exercise
appeared to make LSEs feel significantly better about their rela-
tionships 2 weeks later.
One remaining question to be addressed by future research is
why LSEs reported more positive partner behavior since the first
session of the study. We can think of three possible explanations:
LSEs construed their partners’ behavior more positively only when
they were asked to look back on it, they actually perceived more
positive behavior throughout the 2 weeks, or they in fact elicited
more positive behavior from their partner during that time period.
If they finished the first session of the study feeling happier about
their relationship, they may have gone on to behave in warmer,
kinder ways toward their partner, who in turn may have treated
them better. Obtaining daily diaries and partner reports during this
2-week span could help disentangle these intriguing possibilities.
241
REFRAMING COMPLIMENTS FROM ROMANTIC PARTNERS
Study 3
Given how difficult it has been for past researchers to make
LSEs feel more optimistic about anything, why was our abstraction
manipulation so effective? We propose that the key is the subtlety
of the manipulation. We think that it worked because it managed
to assumptively imply that the compliment must have been mean-
ingful and significant, which avoided activating LSEs’ doubts
about their value to their partners. If LSEs had been explicitly
provided with an opportunity to question the broader meaning and
significance of the feedback, their self-evaluative worries would
likely have undermined their enjoyment of the compliment. We
tested this proposal in Study 3. Our hypothesis is that LSEs would
feel more positively about a compliment from their partner, about
themselves, and about their relationship when they were instructed
to describe the compliment abstractly but not when they were
instructed to reflect on the question of whether it should be
considered abstract. Further, we expected the LSEs would rate
their partners’ positive feedback and the relationship in general
more negatively than would HSEs in the control and question
conditions, but they would be just as positive as HSEs in the
abstract condition. In accordance with the research reviewed in this
article and with our earlier findings, we did not expect HSEs’
ratings to differ between conditions.
Method
Participants and Procedure
Ninety-one undergraduate students in romantic relationships
participated in a two-part study of “Relationship Perceptions” in
exchange for credit for their introductory psychology course. One
participant was excluded for not completing the materials. The
remaining sample of 90 consisted of 39 men and 51 women. Mean
age was 20 years, and mean relationship length was 21 months.
The majority of participants (71) indicated that they were in an
“exclusive dating” relationship. Seven individuals reported their
relationship status as “casual dating,” 6 were “living together,” 2
were “engaged,” and 4 were “married.”
Eighty-five participants completed Part 2. Two participants
were excluded from Part 2 analyses, 1 for failing to complete the
materials properly and 1 for having broken up with her partner
since completing Part 1. Thus, the Part 2 sample included 83
participants (35 men and 48 women) or 92% of the original
sample. The two samples did not differ on SE or on any of the
dependent variables. The study was conducted online, and the
procedure was identical to Study 2.
Materials
All materials, for Parts 1 and 2, were identical to Study 1, with
two exceptions. Memory accessibility was not assessed in this
study, and we replaced the concrete condition with a question
condition.
As in Studies 1 and 2, all participants were first instructed to
think of a time when their partner told them something they liked
about them. They were randomly assigned to one of three condi-
tions. The control condition (N ⫽ 29) and the abstract condition
(N ⫽ 27) were the same as Study 2. The question condition (N ⫽
34) differed very subtly from the abstract condition. Whereas the
abstract condition instructed participants to “Explain why your
partner admired you. Describe what it meant to you and its
significance for your relationship,” the question condition in-
structed participants to “Explain whether you think what your
partner said indicated that he/she admired you. Consider whether
it was meaningful to you and significant for your relationship.”
Thus, the same terms were used (admired, meaningful, and sig-
nificant), but they were posed as questions rather than implied as
assumptions by the instructions.
Results
Part 1
As in Studies 1 and 2, all dependent variables were regressed on
effect-coded condition (abstract, question, or control), SE, and the
Condition ⫻ SE interaction. Temporal distance of the compliment
(M ⫽ 4.69 months) was included as a covariate. As with Study 2,
there were no condition differences for HSEs unless indicated.
The correlations between measured variables are presented in
Table 5, and the predicted values for the various analyses are
reported in Table 6. As in Study 2, only the graph for felt security
is presented (Figure 2). The pattern of means in Figure 2 is so
similar to the pattern of means for the other dependent variables in
Table 5
Correlations Among Measured Variables: Study 3
Variable 123456789
Part 1
1. Trait self-esteem —
2. Compliment positivity .22
*
—
3. State self-esteem .48
**
.61
**
—
4. Felt security .24
*
.70
**
.63
**
—
5. Relationship valuing .17 .64
**
.60
**
.84
**
—
Part 2
6. Compliment positivity .12 .75
**
.38
**
.59
**
.56
**
—
7. Frequency of positive partner behavior .04 .48
**
.32
**
.57
**
.56
**
.62
**
—
8. Felt security .27
*
.64
**
.57
**
.82
**
.76
**
.64
**
.70
**
—
9. Relationship valuing .19 .59
**
.54
**
.75
**
.80
**
.64
**
.67
**
.86
**
—
*
p ⬍ .05.
**
p ⬍ .01.
242
MARIGOLD, HOLMES, AND ROSS
this study that it may be viewed as a prototypical finding to aid in
understanding the information in Table 6.
Coding of abstractness. The correlation between coders’ rat-
ings of the abstractness of the compliment narratives was .57.
There was a main effect of condition on abstraction, ⫽.36,
t(86) ⫽ 3.58, p ⬍ .01, and ⫽.36, t(86) ⫽ 3.64, p ⬍ .01,
respectively, but there was no interaction ( p ⬎ .30). Participants in
the abstract condition wrote more abstract narratives (M ⫽ 3.39)
than did those in the control condition, M ⫽ 1.79, ⫽.62, t(86) ⫽
6.07, p ⬍ .01, but not more than did participants in the question
condition (M ⫽ 3.47). Participants in the question condition also
wrote more abstract narratives than did participants in the control
condition, ⫽.64, t(86) ⫽ 6.47, p ⬍ .01.
Compliment positivity. We expected to replicate the finding
from Study 2 that LSEs would be more enthusiastic about past
positive feedback from their partner when they were instructed to
describe it in an abstract manner as compared with when they were
given no specific instructions. We did not expect the question
condition—in which abstract meaning and significance were sug-
gested but not assumed—to provide the same boost. A marginal
main effect of SE, ⫽.24, t(83) ⫽ 1.89, p ⬍ .07, was qualified
by a Condition ⫻ SE interaction on the compliment positivity
composite, ⫽⫺.25, t(83) ⫽⫺2.22, p ⬍ .05. As expected, LSEs
were less positive about the compliment than were HSEs in the
question condition, ⫽.38, t(83) ⫽ 2.63, p ⫽ .01, and somewhat
less in the control condition, ⫽.48, t(83) ⫽ 1.58, p ⫽ .12, but
not in the abstract condition (⫽⫺.13, ns). As well, LSEs tended
to report more positive thoughts and feelings about the compliment
in the abstract condition than in the control condition or the
question condition, although the simple effects were not significant
( ps ⬎ .17). Thus, asking LSEs to consider whether a compliment
from their partner was meaningful and significant did not have the
same benefit as instructing them to describe the meaning and
significance.
State SE. Consistent with Studies 1 and 2, there were main
effects of condition, ⫽⫺.31, t(86) ⫽⫺2.56, p ⬍ .05, and SE,
⫽.51, t(83) ⫽ 4.51, p ⬍ .01, that were qualified by a Condi-
tion ⫻ SE interaction, ⫽⫺.23, t(83) ⫽⫺2.33, p ⬍ .05, on state
SE. LSEs reported lower state SE than did HSEs in the control
condition, ⫽.78, t(83) ⫽ 2.93, p ⬍ .01, and the question
condition, ⫽.58, t(83) ⫽ 4.58, p ⬍ .01, but not in the abstract
condition (⫽.16, ns). LSEs who were in the abstract condition
reported higher state SE than did LSEs who were in the control
condition, ⫽.46, t(83) ⫽ 2.24, p ⬍ .05, and in the question
condition, ⫽.40, t(83) ⫽ 3.02, p ⬍ .01.
Felt security. As with Study 2, we expected that LSEs who
were instructed to reframe their partners’ compliments abstractly
would be able to internalize this praise and feel more valued by
their partner generally. We did not think that LSEs who were
encouraged to consider whether their partner’s praise might indi-
cate that they admired them would have a corresponding increase
in felt security. A main effect of SE, ⫽.26, t(83) ⫽ 2.04, p ⬍
.05, was qualified by a Condition ⫻ SE interaction, ⫽.25,
t(83) ⫽ 2.28, p ⬍ .05 (Figure 2). LSEs in the abstract condition felt
significantly more secure in their relationship than did LSEs in
either the control condition (who made their ratings prior to
recalling a compliment), ⫽.49, t(83) ⫽ 2.11, p ⬍ .05, or the
question condition, ⫽.33, t(83) ⫽ 2.18, p ⬍ .05. In fact, LSEs
felt just as secure as did HSEs in the abstract condition (⫽⫺.12,
ns). The simple effect of SE was significant in the question
condition, ⫽.46, t(83) ⫽ 3.27, p ⬍ .01. So, LSEs who were
instructed to describe their partners’ compliments abstractly felt
more secure about their partners’ regard than they typically did
(control condition), but LSEs who were asked to evaluate whether
Table 6
Predicted Values for Control, Question, and Abstract Conditions at Low and High Self-Esteem (SE): Study 3
Dependent variable
Low SE High SE
Control Question Abstract Control Question Abstract
Part 1
Compliment positivity ⫺0.62
a
⫺0.24
a
0.34
a
⫺0.02
ab
0.56
b
⫺0.12
ab
State self-esteem 5.05
a
5.16
a
5.87
b
6.31
b
6.10
b
6.13
b
Felt security 4.96
a
5.33
a
6.03
b
5.82
ab
6.25
b
5.81
ab
Relationship valuing ⫺0.51
a
⫺0.20
a
0.34
a
⫺0.11
a
0.54
b
0.06
ab
Part 2
Felt security 5.06
a
5.64
a
6.26
b
6.36
ab
6.64
b
5.98
b
Relationship valuing ⫺0.31
ab
⫺0.40
a
0.43
b
⫺0.07
b
0.48
b
0.05
b
Frequency of positive partner behavior (positive only) 3.14
a
4.01
b
4.21
b
4.10
ab
4.37
ab
3.87
ab
Note. Simple effects test were conducted comparing differences between low self-esteem (LSE) conditions, differences between high self-esteem (HSE)
conditions, and participants with low self-esteem with participants with high self-esteem within each condition for each dependent variable. In each row,
predicted scores for the cells that were compared that do not share subscripts differ at p ⱕ .05. Low and high SE were calculated at ⫾1 SD.
4.8
5.3
5.8
6.3
Low High
Self-Esteem
Felt Security
Abstract
Control
Question
Figure 2. Felt security as a function of condition and self-esteem (⫾1
SD): Study 3
243
REFRAMING COMPLIMENTS FROM ROMANTIC PARTNERS
their partners’ compliments had abstract meaning and significance
did not experience a boost to felt security.
Relationship valuing. The Condition ⫻ SE interaction on the
composite measure of relationship valuing was significant, ⫽
⫺.21, t(83) ⫽⫺1.96, p ⫽ .05. LSEs in the abstract condition
valued their relationship marginally more than did LSEs in the
control, ⫽.44, t(83) ⫽ 1.91, p ⬍ .06, and question conditions,
⫽.29, t(83) ⫽ 1.92, p ⬍ .06. Unexpectedly, HSEs were
significantly more positive about their relationship in the question
condition than in the control condition, ⫽.37, t(83) ⫽ 2.07, p ⬍
.05. The simple effect of SE was significant only in the question
condition, ⫽.43, t(83) ⫽ 3.05, p ⬍ .01. Thus, LSEs in the
abstract condition valued their relationships more than did LSEs in
the other two conditions. The finding for HSEs we suspect to be a
compensatory reaction to a perceived threat; we save further
explanation of this effect for the discussion.
Mediation. We conducted the same mediation analyses that
we did for Study 2. For LSEs, describing positive feedback from
their romantic partner in an abstract manner led them to perceive
the feedback more positively, which raised their state SE (z ⫽
⫺2.11, p ⬍ .05) and made them feel more secure in the relation-
ship (z ⫽⫺2.15, p ⬍ .05). This increased sense of security, in
turn, allowed them to make more generous conclusions about the
relationship by evaluating it more positively (z ⫽⫺2.25, p ⬍ .05).
Verb tense. Unlike Study 2, there were only main effects of
condition on verb tenses. The conditions differed in their propor-
tion of both past tense verbs, ⫽⫺.30, t(86) ⫽⫺2.58, p ⬍ .05,
and present tense verbs, ⫽.30, t(86) ⫽ 2.58, p ⬍ .05. Specif-
ically, participants in the control condition used more past tense
verbs (M ⫽ 62.1%) than did those in the abstract condition, M ⫽
44.1%, ⫽.45, t(86) ⫽ 3.74, p ⬍ .01, or the question condition,
M ⫽ 43.0%, ⫽.36, t(86) ⫽ 3.28, p ⬍ .01. As well, participants
in the control condition used fewer present tense verbs (M ⫽
33.3%) than did those in the abstract, M ⫽ 55.3%, ⫽⫺.46,
t(86) ⫽⫺3.93, p ⬍ .01, and question conditions, M ⫽ 54.9%, ⫽
⫺.40, t(86) ⫽⫺3.70, p ⬍ .01. Although participants in the
question condition generally reported less positive outcomes than
did participants in the abstract condition, their narratives indicated
that they were at least considering the compliments in as present
terms.
Part 2
The Part 2 sample included 26 participants from the abstract
condition, 24 from the control condition and 33 from the question
condition. The time interval between Parts 1 and 2 ranged from 12
to 24 days (M ⫽ 15.93).
Felt security. A main effect of SE, ⫽.34, t(75) ⫽ 2.35, p ⬍
.05, and a Condition ⫻ SE interaction, ⫽⫺.32, t(75) ⫽⫺2.72,
p ⬍ .01, yielded a similar pattern of results as in Part 1. LSEs who
had been in the abstract condition in Part 1 continued to report
greater felt security than did LSEs who had been in the control,
⫽.56, t(75) ⫽ 2.07, p ⬍ .05, and question conditions, ⫽.29,
t(75) ⫽ 1.94, p ⫽ .05. In the abstract condition LSEs did not differ
significantly from HSEs (⫽⫺.15, ns), whereas LSEs were
marginally less secure than HSEs in the control condition, ⫽.66,
t(75) ⫽ 1.80, p ⬍ .08, and significantly less in the question
condition, ⫽.51, t(75) ⫽ 3.58, p ⬍ .01, which we would
typically expect. So, 2 weeks after writing about the abstract
meaning and significance of a compliment from their romantic
partner, LSEs continued to feel more positively regarded by their
partner in general, as indexed by their felt security, than did LSEs
who were made to question the meaning and significance of the
compliment or given no instructions about what to write.
Relationship valuing. The Condition ⫻ SE interaction on re-
lationship valuing was significant, ⫽.27, t(75) ⫽ 2.01, p ⬍ .05.
LSEs who had been in the abstract condition in Part 1 continued to
value their relationship significantly more than LSEs who had
been in the question condition, ⫽.41, t(75) ⫽ 2.69, p ⬍ .01. The
simple effect of SE was significant only in the question condition,
⫽.47, t(75) ⫽ 3.29, p ⬍ .01. Thus, similar to the results for felt
security, LSEs who had been in the abstract condition in Part 1
continued to value their relationship more than did other LSEs.
Frequency of positive partner behavior. The analysis of the
eight-item measure, which included four positive items and four
(reverse scored) negative items, was not significant. Examining
positive and negative items separately, however, we found a sig-
nificant Condition ⫻ SE interaction. Whereas in Study 2, the
Condition ⫻ SE interaction was significant on both the positive
and negative items (making the entire eight-item measure signif-
icant), in Study 3 the interaction was significant only on the
positive items, ⫽⫺.26, t(75) ⫽⫺2.11, p ⬍ .05. Unexpectedly,
on this measure, LSEs who had been in the question condition
were just as high as were LSEs who had been in the abstract
condition, both of whom perceived greater frequency of their
partners’ positive behaviors than did LSEs who had been in the
control condition: abstract to control comparisons, ⫽.59,
t(75) ⫽ 2.06, p ⬍ .05; question to control comparison, ⫽.51,
t(75) ⫽ 1.76, p ⬍ .09. So, in addition to feeling more secure and
positive about their relationships, LSEs who had been in the
abstract condition also reported greater frequency of positive be-
haviors from their partners since completing Part 1 of the study
than did LSEs who had been in the control condition.
Compliment positivity. In Study 2, there was a main effect of
condition on the compliment positivity composite in Part 2, such
that everyone who had been in the abstract condition reported
more positive thoughts and feelings about the compliment than did
participants who had been in the concrete or control conditions. In
this study, no effects attained significance on this measure.
Mediation. We tested whether the Condition ⫻ SE interaction
effect on frequency of positive partner behavior was mediated by
felt security as in Study 2. We also tested whether the interaction
on relationship valuing (which was not found in Part 2 of Study 2)
was mediated by felt security. The results indicate that LSEs’
increase in felt security in the abstract condition indeed mediated
their increase in perceived frequency of positive partner behaviors
(z ⫽⫺2.58, p ⬍ .01) as well as their increase in relationship
valuing (z ⫽⫺2.77, p ⬍ .01).
Coding. There were no significant effects on abstraction or
verb tenses for the Part 2 narratives.
Discussion
In the control condition, LSEs were less enthusiastic about
specific instances of positive feedback from their romantic part-
ners than were HSEs, as usual. They also tended to report lower
state SE, less relationship security, and less relationship valuing.
When LSEs were led to question whether the feedback might be a
244
MARIGOLD, HOLMES, AND ROSS
meaningful and significant indication that their partner admired
them in general, their responses on these measures did not differ
from those of LSEs in the control condition. LSEs benefited from
recounting a compliment only in the abstract condition, where they
were instructed to describe the meaning and significance of the
compliment. Thus, the findings of Study 3 replicated and extended
Study 2 in an important way, by showing that the cognitive
reframing of the positive feedback must be subtly implied in order
to avoid activating LSEs’ self-evaluative concerns.
One unexpected finding was that HSEs reported valuing their
relationships more highly in the question condition than in the
control condition during the initial session. This effect may have
been a compensatory reaction for HSEs, who tend to respond to
potential relationship threats by affirming the relationship (Murray
et al., 1998, 2002). We do not believe that questioning the meaning
and significance of the compliment was a considerable threat to
HSEs. However, when the questioning raised some doubts, HSEs
probably compensated by recruiting additional positive thoughts
and memories in support of their partners’ admiration (Dodgson &
Wood, 1998; Smith & Petty, 1995). Thus, they ended up even
more enthusiastic than usual about the relationship.
General Discussion
Maintaining a secure and satisfying romantic relationship is
particularly challenging for LSEs; they have more doubts about
their partners’ love than do HSEs (Murray et al., 2000), but they
are less likely to benefit from their partners’ expressions of posi-
tive regard (Collins et al., 2006; Murray et al., 1998). The present
findings suggest a reason for optimism, however. In three studies,
we showed that LSEs could be induced to take their partners’ kind
words to heart.
When LSEs described a past compliment from their partner in
an abstract manner—what it meant to them and how it was
significant for their relationship—they reported increased positive
feelings and thoughts related to the specific compliment. These
positive feelings, in turn, increased LSEs’ state SE, security in
their partners’ acceptance, and consequently, evaluations of their
relationships. These increases for LSEs were observed relative to
no specific instructions for describing the compliment (Studies 2
and 3), concrete instructions (describing the specific details about
the time the compliment was given; Studies 1 and 2), and the
abstract instructions posed as a question (considering whether the
compliment was meaningful and significant; Study 3). LSEs also
showed increased use of present tense verbs and decreased use of
past tense verbs when describing the meaning and significance of
the compliment. This finding suggests that LSEs came to view the
compliments as evidence of their partners’ continuing admiration.
Importantly, LSEs’ cognitive reframing of the compliment had
lasting effects. Two to three weeks after the manipulation, LSEs
who had been in the abstract condition wrote narratives about what
they remembered of the compliment that used more present tense
and fewer past tense verbs (Study 1). They also continued to feel
more secure and positive about their relationship in general and
even perceived their partner as behaving more positively toward
them since the first part of the study (Studies 2 and 3). Being
instructed to write about the event in a particular way may have
altered how participants represented it in memory or reconstructed
it at the time of recall (McGregor & Holmes, 1999; Pennebaker &
Francis, 1996). Furthermore, the abstract manipulation may have
given LSEs a new way of thinking about other instances of
positive feedback from their partners, and they kept these more
abstract framings in mind as they considered the quality of their
relationship in general. Further research into these processes is
warranted.
It is important to note that in the control condition, LSEs were
less positive about the compliments than were HSEs. Indeed, LSEs
seem to adopt a rather myopic view when it comes to appraising
positive relationship events. They focus on that one particular
incident and fail to consider its implications for the relationship
more broadly (cf. L. Campbell et al., 2005). This finding supports
our contention that LSEs are relatively unwilling to take the risk of
embracing positive feedback from their partners lest it turn out
that, as suspected, they were not really valued so much after all.
One alternative explanation is that LSEs feel uncomfortable ac-
cepting compliments because they do not believe they are accurate,
and prefer to receive information that verifies their negative self-
views (Swann et al., 2003). However, several studies have shown
that although LSEs are more likely to believe negative feedback
(Swann et al., 2003), they would much rather receive positive
feedback (Murray et al., 2000; Swann, De La Ronde, & Hixon,
1994, for dating but not married couples). Furthermore, the present
research showed quite clearly that under the right circumstances,
LSEs can truly savor positive feedback.
Another alternative explanation for LSEs’ natural inclination to
be relatively unenthusiastic about their partners’ compliments is
that LSEs are unable to recall compliments that are as positive as
HSEs. To address this issue, we had two coders that were blind to
participants’ SE levels rate how positive the compliments were
(focusing on the nature of the compliments and not how positively
they were described, which would be affected by the manipula-
tion). Across the three studies, the correlation between positivity
and SE was r(266) ⫽ .004, ns. Thus, it does not appear that LSEs
recall less positive feedback from their partners than do HSEs.
Rather, LSEs are more hesitant to take positive feedback to heart.
Our studies, however, demonstrate that LSEs can be encouraged
to see their value to their partners, even to believe that they are
loved and accepted as much as HSEs perceive themselves to be.
This is, in fact, the truth: LSEs are regarded just as positively by
their partners as are HSEs (Murray et al., 2000, 2001), at least until
their unwarranted insecurities become problematic for their part-
ners. In all studies, HSEs were relatively unaffected by the ma-
nipulation of the abstractness of the compliment. We suspect that
their confidence in their partners’ love allows them to spontane-
ously make the most of positive feedback.
One question that remains to be addressed is whether these
results would replicate with a married sample (at most, 4% of the
sample was married in these studies). Given that dependence
regulation processes have been demonstrated to operate similarly
in both dating and married couples (Murray et al., 2006), we are
fairly confident that we would observe a similar pattern of results
with a sample composed entirely of married participants. How-
ever, the effects might be weaker. People attend to and weight
more heavily the signs of their partners’ caring and commitment in
early stages of relationships (Fletcher, Fincham, Cramer, & Heron,
1987), so individuals in long-term relationships (whether married
or not) may be less affected by thoughts of a simple compliment.
As well, married couples may be more motivated to have negative
245
REFRAMING COMPLIMENTS FROM ROMANTIC PARTNERS
self-views verified by their partners (Swann et al., 1994) and
therefore might be resistant to the manipulation.
Possible Mechanisms of the Effect
More pressing questions involve the precise mechanism under-
lying the experimental effect. Giving positive feedback to LSEs is
a risky business. Several experiments have shown that it can
backfire and actually make them feel worse (Logel et al., 2006;
Peach & MacDonald, 2004; Wood, Heimpel, Newby-Clark, &
Ross, 2005). Why, then, did we succeed in making things better for
LSEs? Although we do not have enough evidence at this time to
make definitive statements, we have some promising ideas for why
past attempts have failed, whereas our manipulation was success-
ful.
First, in our studies the positive feedback was self-generated;
participants came up with an example of feedback they had re-
ceived and then those who were assigned to the abstract condition
were encouraged to make more meaning of it. They were not
directly told by the researcher that they had scored high on some
positive attribute or that their partner thought very highly of them,
relative to other people. LSEs are less certain than HSEs that upon
comparing themselves to others they will emerge as superior
(Baumeister et al., 1989; Taylor & Brown, 1988), so in some
situations LSEs self-protectively shy away from social comparison
information (see Baumeister et al., 1989, and Wood & Lockwood,
1999, for reviews). But even being presented with a comparison
that is clearly in their favor may be threatening for LSEs if it
prompts them to evaluate how they measure up to others more
generally, and especially to their partner, to whom they typically
feel inferior (Murray et al., 2005).
Second, we did not tell participants that they possessed a spe-
cific positive attribute or resembled an ideal self, for example, that
they were a considerate partner. That kind of positive feedback
would be relatively discrepant with LSEs’ self-conceptions, so
they would automatically try to judge its fit (Wood et al., 2006).
This process could yield thoughts and memories both in support of
and against a favorable conclusion (cf. Showers, 1992). After
being told they are a considerate partner, LSEs may think of times
they acted inconsiderately. Thus, LSEs would imagine themselves
disappointing their partners when they fell short of this ideal
(Logel et al., 2006; Murray et al., 1998). Another way to think of
this issue is that traditional positive feedback manipulations work
in a top-down manner by first giving participants a broad conclu-
sion about a desirable attribute. For LSEs, this would activate a
search for evidence relevant to this conclusion, and they may be
just as likely to find inconsistent as consistent evidence. In con-
trast, we took a bottom-up approach and asked participants to first
recall a specific compliment and then encouraged them to make a
broader conclusion about the compliment’s implications. General-
izing from their own evidence may have prevented LSEs from
activating a search for disconfirming examples. Further research is
needed, however, to determine whether the bottom-up (vs. top-
down) characterization of our manipulation is as crucial as we
suspect.
A third reason why our abstract manipulation worked so well
may be that it presented a safe enhancement opportunity for LSEs.
LSEs are more interested in protecting the self against loss than are
HSEs (Baumeister et al., 1989), taking only those opportunities for
self-enhancement that seem sure to affirm the self (Rudich &
Vallacher, 1999; Wood, Giordano-Beech, Taylor, Michela, &
Gaus, 1994). In our abstraction manipulation, we presented the
broad meaning and significance of the positive feedback as a
foregone conclusion. The assumption inherent in our instructions
was that the compliment was unquestionably an important expres-
sion of positive regard. In Study 3, when we posed meaning and
significance as a question, there was no benefit to LSEs. Thus,
only when there were no doubts about its importance did LSEs feel
safe to embrace their partners’ positive feedback.
Furthermore, instead of attempting to give participants a global
sense of acceptance, our manipulation made one particular quality
that participants had demonstrated in the past appear to have more
global implications. The suggestion that participants’ partners
highly valued something specific about them, not that they highly
valued the participant overall, should be a safer and more accept-
able conclusion for LSEs. Fortunately, they were able to take this
to the next level and make their own conclusions about how much
their partners valued them more generally, as demonstrated by
increased state SE (all studies) and increased felt security (Studies
2 and 3). In fact, the conclusions LSEs drew were impressively
broad—some of the items in the felt security measure included
“My partner loves and accepts me unconditionally” and “I am
confident my partner will always want to stay in our relationship.”
Practical Implications
Clinical researchers have long noted that interventions in marital
therapy must go beyond increasing rates of partners’ positive
behaviors. Individuals should also be trained to attend to these
behaviors and make relationship-enhancing attributions for them.
Otherwise, LSEs and other individuals dissatisfied with their re-
lationships may frame their partners’ positive behavior in a way
that undermines its potentially reinforcing impact (Holtzworth-
Munroe & Jacobson, 1985; Robinson & Price, 1980). On the basis
of the present studies, we similarly suggest that it would not be
sufficient to encourage partners of LSEs to increase their fre-
quency of giving compliments because LSEs tend not to take
compliments to heart. Furthermore, interventions that require
LSEs’ partners to make more effort to reassure LSEs might be
frustrating and tiring for the partners (Van Orden & Joiner, 2006).
Rather, LSEs should be encouraged to abstractly frame and gen-
eralize from their partners’ compliments. After practice in the use
of this reframing technique on memories of past compliments, we
suspect that LSEs could learn to embrace their partners’ current
compliments as well.
Concluding Comments
The present research supports and extends the risk regulation
model proposed by Murray et al. (2006). The central tenet of this
model is that confidence in a partner’s positive regard allows
people to put self-protection motives aside and take the risk of
thinking and behaving in ways that promote the value of the
partner and the relationship. The current studies are the first to
show that experimentally increasing felt security (through abstract
compliment framing) causes LSEs to value their relationships
more highly. Further, these findings help clarify that it is concerns
about acceptance, and not a fixed aspect of LSEs’ personalities,
246
MARIGOLD, HOLMES, AND ROSS
that causes LSEs to be critical toward their partners. Recent
research in attachment theory has similarly shown that priming
attachment security increases compassion and altruistic behavior,
regardless of dispositional attachment style (Mikulincer, Shaver,
Gillath, & Nitzberg, 2005).
In future research we plan to examine whether the abstract
compliment manipulation will function as a buffer against rela-
tionship threats for LSEs. Typically, LSEs self-protectively de-
crease closeness and devalue their relationships when they feel
rejected (Murray et al., 1998, 2002), actions that ultimately under-
mine their relationships (Murray, Griffen, et al., 2003). We believe
that the boost to felt security engendered by an abstract compli-
ment description will prevent LSEs from behaving in this
relationship-destructive manner.
References
Aiken, L. S., & West, S. G. (1996). Multiple regression: Testing and
interpreting interactions. Thousand Oaks, CA: Sage.
Baldwin, M. W., & Sinclair, L. (1996). Self-esteem and “if . . . then”
contingencies of interpersonal acceptance. Journal of Personality and
Social Psychology, 71, 1130–1141.
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable
distinction in social psychological research: Conceptual, strategic, and
statistical considerations. Journal of Personality and Social Psychology,
51, 1173–1182.
Bartholomew, K., & Horowitz, L. M. (1991). Attachment styles among
young adults: A test of a four-category model. Journal of Personality
and Social Psychology, 61, 226 –244.
Baumeister, R. F., Campbell, J. D., Krueger, J. I., & Vohs, K. D. (2003).
Does high self-esteem cause better performance, interpersonal success,
happiness, or healthier lifestyles? Psychological Science in the Public
Interest, 4, 1– 44.
Baumeister, R. F., Tice, D. M., & Hutton, D. G. (1989). Self-presentational
motivations and personality differences in self-esteem. Journal of Per-
sonality, 57, 547–579.
Brennan, K. A., Clark, C. L., & Shaver, P. R. (1998). Self-report measure-
ment of adult attachment: An integrative overview. In J. A. Simpson &
W. S. Rholes (Eds.), Attachment Theory and Close Relationships (pp.
46 –76). New York: Guilford Press.
Brennan, K. A., & Morris, K. A. (1997). Attachment styles, self-esteem,
and patterns of seeking feedback from romantic partners. Personality
and Social Psychology Bulletin, 23, 23–31.
Brown, J. D., & Dutton, K. A. (1995). The thrill of victory, the complexity
of defeat: Self-esteem and people’s emotional reactions to success and
failure. Journal of Personality and Social Psychology, 68, 712–722.
Cameron, J. J., Anthony, D. B., Gaetz, R., & Balchen, S. (2006). The
impact of self-esteem on liking inference and cue detection. Manuscript
in preparation.
Campbell, J. D., Trapnell, P. D., Heine, S. J., Katz, I. M., Lavallee, L. F.,
& Lehman, D. R. (1996). Self-concept clarity: Measurement, personality
correlates, and cultural boundaries. Journal of Personality and Social
Psychology, 70, 141–156.
Campbell, L., Simpson, J. A., Boldry, J., & Kashy, D. A. (2005). Percep-
tions of conflict and support in romantic relationships: The role of
attachment anxiety. Journal of Personality and Social Psychology, 88,
510 –531.
Camper, P. M., Jacobson, N. S., Holtzworth-Munroe, A., & Schmaling,
K. B. (1988). Causal attributions for interactional behaviors in married
couples. Cognitive Therapy and Research, 12, 195–209.
Collins, N. L., & Feeney, B. C. (2004). Working models of attachment
shape perceptions of social support: Evidence from experimental and
observational studies. Journal of Personality and Social Psychology, 87,
363–383.
Collins, N. L., Ford, M. B., Guichard, A. C., & Allard, L. M. (2006).
Working models of attachment and attribution processes in intimate
relationships. Personality and Social Psychology Bulletin, 32, 201–219.
Collins, N. L., & Read, S. J. (1990). Adult attachment, working models,
and relationship quality in dating couples. Journal of Personality and
Social Psychology, 58, 644– 663.
Dodgson, P. G., & Wood, J. V. (1998). Self-esteem and the cognitive
accessibility of strengths and weaknesses after failure. Journal of Per-
sonality and Social Psychology, 75, 178 –197.
Fincham, F. D., & Bradbury, T. N. (1993). Marital satisfaction, depression,
and attributions: A longitudinal analysis. Journal of Personality and
Social Psychology, 64, 442–452.
Fletcher, G. J. O., Fincham, F. D., Cramer, L., & Heron, N. (1987). The
role of attributions in the development of dating relationships. Journal of
Personality and Social Psychology, 53, 481– 489.
Heatherton, T. F., & Polivy, J. (1991). Development and validation of a
scale for measuring state self-esteem. Journal of Personality and Social
Psychology, 60, 895–910.
Heine, S. J., Lehman, D. R., Markus, H. R., & Kitayama, S. (1999). Is there
a universal need for positive self-regard? Psychological Review, 106,
766 –794.
Holtzworth-Munroe, A., & Jacobson, N. S. (1985). Causal attributions of
married couples: When do they search for causes? What do they con-
clude when they do? Journal of Personality and Social Psychology, 48,
1398 –1412.
Kernis, M. H., Brockner, J., & Frankel, B. S. (1989). Self-esteem and
reactions to failure: The mediating role of overgeneralization. Journal of
Personality and Social Psychology, 57, 707–714.
Leary, M. R., Tambor, E. S., Terdal, S. K., & Downs, D. L. (1995).
Self-esteem as an interpersonal monitor: The sociometer hypothesis.
Journal of Personality and Social Psychology, 68, 518–530.
Logel, C., Spencer, S. J., Wood, J. V., Holmes, J. G., & Zanna, M. P.
(2006). Guarded and responsive low self-esteem: How implicit self-
esteem functions among low explicit self-esteem individuals. Manuscript
submitted for publication.
Lomore, C. D., Spencer, S. J., & Holmes, J. G. (in press). Shared values
affirmation as a strategy to enhance low self-esteem women’s relation-
ship perceptions. Self and Identity.
McFarland, C., & Ross, M. (1982). Impact of causal attributions on
affective reactions to success and failure. Journal of Personality and
Social Psychology, 43, 937–946.
McGregor, I., & Holmes, J. G. (1999). How storytelling shapes memory
and impressions of relationship events over time. Journal of Personality
and Social Psychology, 76, 403– 419.
Mikulincer, M., Shaver, P. R., Gillath, O., & Nitzberg, R. A. (2005).
Attachment, caregiving, and altruism: Boosting attachment security in-
creases compassion and helping. Journal of Personality and Social
Psychology, 89, 817– 839.
Murray, S. L., Bellavia, G. M., Rose, P., & Griffin, D. W. (2003). Once
hurt, twice hurtful: How perceived regard regulates daily marital inter-
actions. Journal of Personality and Social Psychology, 84, 126 –147.
Murray, S. L., Griffin, D. W., Rose, P., & Bellavia, G. (2003). Calibrating
the sociometer: The relational contingencies of self-esteem. Journal of
Personality and Social Psychology, 85, 63– 84.
Murray, S. L., Holmes, J. G., & Collins, N. L. (2006). Optimizing assur-
ance: The risk regulation system in relationships. Psychological Bulletin,
132, 641– 666.
Murray, S. L., Holmes, J. G., & Griffin, D. W. (1996). The benefits of
positive illusions: Idealization and the construction of satisfaction in
close relationships. Journal of Personality and Social Psychology, 70,
79 –98.
Murray, S. L., Holmes, J. G., & Griffin, D. W. (2000). Self-esteem and the
quest for felt security: How perceived regard regulates attachment pro-
cesses. Journal of Personality and Social Psychology, 78, 478 – 498.
247
REFRAMING COMPLIMENTS FROM ROMANTIC PARTNERS
Murray, S. L., Holmes, J. G., Griffin, D. W., Bellavia, G., & Rose, P.
(2001). The mismeasure of love: How self-doubt contaminates relation-
ship beliefs. Personality and Social Psychology Bulletin, 27, 423–436.
Murray, S. L., Holmes, J. G., MacDonald, G., & Ellsworth, P. C. (1998).
Through the looking glass darkly? When self-doubts turn into relation-
ship insecurities. Journal of Personality and Social Psychology, 75,
1459 –1480.
Murray, S. L., Rose, P., Bellavia, G., Holmes, J. G., & Kusche, A. G.
(2002). When rejection stings: How self-esteem constrains relationship-
enhancement processes. Journal of Personality and Social Psychology,
83, 556 –573.
Murray, S. L., Rose, P., Holmes, J. G., Derrick, J., Podchaski, E. J.,
Bellavia, G., & Griffin, D. (2005). Putting the partner within reach: A
dyadic perspective on felt security in close relationships. Journal of
Personality and Social Psychology, 88, 327–347.
Peach, J., & MacDonald, T. K. (2004). When good news is bad: Adult
romantic attachment styles and responses to positive feedback about
one’s relationship. Unpublished senior honors thesis, Queen’s Univer-
sity, Kingston, Ontario, Canada.
Pennebaker, J. W., & Francis, M. E. (1996). Cognitive, emotional, and
language processes in disclosure. Cognition and Emotion, 10, 601–626.
Pennebaker, J. W., Francis, M. E., & Booth, R. J. (2001). Linguistic inquiry
and word count: LIWC 2001. Mahwah, NJ: Erlbaum.
Robinson, E. A., & Price, M. G. (1980). Pleasurable behavior in marital
interaction: An observational study. Journal of Consulting and Clinical
Psychology, 48, 117–118.
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton,
NJ: Princeton University Press.
Rudich, E. A., & Vallacher, R. R. (1999). To belong or to self-enhance?
Motivational bases for choosing interaction partners. Personality and
Social Psychology Bulletin, 25, 1387–1404.
Semin, G. R., & De Poot, C. J. (1997). The question-answer paradigm: You
might regret not noticing how a question is worded. Journal of Person-
ality and Social Psychology, 73, 472– 480.
Semin, G. R., & Fiedler, K. (1988). The cognitive functions of linguistic
categories in describing persons: Social cognition and language. Journal
of Personality and Social Psychology, 54, 558–568.
Showers, C. J. (1992). Compartmentalization of positive and negative
self-knowledge: Keeping bad apples out of the bunch. Journal of Per-
sonality and Social Psychology, 62, 1036 –1049.
Smith, S. M., & Petty, R. E. (1995). Personality moderators of mood
congruency effects on cognition: The role of self-esteem and negative
mood regulation. Journal of Personality and Social Psychology, 68,
1092–1107.
Sobel, M. E. (1982). Asymptotic intervals for indirect effects in structural
equations models. In S. Leinhart (Ed.), Sociological methodology (pp.
290 –312). San Francisco: Jossey-Bass.
Swann, W. B., Jr., Rentfrow, P. J., & Guinn, J. S. (2003). Self-verification:
The search for coherence. In M. R. Leary & J. P. Tangney (Eds.),
Handbook of self and identity (pp. 367–383). New York: Guilford Press.
Swann, W. B., Jr., De La Ronde, C., & Hixon, G. J. (1994). Authenticity
and positivity strivings in marriage and courtship. Journal of Personality
and Social Psychology, 66(5), 857– 869.
Taylor, S. E., & Brown, J. D. (1988). Illusion and well-being: A social
psychological perspective on mental health. Psychological Bulletin, 103,
193–210.
Van Orden, K. A., & Joiner, T. E., Jr. (2006). The inner and outer turmoil
of excessive reassurance seeking: From self-doubts to social rejection. In
K. D. Vohs & E. J. Finkel (Eds.), Self and relationships: Connecting
interpersonal and intrapersonal processes (pp. 104–129). New York:
Guilford Press.
Wood, J. V., Anthony, D. B., & Foddis, W. F. (2006). Should people with
low self-esteem strive for high self-esteem? In M. H. Kernis (Ed.),
Self-esteem issues and answers: A source book of current perspectives
(pp. 288 –296). New York: Psychology Press.
Wood, J. V., Giordano-Beech, M., Taylor, K., Michela, J. L., & Gaus, V.
(1994). Strategies of social comparison among people with low self-
esteem: Self-protection and self-enhancement. Journal of Personality
and Social Psychology, 67, 713–731.
Wood, J. V., Heimpel, S. A., Newby-Clark, I. R., & Ross, M. (2005).
Snatching defeat from the jaws of victory: Self-esteem differences in the
experience and anticipation of success. Journal of Personality and
Social Psychology, 89, 764–780.
Wood, J. V., & Lockwood, P. (1999). Social comparisons in dysphoric and
low self-esteem people. In R. M. Kowalski & M. R. Leary (Eds.), The
social psychology of emotional and behavioral problems: Interfaces of
social and clinical psychology. (pp. 97–135). Washington, DC: Ameri-
can Psychological Association.
Received May 19, 2006
Revision received July 21, 2006
Accepted July 28, 2006 䡲
248
MARIGOLD, HOLMES, AND ROSS
A preview of this full-text is provided by American Psychological Association.
Content available from Journal of Personality and Social Psychology
This content is subject to copyright. Terms and conditions apply.