Self and Identity, 8: 45 – 62, 2009

http://www.psypress.com/sai
ISSN: 1529-8868 print/1529-8876 online
DOI: 10.1080/15298860802079786

Developing and Testing a Self-affirmation

Manipulation

LUCY NAPPER
PETER R. HARRIS
TRACY EPTON
University of Sheffield, Sheffield, UK

We developed a new manipulation of self-affirmation that is simple to use, can be

employed when circumstances prevent pre-screening or using essay-based methods,
and has a highly equivalent control task. In the process, we addressed theoretical and
empirical questions highlighted in a recent review as requiring attention. In
comparison to five controls, the manipulation promoted positive changes on
measures designed to test whether people are self-affirmed. It produced a similar
pattern on these measures to an existing self-affirmation manipulation. The matched
control also performed well. There was no evidence of mediation by mood or state
self-esteem or moderation by trait self-esteem. We discuss the potential of the
method and associated dependent measures in future self-affirmation research.

Keywords: Self-affirmation; Self-esteem; Defensiveness; Values; Health-risk

information.

Most people prefer new information to support rather than challenge their
important beliefs. This preference for congenial information has been shown for
beliefs of many kinds (Sherman & Cohen, 2006). It has both theoretical and applied
significance. In the domain of health, for instance, it poses a significant obstacle to
health behavior change when those at risk of disease reject important but
uncongenial information about their risk status and its health implications (e.g.,
Liberman & Chaiken, 1992).
Self-affirmation theory (Steele, 1988) suggests that such defensive responses arise
when information threatens the person’s ‘‘self-integrity’’, or sense of being
‘‘adaptively and morally adequate’’ (Steele, 1988, p. 262). Encountering such threats
arouses a need to restore self-integrity (Steele, 1988). This can be achieved in several
ways but, according to the theory, one way of achieving it is by self-affirming in

Received 19 April 2007; accepted 25 March 2008; first published online 24 May 2008.
Lucy Napper is now at the Center for Behavioral Research and Services, California State University,
Long Beach, CA, USA. The self-affirmation, matched control and journey control tasks, together with
instructions for their use and the response measures used in Study 2, are available from the corresponding
author.
Thanks to Anna Good for her help with collecting data and Stacey Lavda for her help with coding.
Correspondence should be addressed to: Peter Harris, Department of Psychology, University of
Sheffield, Western Bank, Sheffield S10 2TP, UK. E-mail: p.harris@sheffield.ac.uk

Ó 2008 Psychology Press, an imprint of the Taylor & Francis Group, an Informa business
46 L. Napper et al.

domains unrelated to the threat. Thus, for example, offering people the opportunity
to reflect upon how kind they are could make them less defensive and more open-
minded about the health risks of the caffeine they drink (Reed & Aspinwall, 1998).
This and other predictions from self-affirmation theory have been subject to
extensive tests. Self-affirmation has been shown to reduce: attitude change in
dissonance paradigms (Steele, 1988), the tendency to engage in downward social
comparisons when faced with a threat (Spencer, Fein, & Lomore, 2001), prejudiced
evaluations (Fein & Spencer, 1997), and biased processing of counter-attitudinal or
threatening health information (Cohen, Aronson, & Steele, 2000; Harris & Napper,
2005; Sherman, Nelson, & Steele, 2000). Such findings suggest that self-affirmation
theory has significant potential to contribute to our understanding of the
psychological processes underlying self-defense.
The dominant self-affirmation paradigm is experimental: participants are
randomized into self-affirmation or control conditions. This raises two issues: (a)
how to establish that a manipulation has successfully induced self-affirmation and
(b) what comprises an adequate control condition. Surprisingly little attention has
been paid to these issues (McQueen & Klein, 2006). They are addressed in this paper
in the course of reporting three studies designed to develop and test a new method of
self-affirming that has some practical advantages over existing methods and that has
the potential for what we call here greater ‘‘control equivalence.’’ In the process of
developing this method we also test two theoretical issues that emerged as requiring
more data from McQueen and Klein’s (2006) recent systematic review: whether the
effects of self-affirming are mediated by mood or state self-esteem and whether they
are moderated by trait self-esteem.
Given that the timing of the self-affirmation may affect the processes involved
(e.g., Brinol, Petty, Gallardo, & DeMarree, 2007) it is important to make clear from
the outset that the current studies belong to the tradition of self-affirmation research
in which the manipulation comes before rather than after the threat (e.g., Reed &
Aspinwall, 1998; Sherman et al., 2000, Study 2). In this approach participants are (1)
given the self-affirmation manipulation shortly before being exposed to (2) the
threatening information and then (3) any post-threat dependent measures. In Studies
1 and 2 we assess the effects of completing the self-affirmation and control measures
in the absence of threat (i.e., at stage 1 of the above chain); in Study 3, in a small
pilot test, we probe effects at stages 2 and 3.

Methods of Self-affirming
Researchers have adopted a variety of ways of inducing self-affirmation (see
McQueen & Klein, 2006). These include the completion of value scales (e.g., Steele &
Liu, 1983), writing about a central value (e.g., Sherman et al., 2000) and giving
positive personality feedback (e.g., Cohen et al., 2000). The variety of methods is
testament to the ‘‘flexibility’’ of self-affirmation (Sherman & Cohen, 2006).
The commonest manipulations focus on a specific personal strength or value, with
many studies using some version of the Allport–Vernon–Lindzey values scales
(AVL; Allport, Vernon, & Lindzey, 1960). Self-affirmation studies using these scales
typically ask participants first to rank or complete an inventory to indicate their
important values and then use this information either to pre-select participants (e.g.,
Steele & Liu, 1983) or assign to them a subscale corresponding to their selected value
(e.g., Tesser & Cornell, 1991). Although the AVL scales have been used successfully
in a number of experimental settings (e.g., Sherman et al., 2000; Spencer et al., 2001),
 Manipulating Self-affirmation 47

it is not always feasible to establish participants’ core values and assign each a
specific subscale (as, for instance, when running large numbers of participants
simultaneously). The language used on the items is also antiquated and can be
difficult for populations with low literacy or education levels (McQueen & Klein,
2006). Time constraints can limit the use of alternative methods, such as requiring
participants to write essays about a central value.
One alternative is to give all participants the opportunity to affirm the same value or
characteristic. For example, Reed and Aspinwall (1998) asked participants to recall
and describe occasions on which they were kind to others. In not requiring questions
to be tailored to individuals, this method has some practical advantages. However,
there are also clearly problems in assuming that one value is important to everyone.
There is a need for a method that does not involve pre-testing or assignment to
different scales but is not restricted to a single value or characteristic and that
preferably uses contemporary language. Such a method, especially one that proved
suitable for use in large-scale and more applied settings, would provide a useful
addition to the available methods of self-affirmation. This is one of the goals of the
studies reported here.

Control Tasks
Another neglected topic in self-affirmation research is the choice of control task. The
goal is, of course, to create a control that is as equivalent to the self-affirmation
manipulation as possible (‘‘control equivalence’’) but that lacks the affirmational
‘‘active ingredient’’. In practice, this can be hard to achieve.
The AVL scales have a clear advantage in this regard, as the control task—
completing a subscale about an unimportant value—appears to have high control
equivalence without being self-affirmational. Controls for other methods are less
adequate, often sacrificing equivalence to the goal of making the task non-
affirmational. For example, controls for essay techniques often vary on more than
one dimension from the experimental condition, such as when participants are asked
to write about how a value unimportant to them may be important to someone else
(e.g., Fein & Spencer, 1997). Some controls are completely different from the
experimental task, such as when participants have to indicate agreement with trivial
opinion statements (Reed & Aspinwall, 1998) or recall food and drink consumption
(Cohen et al., 2000); some studies have no control task at all (e.g., Klein, Blier, &
Janze, 2001). The use of such non-equivalent control conditions limits the ability to
attribute effects unequivocally to self-affirmation.
The search is on for a method of self-affirmation that combines universal
applicability with control equivalence, but does not require individual tailoring. The
following studies describe the development of a self-affirmation technique designed
to achieve this. It is designed to have the following properties: (a) the ability to make
salient central and positive aspects of the self-concept while being (b) easy and
practical to administer, (c) using values and strengths expressed in ways appropriate
and meaningful to modern samples, and (d) having available a highly equivalent
control condition.

The New Method

The Values in Action (VIA) Strengths scale (Peterson & Seligman, 2004) is a
contemporary alternative to the AVL scales. It consists of 250 items measuring an
48 L. Napper et al.

extensive range of values and strengths—expressed in contemporary language—

under six core headings: Wisdom and Knowledge, Courage, Humanity, Justice,
Temperance, and Transcendence. We used the VIA to develop a simple, self-report
questionnaire to act as a potential manipulation of self-affirmation. We designed a
control task to be as similar as possible to this manipulation. Instead of rating
themselves, participants were asked to rate a well-known celebrity on the same
attributes.1 Whereas the control task therefore lacks the focus on personal values
and strengths, in all other respects it is identical. We assess here the adequacy of this
control by comparing responses to those made by participants completing self-
affirmation tasks or other control tasks.

Establishing that People are Self-affirmed

A further problem hampering the development of control tasks stems from the very
flexibility of self-affirmation: it is possible to imagine how almost any task can be
self-affirming. This brings us to the criteria for testing when the self-affirmational
mindset has been successfully induced. By examining such criteria we can establish
whether a control task is typically used to self-affirm or not, regardless of its
apparent potential for self-affirming.2
Unfortunately, how to do this empirically is another relatively neglected topic in
self-affirmation research. McQueen and Klein found that only about half of the
studies using the AVL scales checked whether those in the experimental group
endorsed more items consistent with the value than those in the control group. Steele
and Liu (1983) tested whether participants made more positive ratings of their self-
concepts after focusing on their central value. Two studies (Cohen et al., 2000;
Sherman et al., 2000) found increases, albeit marginal (Cohen et al., 2000) on a single
self-report item. Researchers using essay-based methods have sometimes used
independent raters to rate essay content (e.g., Harris & Napper, 2005). Based on
these approaches and Steele’s (1988) theorizing, we therefore developed a number of
self-report items to test whether our new manipulation made positive, central, and
valued aspects of the self-concept more salient than did the matched control and
other control tasks. These items were tested in Study 1 and refined and extended in
Study 2.

Study 1
In Study 1 we tested the new manipulation and matched control alongside two
existing tasks: a self-affirmation manipulation that has been shown to reduce
defensiveness in cigarette smokers (Harris, Mayle, Mabbott, & Napper, 2007) and a
control task developed expressly to address concerns that participants may use any
self-reflective writing opportunity to self-affirm (Cohen et al., 2000). These tasks
provided rigorous benchmarks against which initially to test our new tasks.

Method
The self-affirmation manipulation. This comprised 32 items adapted from the
Values in Action (VIA) Strengths scale (see Table 1). The questionnaire was designed
to focus participants’ minds on, rather than to measure, their important self-values
(i.e., to function not as a values scale but as a mindset manipulation). Items were
chosen (by the first author in consultation with the second) to cover a wide range of
 Manipulating Self-affirmation 49

TABLE 1 A Sample of the Items Comprising the New Method

VIA value Example item
Wisdom and Knowledge ‘‘Being able to come up with new and different ideas
and ways of doing things is one of my strong
points.’’
‘‘I value my ability to think critically.’’
Courage ‘‘I must stand up for what I believe in, even in the
face of strong opposition.’’
‘‘I always admit when I am wrong.’’
Humanity ‘‘I am never too busy to help a friend.’’
‘‘I go out of my way to cheer up people who appear
down.’’
Justice ‘‘I treat all people equally, regardless of who they
might be.’’
‘‘I really enjoy being part of a group.’’
Temperance ‘‘I never seek vengeance.’’
‘‘I do not act as though I am a special person.’’
Transcendence ‘‘I experience deep emotions when I see beautiful
things.’’
‘‘Despite challenges, I always remain hopeful about
the future.’’
Note: The items were adapted from Peterson, C., & Seligman, M. E. P. (2004). Character
strengths and virtues: A handbook and classification. Oxford, UK: Oxford University Press.
Copyright Ó 2004 by Values in Action Institute. Reprinted with permission of Values in
Action Institute.

strengths and values. Participants were told that the task was designed to measure
their personal strengths and asked to ‘‘choose the response that most closely reflects
your thoughts’’ using the 5-point scale used on the VIA scale (very much like me/like
me/neutral/unlike me/very much unlike me).
In Study 1, this task was tested against an existing self-affirmation manipulation
and two control conditions. The alternative self-affirmation manipulation was a
directed positive thinking task, based on one employed by McGuire and McGuire
(1996), in which participants were asked to ‘‘write down as many of your desirable
characteristics that you can think of.’’ The task was presented as a personal recall
exercise concerning how easy people found it to recall their positive self-
characteristics. This task has been used previously and shown to reduce defensive
processing in an at-risk group (Harris et al., 2007).
The first control was the control task matched to the VIA-based self-affirmation
manipulation. Participants completed the same set of items, but about a celebrity.
They were told that the task was ‘‘designed to measure the way in which people make
judgments about the personal strengths of other people’’ and in their case to answer
the questions thinking about the qualities of David Beckham. Reponses were given
on the equivalent 5-point scale, with ‘‘him’’ replacing ‘‘me.’’ In all key respects the
instructions were identical in the VIA-based tasks. The second control was Cohen
et al.’s (2000) food control in which participants were asked to list everything they
had eaten in the previous 48 hours accurately and in as much detail as they could.
50 L. Napper et al.

Participants. These were 400 undergraduate and high-school students recruited

by opportunity sampling in class settings or on the university campus (MAGE ¼ 19.1
years). The majority of the sample was female (N ¼ 300).

Measures. Items (Table 2) were designed to assess the extent to which the tasks
had focused participants’ attention on positive, valued and important self-aspects:
e.g., ‘‘The questionnaire made me . . . think about positive aspects of myself,’’ ‘‘. . .
think about my values (the principles and standards by which I try to live my life)’’
(Strongly disagree/Disagree/Neutral/Agree/Strongly agree). We also included the
self-feeling item used by Cohen et al. (2000) and Sherman et al. (2000), ‘‘How do you
feel about yourself?’’ 0 (poorly) to 6 (extremely positive). A subgroup of participants
(N ¼ 212) next completed the adapted Mood Adjective Checklist (Raghunathan &
Trope, 2002). Two adjectives measured positive mood (happy/elated) and two
negative mood (sad/depressed) 0 (Definitely does not apply to my feeling at this
moment) to 3 (definitely does apply to my feeling at the moment). Another subgroup
(N ¼ 188) instead completed the 20-item Current Thought Scale (Heatherton &
Polivy, 1991) state self-esteem measure (Not at all/A little bit/Somewhat/Very much/
Extremely). The final item, completed by all participants, was Robins, Hendrin, and
Trzesniewski’s (2001) single-item measure of trait self-esteem (SISE), ‘‘I have high
self-esteem’’ (5-point scale, not very true of me to very true of me). This measure has
been designed as a practical alternative to the Rosenberg Self-Esteem Scale
(Rosenberg, 1965) and, given the time constraints operating in the studies reported
here, was chosen for its brevity. In addition to these measures, participants in the
matched control condition were asked, ‘‘What is your overall opinion of David

TABLE 2 Responses to the Principal Outcome Measures by Condition, Study 1

Matched Food
McGuire SA VIA SA control control
Measure (N ¼ 103) (N ¼ 100) (N ¼ 96) (N ¼ 101)
The questionnaire made me*:
Think about positive 2.92a (0.76) 2.71a (0.77) 1.67b (0.89) 1.97c (0.90)
aspects of self
Focus my attention on 2.56a (0.89) 2.61a (0.72) 1.40b (0.84) 1.81c (0.93)
who I am
Aware of things I value 2.72a (0.80) 2.42a (0.78) 1.67b (1.03) 1.64b (0.81)
about myself
Think about things 2.58a (0.83) 2.54a (0.75) 1.80b (1.04) 1.82b (1.31)
personally important
to me
Think about my values 2.07a (1.02) 2.04a (0.81) 1.26b (1.03) 1.27b (1.17)
Self-feeling 3.36 (1.02) 3.52 (0.82) 3.47 (1.02) 3.49 (1.03)
Positive affect 1.44ab (0.71) 1.58a (0.64) 1.50ab (0.65) 1.17b (0.68)
Negative affect 0.42 (0.64) 0.36 (0.47) 0.38 (0.65) 0.55 (0.73)
Total state self-esteem 68.63 (12.13) 68.34 (10.63) 66.24 (14.07) 71.04 (8.84)
Note: Sample sizes given in table are maximum Ns. Means sharing subscripts do not differ
significantly (alpha ¼ .05) on post hoc tests (Tukey’s HSD) following significant main effect.
*Items scored 0 to 6.
 Manipulating Self-affirmation 51

Beckham?’’ (0 ¼ Extremely negative, 3 ¼ Neutral, 6 ¼ Extremely positive). The mean

was mildly positive (M ¼ 3.5). Ratings were not significantly associated with any
outcome variable, although the correlation with whether participants had focused on
personally important things approached significance, r(84) ¼ .20, p ¼ .07.

Procedure. Participants took part in a psychology project on personal strengths

[personal recall]. They were randomly assigned to condition and—because of time
limits beyond the researchers’ control—completed either questions measuring mood
or state self-esteem. Time limits also meant that not all participants completed all
measures. This is reflected in variations in the degrees of freedom in some analyses.

Results
Effects of the self-affirmation manipulation. Analysis comprised one-way
ANOVAs, with Questionnaire Condition (VIA self-affirmed, McGuire self-affirmed,
matched control, food control) as the between-participants variable, followed by
Tukey HSD post hoc tests after significant main effects. The randomization check
showed that participants did not differ in trait self-esteem (VIA self-affirmation,
3.3 ¼ M ¼ 3.5, food control), F(3, 344) ¼ 1.12, ns. Adding gender to the analyses
reported below did not qualify any of the reported effects and revealed only one
main effect of gender (females reported significantly lower scores on the self-feeling
item).
Mean responses of the groups to the principal outcome measures are in Table 2.
There were significant main effects of condition on thinking more about positive self-
aspects, F(3, 396) ¼ 50.97, focusing attention on who I am, F(3, 394) ¼ 47.74,
awareness of things I value about myself, F(3, 385) ¼ 39.37, thinking about things
personally important to me, F(3, 385) ¼ 18.03, and thinking about my values, F(3,
306) ¼ 14.09 (all ps 5 .001). Post hoc analysis showed that the self-affirmation
conditions did not differ significantly from each other but had significantly higher
scores than either control condition on all these dependent variables (Table 2). The
two control conditions differed from each other on two: the food control made
participants think more about positive self-aspects and about who they were than
did the matched control. (It is worth noting that whereas for the self-affirmation
conditions the means for the measures fall on the agree side of the neutral midpoint,
2, for the control conditions they fall on the disagree side.)
There were no significant differences in ratings on the measures of self-feeling, F(3,
380) 5 1, negative affect, F(3, 198) ¼ 1.06, ns, or state self-esteem, F(3, 135) 5 1, but
the groups did differ in positive mood, F(3, 200) ¼ 3.45, p 5 .05: the food control
resulted in a less positive mood than the other tasks (Table 2).

Dispositional self-esteem as a moderator. Moderated regressions (Baron &

Kenny, 1986) were used to examine whether dispositional self-esteem interacted
with the self-affirmation manipulation. The mean of the five items measuring
positive, valued and important self-aspects was calculated (Cronbach’s alpha ¼ .86)
and analyzed using two-step hierarchical regression analyses, followed by tests of
simple slopes (Aiken & West, 1991). At step one, the main effects of condition
(control ¼ 0, self-affirmation ¼ 1) and self-esteem were entered, and the condition6
self-esteem interaction at step two.3 (The data from the two control conditions were
combined and both predictors were mean centered.) The regression equation was
significant at step 2, F(3, 241) ¼ 32.71, p 5 .001, R2 ¼ .29. Along with a main effect of
52 L. Napper et al.

condition, b ¼ .51, p 5 .001, trait self-esteem also predicted significantly higher

ratings on this measure, b ¼ .19, p 5 .001. However, the interaction term was not
significant, b ¼ .09, ns, nor did adding this term lead to a significant increase in
explained variance. Thus there was no evidence here that participants lower in self-
esteem were less likely to take the opportunity to self-affirm. In equivalent analyses
of mood and state self-esteem, condition predicted significantly more positive affect,
b ¼ .22, p 5 .01, and trait self-esteem predicted significantly more positive affect,
b ¼ .19, p 5 .05, less negative affect, b ¼ 7.28, p 5 .001, and more state self-esteem,
b ¼ .77, p 5 .001. Again, the interaction term was not significant in any analysis.4

Discussion
In contrast to two control conditions, one designed to closely match the new self-
affirmation task and one used previously in the literature, the new self-affirmation
manipulation focused the minds of participants more on important, positive and
valued self-aspects. At the same time, the pattern of responses it produced across the
range of dependent measures matched closely those of an established method of self-
affirming. Thus, the self-affirmation manipulation has passed its initial test. The
matched control task also appears to have functioned well, being comparable to the
food control on most dependent variables. Indeed, where it did differ, it arguably
performed better, being less likely to make participants aware of themselves or think
about positive self-aspects. In common with other self-affirmation manipulations,
neither manipulation affected negative affect or state self-esteem. Thus, there is no
support in these data for affect or state self-esteem as mediators of the effects of
condition on the principal dependent variables. Neither did trait self-esteem
moderate the effects of self-affirmation. Following these promising initial findings,
study two provides further tests of the new self-affirmation and matched control
conditions.

Study 2
In Study 2, we tested the new self-affirmation and matched control tasks against
three additional control tasks, including two from the literature: writing an essay
about a value unimportant to the self (e.g., Harris & Napper, 2005; Sherman et al.,
2000) and the personal opinion survey used by Reed and Aspinwall (1998). We also
developed a new control task (the ‘‘journey’’ control) and tested this for future use
with the current self-affirmation manipulation in circumstances where it is
inappropriate to use the matched control task.

Method
Participants. Participants were 246 high school and university students aged 16–
27 (M ¼ 18.7 years) recruited in classroom or laboratory settings. The majority of the
sample was female (N ¼ 179).

Self-affirmation and control tasks. The new self-affirmation and matched control
tasks were as in Study 1. Instructions for the essay-control condition were as in
Harris and Napper (2005): Participants were asked to choose their least important
value and write about why it might be important to another student. Those in the
opinion survey condition (Reed & Aspinwall, 1998) were asked their opinions on a
 Manipulating Self-affirmation 53

range of trivial opinion issues, such as, ‘‘I think that winter is the most satisfying
season of the year’’ (yes/no) and to elaborate when they answered yes. Participants in
the journey control condition were told that the task was designed to see how easy
people find it to recall familiar items and were asked to spend a few minutes writing
down the shops and buildings they pass on a journey they travel regularly. They were
asked to describe everything they could recall about the shops and buildings.
Participants were randomly assigned to condition. No participant reported problems
completing any of the above tasks.

Measures. Seven bipolar scales assessed self-appraisal (Cronbach’s alpha ¼ .74).

The items followed the stem, ‘‘The questionnaire made me think about . . .’’ and
were: negative aspects of myself/positive aspects of myself; things that are
not important to me/things that are important to me; things I don’t like about
myself/things I like about myself; things I am bad at/things I am good at; things I don’t
value about myself/things I value about myself; my failings/my successes; things I’d like
to change about myself/things I wouldn’t like to change about myself (scored 0 to 6,
mid point 3 ¼ not at all). Principal component analysis extracted only one
component (eigenvalue 4 1), which explained 42% of the scale variance. Two
unipolar scales assessed awareness of self and values (Cronbach’s alpha ¼ .74). These
items followed the stem, ‘‘The questionnaire made me aware of . . .’’ and were: who I
am and my values (the principles and standards by which I try to live my life)
(0 ¼ Strongly disagree/Disagree/Neutral/Agree/Strongly agree ¼ 4). To address con-
cerns about whether the matched control task induces a sense of inferiority, six
bipolar scales were added to assess inferiority relative to the other conditions
(Cronbach’s alpha ¼ .82). The items followed the stem, ‘‘Completing the task made
me feel . . .’’ and were: foolish/clever; inadequate/adequate; bad/good; unimportant/
important; inferior/superior; unattractive/attractive (scored 0 to 6, mid point 3 ¼ not
at all). Three items assessed knowledge about and similarity to the celebrity among
participants in the matched control condition: ‘‘How confident are you in the
accuracy of your judgments about David Beckham?’’; ‘‘How easy was it for you to
make judgments about David Beckham’s strengths?’’; and ‘‘How similar do you feel
to David Beckham?’’; 0 (not at all confident/easy/similar) to 6 (extremely confident/
easy/very similar; Cronbach’s alpha ¼ .76). Finally, participants in all conditions
completed the self-esteem and mood items from Study 1. A subset of respondents in
the control conditions also completed the Rosenberg Self-Esteem Scale (Rosenberg,
1965). The two self-esteem measures correlated significantly positively, r(100) ¼ .64,
p 5 .001.

Procedure. Participants completed the tasks either alone or in groups of 5–10.

They were told to work through the questions in order and at their own pace without
conferring.

Results
Analysis comprised one-way univariate ANOVAs, with Questionnaire Condition
(self-affirmed, matched, essay, journey, opinion) as the between-participants
variable, followed by Tukey HSD post hoc tests after significant main effects. The
randomization check showed that conditions did not differ in mean levels of trait
self-esteem, whether measured using the SISE (3.1 ¼ M ¼ 3.3), F(4, 212) 5 1, or
Rosenberg (1.9 ¼ M ¼ 2.1), F(2, 98) ¼ 1.17, ns, measures. Adding gender to the
54 L. Napper et al.

analyses reported below did not qualify any of the reported effects and revealed only
one main effect of gender (females reported significantly lower state self-esteem).
Mean responses of the groups to the principal outcome measures are in Table 3.
There were significant main effects of condition on all three dependent variables: self-
appraisal, F(4, 240) ¼ 3.14, p 5 .05, awareness of self and values, F(4, 225) ¼ 10.60,
p 5 .001, and inferiority, F(4, 230) ¼ 2.80, p 5 .05. Post hoc analyses revealed that
participants in the self-affirmation condition were significantly more positive than
those in each of the control conditions on the self-appraisal measure (see Table 3). In
contrast, none of the control conditions differed from each other on this measure.
The self-affirmation condition was also significantly more likely to induce awareness
of self and values than either the matched or journey controls, but not than the essay
or opinion survey controls. The matched and journey controls were, in turn,
significantly less likely to induce such awareness than the essay or opinion survey
controls. The self-affirmation condition was also significantly less likely than the
essay control to induce inferiority.
As in Study 1, there were no significant effects of condition on state self-esteem,
F(4, 182) 5 1, ns. There were, however, significant main effects of condition on both
positive, F(4, 186) ¼ 3.63, p 5 .01, and negative affect, F(4, 187) ¼ 7.70, p 5 .001.
Post hoc analyses revealed that in each case the main effect was generated principally
by differences between control conditions rather than between self-affirmation and
control conditions: completing the matched control led to more positive affect than
the other three controls, whereas completing the essay control led to more negative
affect than any other condition.
Higher ratings of knowledge about and similarity to David Beckham were
associated with higher self-appraisal, r(67) ¼ .33, p 5 .01, but not with awareness of
self and values, r(54) ¼ .05, ns, or inferiority, r(59) ¼ .04, ns.
Dispositional self-esteem as a moderator. As in Study 1, moderated regressions
tested whether the effects of self-affirmation on the outcome measures were
moderated by dispositional self-esteem. Data from all control conditions were
combined into mean scores and analyzed using two-step hierarchical regression
analyses. At step one, the main effects of condition (control ¼ 0, self-affirmation ¼ 1)
and self-esteem were entered, after mean centering, and the condition6self-esteem
interaction at step two. All regression equations were significant at step 2, F(3,
212) ¼ 5.25, p 5 .05, R2 ¼ .07 (self-appraisal), F(3, 199) ¼ 3.18, p 5 .05, R2 ¼ .05
(awareness of self and values), F(3, 203) ¼ 5.13, p 5 .01, R2 ¼ .07 (inferiority). In no
case was the interaction term significant. Condition significantly predicted greater
self-appraisal, b ¼ .22, p 5 .001, awareness of self and values, b ¼ .20, p 5 .01, and
less inferiority, b ¼ .18, p 5 .01. Trait self-esteem (SISE) predicted significantly less
inferiority, b ¼ .17, p 5 .05, but was not a significant independent predictor of self-
appraisal, b ¼ .08, ns, or awareness of self and values, b ¼ .03, ns. Thus, as in Study 1,
there was no evidence here that participants lower in self-esteem were less likely to
take the opportunity to self-affirm. In equivalent analyses condition did not predict
mood or state self-esteem, but trait self-esteem (SISE) predicted significantly more
positive affect, b ¼ .35, less negative affect, b ¼ 7.32, and more state self-esteem,
b ¼ .58 (all ps 5 .001).

Discussion
The self-affirmation manipulation led to more positive self-appraisal and greater
awareness of self and values, but did not affect ratings of inferiority, mood or state
TABLE 3 Responses to the Principal Outcome Measures by Condition, Study 2
Self affirmation Matched control Journey control Essay control Opinion control
Measure (N ¼ 47) (N ¼ 70) (N ¼ 40) (N ¼ 45) (N ¼ 44)
Self-appraisal* 3.56a (0.73) 3.17b (0.57) 3.13b (0.54) 3.12b (0.81) 3.14b (0.94)
Awareness of self/values* 2.71a (0.72) 2.25bc (0.84) 1.82b (1.04) 2.88a (0.70) 2.49ac (0.82)
Inferiority* 3.42a (0.61) 3.12ab (0.53) 3.15ab (0.54) 2.99b (0.67) 3.14ab (0.80)
Positive affect 1.38ab (0.87) 1.60a (0.87) 1.13b (0.47) 1.10b (0.53) 1.19b (0.54)
Negative affect 0.71a (0.75) 0.62a (0.84) 0.42a (0.54) 1.16b (0.87) 0.39a (0.55)
Total state self-esteem 70.12a (11.52) 69.18a (9.47) 68.38a (9.79) 67.92a (9.15) 69.61a (10.95)
Manipulating Self-affirmation

Note: Sample sizes given in table are maximum Ns. Means sharing subscripts do not differ significantly (alpha ¼ .05) on post hoc tests (Tukey’s HSD).
a
Scored 0 to 6. Higher scores indicate more positive appraisal, more awareness, affect and esteem, but less inferiority.
55
56 L. Napper et al.

self-esteem. The new self-affirmation manipulation has therefore passed its second
empirical test, again promoting significant positive changes on the critical measures
of important and valued self-aspects. The two new controls, the matched and
journey controls, also appear to have functioned well, being comparable to the essay
and opinion survey controls in self-appraisal, but less likely than these controls to
induce awareness of self and values. There was little evidence here that the matched
control task induced a sense of inferiority. There was, however, a tendency for those
with more knowledge about and similarity to David Beckham to have higher self-
appraisal scores, which suggests that a small subset of participants may have found
this control task self-affirming. Researchers should avoid control targets with whom
large numbers of participants will identify and include measures of identification
with the target when using this task. The new journey control condition appears to
provide a viable alternative when circumstances prevent the use of the matched
control.
Once again, there was no evidence here that trait self-esteem moderated the extent
to which participants self-affirmed.

Study 3
Study 3 was designed to provide a further test of the self-affirmation and matched
control. In this study, in a small sample pilot test, we examined the pattern of
thoughts high-risk participants generate when facing a threat to their self-integrity
after completing either the new self-affirmation or matched control tasks. Previous
research has shown that self-affirming can reduce the extent to which at-risk groups
are biased when facing uncongenial information, such as information about a threat
to their health (Harris & Napper, 2005; Sherman et al., 2000). Consequently, we
probed here to see whether we could detect shifts in the balance of thoughts
expressing greater or lesser defensiveness—such as thoughts implying less or more
message acceptance—when a high-risk group subsequently read a health message
after completing the tasks. As well as further attesting to the functionality of the new
self-affirmation method, such data will enhance our picture of the thought patterns
generated by such groups in response to threatening health-risk information.

Method
Young female students (N ¼ 35) in the top quartile of alcohol consumers on a pre-
test given at the beginning of the academic year were recruited to take part and
randomly assigned to complete either the self-affirmation or matched control tasks.5
No participant was aware that they had been selected because of their pre-test scores
The experimenter remained blind to condition throughout.
On arrival, participants were given the measure used by Harris and Napper (2005)
to assess alcohol consumption. Mean weekly alcohol consumption did not differ
between conditions (control 16.8 ¼ M ¼ 17.0 self-affirmed) and was above the UK
Government recommended maximum of 14 units. (A UK unit is 8 grams of alcohol,
or approximately half a pint of beer, a measure of spirits, or glass of wine.)
After completing the manipulation, participants were provided with a health
message (based on that used in Harris & Napper, 2005) outlining the risks of breast
cancer from alcohol and a sheet describing the thought-listing task. They were told
that the researchers were interested in their reactions to the content of the article and
instructed to write down as they read the article ‘‘anything and everything’’ that went
 Manipulating Self-affirmation 57

through their mind. After the thought-listing task participants completed a brief
questionnaire that included three sets of opposite-paired items, measured using 7-
point scales. Two items assessed risk perception, ‘‘How likely do you think YOU will
be to develop breast cancer from drinking alcohol compared with the average
student of your age and sex?’’ (much less likely than average/less likely than average/
slightly less likely than average/average/slightly more likely than average/more likely
than average/much more likely than average) (comparative risk) and later: ‘‘I do not
think that I will develop breast cancer as a result of my drinking’’ (strongly disagree/
strongly agree) (own risk). Two items assessed thinking about the information, ‘‘I
thought deeply about the information’’ and ‘‘I tried not to think about how the
article applied to me’’ (not at all/very much). Two items assessed perceived relevance,
‘‘The content of the article was . . . relevant to me’’, ‘‘. . . relevant to the average
University of Sheffield student of my age and sex’’ (strongly disagree/strongly agree).

Results
Coding categories were developed from the literature on defensive processing
(Blumberg, 2000; Croyle & Sande, 1988; Raghunathan & Trope, 2002) and used to
code participants’ thought listings. Each set of codings was designed to comprise a
related pair. The categories were: thoughts (a1) in support of the message, (a2)
counter-arguing the message; (b1) taking the message seriously, (b2) minimizing or
downplaying the message; (c1) accepting that the message was personally relevant,
(c2) downplaying or rejecting the personal relevance of the message. Thoughts coded
as irrelevant were not included in the analysis. Thoughts were also coded for valence.
Two independent judges, blind to condition, analyzed the thought listings. Their
ratings corresponded highly (r ¼ .72 to .94) and were therefore combined into a mean
rating per category for analysis (Table 4).6
Data for each pair were analyzed using one-way, repeated measures ANOVA
within condition, with the pair of related categories in each case comprising a
two-level repeated-measures variable. Both self-affirmed, F(1, 17) ¼ 9.39, p 5 .01,
and non-affirmed, F(1, 16) ¼ 12.36, p 5 .05, participants made fewer statements
supporting than counter-arguing the evidence. Participants in the self-affirmed
condition made significantly more statements accepting than denying personal
relevance, F(1, 17) ¼ 8.33, p 5 .01; non-affirmed participants did not, F(1, 16) 5 1.
There were no differences in either condition in the number of statements taking the
message seriously or minimizing the issue, F(1, 17) ¼ 1.11, ns (self-affirmed), F(1,
16) 5 1 (non-affirmed), or coded as being positive or negative, F 5 1 (both
conditions). The paired questionnaire items were analyzed in the same way.
Participants in the self-affirmed condition reported thinking deeply about the
information more than trying not to think about it, F(1, 17) ¼ 9.30, p 5 .01; the non-
affirmed participants did not, F(1, 16) ¼ 1.09, ns. Self-affirmed participants did not
see the article as being less relevant to them than to the average student, F(1, 17) 5 1,
whereas this difference was closer to significance in the non-affirmed condition,
F(1, 16) ¼ 3.13, p ¼ .096. Self-affirmed participants did not see themselves as less at
risk, F(1, 17) ¼ 1.24, ns; non-affirmed participants did, F(1, 16) ¼ 12.36, p 5 .01.
The responses of both groups were also tested for optimistic bias (perceiving self
risk as significantly less than average risk) on the comparative risk measure using a
one-sample t-test to test whether the sample mean differed significantly from the
mid-point of the scale (3). The self-affirmed group did not display optimistic bias,
t(17) ¼ 70.62, ns; the non-affirmed group did, t(16) ¼ 73.82, p 5 .01.
58 L. Napper et al.

TABLE 4 Thought Listing Codings and Questionnaire Responses, Study 3

SA NA
(N ¼ 18) (N ¼ 17) d
Thought listing
Supporting the evidence 0.53 (0.74) 0.59 (0.78) 0.08
Counter-arguing the evidence 1.83 (1.33) 2.09 (1.84) 0.16
Issue considered genuine and serious 2.67 (1.62) 2.53 (1.75) 0.08
Minimizing issues 1.94 (1.84) 2.09 (1.83) 0.08
Personal relevance 2.53 (2.00) 1.82 (2.05) 0.35
Denying/reducing personal relevance 1.00 (0.97) 1.47 (1.28) 0.42
Unrelated to the article 2.69 (2.95) 2.26 (2.25) 0.16
Negative statements 4.78 (2.50) 5.59 (3.81) 0.25
Positive statements 5.72 (2.82) 4.94 (3.22) 0.26
Questionnaire items
Thought deeply about the information 3.78 (1.40) 3.47 (1.28) 0.23
Tried not to think about the information 2.06 (1.47) 2.88 (1.69) 0.52
Content relevant to self 4.72 (1.07) 3.94 (1.56) 0.59
Content relevant to others 4.83 (1.04) 4.24 (1.44) 0.47
Comparative risk 2.83 (1.15) 1.94 (1.14) 0.78
Own risk 3.39 (1.20) 3.94 (1.64) 0.38
Note: SA ¼ self-affirmed; NA ¼ non-affirmed. Cohen’s d is included as an estimate of effect
size for the between-group main effect. Cohen (1988) suggested d ¼ 0.2 be considered small,
d ¼ 0.5 medium and d ¼ 0.8 large effects, respectively.

Discussion
The findings provide insights into the thought patterns of a group of high-risk people
reading an uncongenial health message. In both conditions participants generated
more thoughts counter-arguing than supporting the message. However, even in such
a high-risk group, the new self-affirmation task appeared to affect the balance of the
participants’ thoughts and responses, rendering them less biased: they made more
statements accepting than denying personal relevance, reported thinking deeply
about the information more than trying not to think about it, and were not
optimistically biased about the risk of developing breast cancer from alcohol.
Participants completing the matched control task displayed a pattern of thoughts
more indicative of defensiveness and, despite being in the top quartile of female
drinkers, saw themselves as being less at risk than the average female student of
developing breast cancer from alcohol. These initial findings support a future study
with a bigger sample (perhaps broader in terms of risk status) that tests for between-
group effects. Nevertheless, there is evidence even here of a detectable influence of
the new self-affirmation task on the pattern of thoughts generated by an at-risk
group in the face of threat, findings that lend further support to the functionality of
the new tasks.

General Discussion
Across three studies, the new tasks stood up well to testing. In Studies 1 and 2, the
new self-affirmation task promoted significant positive changes on dimensions
 Manipulating Self-affirmation 59

suggested in the literature to be indicative of self-affirming. It did this in contrast to

five different control conditions, including three used in previous studies. In Study 1,
its effects corresponded closely to those of a method of self-affirmation that has been
shown previously to reduce defensive processing. In Study 3, there were detectable
effects of completing the task on the thoughts expressed by a group of high-risk
participants, suggesting they were less defensive when reading a threatening health
message. The matched control task also stood up well to test. In Studies 1 and 2 it
promoted responses that were either equivalent to those of the other control tasks or
superior. In Study 3, participants who had completed the task appeared to remain
largely defensive about the health message.
In the process of developing and testing the new methods, we also explored
theoretical questions concerning mediation, by mood and state self-esteem, and
moderation, by trait self-esteem, finding no evidence here for either. Below we
discuss a number of issues arising from what we have done and consider what might
be investigated next.
Approaches to assessing the impact of self-affirming on defensiveness vary in
whether the self-affirmation comes before or after exposure to the threat, and thus
whether people are aware of the threat when self-affirming. The approach adopted
here reflects the procedures typically employed in studies that assess the impact of
self-affirming upon message processing, in which participants are given the self-
affirmation manipulation first (often in the guise of a separate experiment) and only
subsequently exposed to the threatening message (e.g., Harris & Napper, 2005; Reed
& Aspinwall, 1998). Now that we have established that it affects relevant dependent
measures, the new method needs to be tested for its effectiveness at the next stage of
this process, reducing defensive responding. The findings of Study 3 go some way
towards demonstrating this, but this is only a pilot study and it was not possible to
test the between-group interactions for statistical significance.7 In further research
(Klein, Harris, & Napper, 2008) we have demonstrated that the new self-affirmation
method also reduces defensiveness to threat. Among other things, participants who
completed this task and were then presented with threatening information about
their caffeine consumption oriented quicker to the threatening material, were more
accurate in their recall of items that had been in the message—suggesting that they
paid more attention to the message—and, one week later, recalled less risk-
disconfirming information, than their non-affirmed counterparts.
Nevertheless, further tests of the capacity of this particular method to reduce
defensiveness would be useful. At the same time, a next critical step is to examine
whether the attributes shown here to be raised by self-affirming mediate subsequent
reductions in defensive processing. Assessing this would help advance our under-
standing of the processes by which prior self-affirming facilitates greater open-
mindedness. That is, although we created these items to measure the impact of the
self-affirmation manipulations, if they do capture the essence of what it is to be self-
affirmed, then they potentially comprise its ‘‘active ingredients’’—the dimensions
raised or made salient by self-affirming that facilitate more open-minded and
balanced message appraisal.
Consequently, testing this—with other self-affirmation manipulations as well—is
important. It is not, however, without its complexities, as completing items designed
to assess whether people are self-affirmed may itself be self-affirming and thus erode
or even eliminate differences between experimental and control groups in defensive
responding. Whether this is also the case for these items should therefore be tested
before they are placed between a self-affirmation manipulation and a threat message.
60 L. Napper et al.

There is also the danger that making people aware that their values and other
important self-aspects have been made salient may eliminate the effectiveness of the
self-affirmation manipulation, as awareness has been argued to be a critical
moderator of the effectiveness of self-affirmation (Sherman & Cohen, 2006). Solving
these issues and exploring whether these attributes mediate the impact of self-
affirmation on message processing is an important task for future research.
The current findings add to evidence that self-affirmation manipulations typically
do not boost explicit measures of self-esteem or positive affect (Schmeichel &
Martens, 2005; Sherman & Cohen, 2006). To help us map what is distinctive about
the consequences of self-affirming, the measures developed here could be used next in
studies in which self-esteem is boosted or self-enhancement encouraged to establish
how the patterns of responding to the items following such manipulations compare
to responses after self-affirming.
In Studies 1 and 2 we demonstrated that those lower in trait self-esteem are as
ready as those higher in trait self-esteem to respond positively when required to
complete a self-affirmation manipulation, thus demonstrating no difference in
baseline ability to self-affirm. Research needs to establish when trait self-esteem
moderates the impact of self-affirming in response to threat (McQueen & Klein,
2006).
To accompany the new self-affirmation manipulation we developed a control task
that is very high in control equivalence. The data suggest that this task functions
well. However, there will be circumstances in which researchers feel unable to select a
suitable comparison target, such as when large numbers of participants identify with
the target. For those concerned that no target is sufficiently neutral, we also
developed and tested here a further control condition, the journey control that—
while it lacks the same level of control equivalence—emerged as one of the better
functioning control conditions from Study 2.
It is, of course, possible to use the new self-affirmation method with any control
task. It was designed to add to existing methods one with strengths for use in
circumstances such as large group settings or where literacy or other difficulties
present barriers to the use of other available methods. As such we believe the data
presented here support its use. No other method of self-affirming has been subject to
such stringent test. At the same time we have developed a set of measures that
researchers can use when developing other manipulations and controls and in future
investigations of the effects of self-affirming.

Notes
1. For the purpose of these studies this was the England soccer player David Beckham.
However, this questionnaire is designed for use with any target familiar to a sample
and about whom they would be happy to make such judgments (subject to the caveats
mentioned in the general discussion).
2. This is a critical distinction. The question is not whether the task looks as if it
could be used to self-affirm—given the flexibility of self-affirmation almost any task
has this potential—but whether participants typically self-affirm when they undertake
it.
3. These analyses preserve the continuity of the measures. In accordance with the
recommendations of Aiken and West (1991), the independent variables involved in all
moderator analyses were mean centered. Mean-centering involves subtracting the
mean value from the raw scores and reduces multicollinearity (see Aiken & West,
1991).
 Manipulating Self-affirmation 61

4. Equivalent analyses of the outcome measures using the McGuire task

produced very similar findings, except that condition did not predict positive affect,
b ¼ 7.10, ns.
5. Although this group represents a relatively large proportion of the sample of females
in the year, in absolute terms it is quite small and this, of course, limited the statistical
analyses we could meaningfully conduct on the data.
6. Copies of the instructions given to participants and coders can be obtained from the
corresponding author.
7. Using these data to calculate effect size estimates (Z2) for the interaction effects
indicates the presence of medium-sized effects (Cohen, 1988), explaining 6–10% of the
variance.

References
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions.
Newbury Park, CA: Sage.
Allport, G. W., Vernon, P. E., & Lindzey, G. (1960). Study of values (3rd ed.). Boston:
Houghton-Mifflen.
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.
Blumberg, S. J. (2000). Guarding against threatening HIV prevention messages: An
information-processing model. Health Education & Behavior, 27(6), 780–795.
Brinol, P., Petty, R. E., Gallardo, I., & DeMarree, K. G. (2007). The effect of self-affirmation
in nonthreatening persuasion domains: Timing affects the process. Personality and Social
Psychology Bulletin, 33(11), 1533–1546.
Cohen, G. L., Aronson, J., & Steele, C. M. (2000). When beliefs yield to evidence: Reducing
biased evaluation by affirming the self. Personality and Social Psychology Bulletin, 26(9),
1151–1164.
Cohen, J. (1988). Statistical power analysis for behavioral sciences (2nd ed.). Hillsdale, NJ:
Lawrence Erlbaum Associates, Inc.
Croyle, R. T., & Sande, J. N. (1988). Denial and confirmatory search: Paradoxical
consequences of medical diagnosis. Journal of Applied Social Psychology, 18, 473–490.
Fein, S., & Spencer, S. J. (1997). Prejudice as self-image maintenance. Journal of Personality
and Social Psychology, 73(1), 31–44.
Harris, P. R., Mayle, K., Mabbott, L., & Napper, L. (2007). Self-affirmation reduces smokers’
defensiveness to graphic on-pack cigarette warning labels. Health Psychology, 26, 437–
446.
Harris, P. R., & Napper, L. (2005). Self-affirmation and the biased processing of health-risk
information. Personality and Social Psychology Bulletin, 31(9), 1250–1263.
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.
Klein, W. M. P., Blier, H. K., & Janze, A. M. (2001). Maintaining positive self-evaluations.
Motivation and Emotion, 25(1), 23–40.
Klein, W. M. P., Harris, P. R., & Napper, L. (2008). Self-affirmation and cognitive orientation
to threat. Manuscript in preparation.
Liberman, A., & Chaiken, S. (1992). Defensive processing of personally relevant health
messages. Personality and Social Psychology Bulletin, 18, 669–679.
McGuire, W. J., & McGuire, C. V. (1996). Enhancing self-esteem by directed-thinking tasks.
Journal of Personality and Social Psychology, 70, 1117–1125.
McQueen, A., & Klein, W. M. P. (2006). Experimental manipulations of self-affirmation: A
systematic review. Self and Identity, 5(4), 289–354.
Peterson, C., & Seligman, M. E. P. (2004). Character strengths and virtues: A handbook and
classification. Oxford, UK: Oxford University Press.
62 L. Napper et al.

Raghunathan, R., & Trope, Y. (2002). Walking the tightrope between feeling good and being
accurate. Journal of Personality and Social Psychology, 83, 510–525.
Reed, M. B., & Aspinwall, L. G. (1998). Self-affirmation reduces biased processing of health-
risk information. Motivation and Emotion, 22(2), 99–132.
Robins, R. W., Hendin, H. M., & Trzesniewski, K. H. (2001). Measuring global self-esteem:
Construct validation of a single-item measure and the Rosenberg Self-Esteem Scale.
Personality and Social Psychology Bulletin, 27, 151–161.
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton
University Press.
Schmeichel, B. J., & Martens, A. (2005). Self-affirmation and mortality salience: Affirming
values reduces worldview defense and death-thought accessibility. Personality and Social
Psychology Bulletin, 31, 658–667.
Sherman, D. A. K., & Cohen, G. L. (2006). The psychology of self-defense: Self-affirmation
theory. Advances in Experimental Social Psychology, 38, 183–242.
Sherman, D. A. K., Nelson, L. D., & Steele, C. M. (2000). Do messages about health risks
threaten the self? Increasing the acceptance of threatening health messages via self-
affirmation. Personality and Social Psychology Bulletin, 26(9), 1046–1058.
Spencer, S. J., Fein, S., & Lomore, C. D. (2001). Maintaining one’s self-image vis-à-vis others.
Motivation and Emotion, 25(1), 41–65.
Steele, C. M. (1988). The psychology of self-affirmation: Sustaining the integrity of the self. In
L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 21, pp. 261–302).
New York: Academic Press.
Steele, C. M., & Liu, T. J. (1983). Dissonance processes as self-affirmation. Journal of
Personality and Social Psychology, 45(1), 5–19.
Tesser, A., & Cornell, D. P. (1991). On the confluence of self processes. Journal of
Experimental Social Psychology, 27, 501–526.