Journal of Social and Clinical Psychology, Vol. 29, No. 6, 2010, pp.

668-700

PERRY ET AL.
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE

Bad Starts and Better Finishes:

Attributional Retraining and
Initial Performance in
Competitive Achievement Settings
Raymond P. Perry
University of Manitoba

Robert H. Stupnisky
Laval University

Nathan C. Hall
University of Maryland, College Park

Judith G. Chipperfield
University of Manitoba

Bernard Weiner
University of California at Los Angeles

Transitions to new achievement settings are often accompanied by unfamiliar

learning conditions wherein individuals experience unanticipated failures and
engage in dysfunctional explanatory thinking. To counter these developments,
attributional retraining (AR) was presented to 457 first-year students following

This study was supported by research grants to the first author from the Social
Sciences and Humanities Research Council of Canada (501-2002-0059; 410-2007-2225)
and the Alexander von Humboldt Foundation, by graduate fellowships to the second
and third authors from the Social Sciences and Humanities Research Council of Canada,
and by a research grant to the fourth author from the Canadian Institutes for Health
Research (SSC-42790).
The authors wish to acknowledge the efforts of Steve Hladkyj and Joelle Ruthig
in contributing to the development of this project, and to thank Reinhard Pekrun for
his comments on an earlier draft. Parts of the research were presented at the annual
meeting of the Society for Personality and Social Psychology, February, 2006.
Address correspondence to Raymond P. Perry, Department of Psychology, The
University of Manitoba, Winnipeg, Manitoba, Canada, R3T 2N2; E-mail: rperry@
cc.umanitoba.ca

668
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 669

an initial test in a two-semester course. A Semester 1 AR treatment (no, yes) and

initial-test-performance (low, average, high) 2 x 3 quasi-experimental design was
used to assess Semester 2 attributions, emotions, and performance outcomes. AR
encouraged all students to endorse controllable attributions and de-emphasize
uncontrollable attributions in explaining achievement outcomes in Semester
2. For low- and average-initial-performance students, AR improved subsequent
in-class tests, final course grades, and first-year GPAs. Higher initial-test-perfor-
mance was related to positive emotions and better achievement in Semester 2.
The discussion focused on the implications of AR for attributional thinking in
unfamiliar achievement settings.

Decades of evidence point to the negative effects of failure out-

comes and unpredictable (noncontingent) events on human behav-
ior (Glass & Singer, 1971; Schultz, 1976; Skinner, 1996; Wortman &
Brehm, 1975). To account for these effects, researchers have focused
on how individuals explain failure and unpredictability using caus-
al attributions. Dysfunctional attributional thinking is believed to
impair an array of human endeavors, from goal striving in achieve-
ment settings to physical and psychological well-being in social
interactions (Abramson, Seligman, & Teasedale, 1978; Perry, 1991;
Seligman, 1975; Weiner, 1972, 1979, 1985, 1995).
With regard to achievement settings, dysfunctional attributional
thinking is especially likely during times of transition such as from
high school to college. These transitions are characterized by unfa-
miliarity and more challenging learning conditions that exacerbate
failure and unpredictability. In these situations, dysfunctional at-
tributional thinking erodes goal striving through the development
of problematic outcome-attribution associations such as attributing
success to good luck or failure to bad teaching, implying that the
causes of success and failure are uncontrollable by the individual.
These developments are supported by Perry, Stupnisky, Daniels,
and Haynes’ (2008) “school-to-university transitions” study involv-
ing five different cohorts of first-year students that entered univer-
sity over a 10-year period. They found that students typically en-
dorsed multiple attributions to explain poor performance just after
making the transition to a new learning environment and that over
40% showed some form of dysfunctional attributional thinking.
These 40% were made up of relinquished control students who relied
on bad luck, low ability, test difficulty, and poor teaching to explain
poor performance, and devalued control students who discounted ef-
fort and strategy. As such, the former group is emphasizing uncon-
trollable causes, whereas the latter group is relying on controllable
670 PERRY ET AL.

causes to explain unsatisfactory performance. From Weiner’s (1985,

2006) perspective, both groups are exhibiting dysfunctional attribu-
tional thinking because each attribution combination impairs moti-
vation and goal striving. For these students, attributional retraining
(AR) represents a viable treatment to instill more adaptive thinking
patterns.
In the present study, an AR treatment, designed to promote con-
trollable attributions based on Weiner’s (1972, 1979, 1985, 1995) the-
ory, was administered to students at the start of their undergradu-
ate studies. Our guiding premise was that unfamiliar and adverse
learning conditions in school-to-university transitions can foster
uncertainty, laying the grounds for dysfunctional attributional
thinking. Substantial differences in study strategies, note-taking,
time-management, autonomy, etc., between high school and college
may lead first-year students to underestimate the controllability of
academic success. This puts them at risk to develop dysfunctional
attributional thinking following failure and to experience motiva-
tional deficits. Assuming that first-year students are in a state of
flux and that their attributions are especially malleable during this
transition phase, these students are well suited to benefit from attri-
butional retraining (Perry, 2003; Perry, Hall, & Ruthig, 2005; Perry,
Hechter, Menec, & Weinberg, 1993).

Attributional Thinking in Achievement Settings

Accumulating research shows that attributional retraining (AR) can

improve performance in achievement settings; however, notable
gaps exist in the literature (Perry & Hall, 2009; Perry et al., 1993;
Wilson, Damiani, & Shelton, 2002). Multiple attributions are rarely
measured in AR field studies, and when they have been, they were
not examined as integral combinations. Rather, attributions have
typically been analyzed individually, ruling out the assessment of
AR effects on more complex attributional thinking (e.g., Haynes,
Ruthig, Perry, Stupnisky, & Hall, 2006; Van Overwalle & De Metse-
naere, 1990). These studies preclude a more fine-grained analysis
that would show, for example, that AR can result in an emphasis on
effort and strategy and a de-emphasis on test difficulty, to promote
a form of complex attributional thinking that focuses on the control-
lability of performance. In response, the present study explored the
effects of an AR treatment on several attributions so that combina-
tions of attributions could be observed.
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 671

Our study also differs from the few AR field studies that examined
multiple attributions in its focus on students’ initial learning experi-
ences in relation to school-to-university transitions. Some studies
have investigated AR and performance later in a course; however,
they did not consider performance at the beginning when students
are most unfamiliar with their new learning conditions (Andrews
& Debus, 1978; Heller & Ziegler, 1996; Van Overwalle & De Metse-
naere, 1990; Wilson & Linville, 1982). Others administered AR at
the start of the academic year, but did not consider AR and initial
performance within the same study, so that AR efficacy was not as-
sessed for different performance groups (Hall, Perry, Chipperfield,
Clifton, & Haynes, 2006; Ruthig, Perry, Hall, & Hladkyj, 2004). None
of these studies involving school-to-university transitions examined
initial learning experiences and complex attributional thinking.
Finally, researchers have rarely considered achievement-related
emotions in assessing AR treatment effects, and less so in combina-
tion with attributions, precluding an analysis of attribution-emo-
tion linkages in relation to motivation and performance. Weiner’s
(1985, 2000) theory predicts this linkage between initial perfor-
mance outcomes, emotion, and achievement, as in the case when a
lack of effort attribution (controllable) for failure engenders feelings
of guilt, or a low ability attribution (uncontrollable) produces feel-
ings of hopelessness. More recently, Pekrun and colleagues propose
that emotions are pivotal to performance in achievement settings
(Pekrun, 2000; Pekrun, Frenzel, Goetz, & Perry, 2007; Pekrun, Goetz,
Titz, & Perry, 2002). Deactivating emotions such as worry erode per-
formance, whereas activating emotions like pride and hope foster
achievement outcomes. Moreover, Fredrickson (2001) points to the
salutary effects of positive emotions undoing the detrimental effects
of negative emotions.

Attributional Retraining and Initial

Differences in Performance

Given that transitions to new achievement settings are unavoidable,

AR treatments provide an opportunity to attenuate dysfunctional
attributional thinking. Our study was restricted to first-year stu-
dents based on the premise that dysfunctional attributional think-
ing is exacerbated by the shift from familiar to novel, more com-
petitive learning conditions. Adverse learning conditions in school-
672 PERRY ET AL.

to-university transitions include unsatisfactory performance on

the first test in a course which would highlight the uniqueness and
uncertainty of the new learning conditions and promote dysfunc-
tional attributional thinking. Perry et al.’s (2008) results support this
assumption in that students endorsed indeterminate combinations
of causes that differed in controllability, stability, and locus, and as
such, undermined motivation and performance.
In accordance with Weiner’s theory, it was predicted that AR
would instill adaptive attributional thinking, positive emotions, and
better performance in students whose initial academic experiences
were unsatisfactory. Students disappointed with their first evalua-
tion in unfamiliar learning conditions are more likely to search for
explanations and hence are more amenable to an AR treatment. Our
AR treatment was designed to foster controllable attributions (effort
and strategy), recognizing that the intent of AR treatments can dif-
fer, from modifying single attributions to changing the dimensional
properties of attributions. Some treatments, for example, encour-
age effort attributions instead of ability attributions; others attempt
to shift the dimensional property of ability from stable to unstable,
so that ability is depicted as changeable over time (Wilson & Lin-
ville, 1982). Still others promote controllable causes for negative
outcomes, or discourage uncontrollable causes (cf., Haynes, Perry,
Stupnisky, & Daniels, 2009).
In the present study, the AR treatment was designed to promote
controllable attributions, positive affect, and performance in stu-
dents several months after being administered. However, these
positive outcomes were expected to be qualified by the initial per-
formance of students on entering university as determined by a
class test at the start of the academic year. Three initial-performance
groups (low, average, high) were identified to reflect groupings
typically employed by students for social comparison purposes, by
teachers to estimate student progress, and by educational institu-
tions for administrative decisions. Low-performance students were
predicted to be receptive to AR because failure conditions (negative,
important, unexpected) heighten causal search and encourage stu-
dents to be more attentive to the content of the treatment. Thus, AR
was expected to foster controllable attributions, positive emotions,
and better performance in students receiving the treatment more so
than those not receiving the treatment.
Average-performance students were of particular interest be-
cause little is known regarding AR effects on these students, yet
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 673

they make up the majority of students in achievement settings. If

they perceive their performance on their first test just after entering
university as inconsistent with having met stringent admissions cri-
teria, then causal search should increase because the outcome was
negative and unexpected. If so, AR would also benefit them relative
to their counterparts not receiving the treatment. Finally, AR was
not expected to impact high-performance students because success
on the first test would activate less causal search, and having been
successful in high school, they may already have acquired the at-
tributions specified in the treatment (Perry & Struthers, 1994; Perry
et al., 2008).

Method
Participants

The participants were students in a two-semester, multi-section

Introductory Psychology course at a research-intensive university.
Only first-year students were selected (N = 459) based on informa-
tion provided in a Time 1 questionnaire (October) assessing demo-
graphic information including age, gender, high school grades, year
in program, etc. (285 females; 172 males; two did not indicate gen-
der; 92.4% between 17-22 years). Of those who answered the Time
1 questionnaire, 78% (n = 359) responded to a Time 2 questionnaire
(March), and 98% (n = 451) provided Time 3 (June) academic per-
formance information related to class tests, final course grades, and
overall GPAs. Time 2 attrition (22%) was primarily due to forget-
fulness, illness, and completion of experimental credits, and was
consistent with prior studies (Hall et al., 2006; Hall, Hladkyj, Perry,
& Ruthig, 2004: 17%, 26% respectively).
Attrition rates for the AR and No-AR conditions were computed
separately to determine whether they differed over the academic
year. The analyses were based on comparing students’ participation
in the Time 1 versus Time 2 questionnaire sessions, and comparing
their Time 1 questionnaire participation to providing Time 3 GPAs
at the end of the year. Chi-square analyses revealed that attrition
in the AR condition between Time 1 and Time 2 was no different
than in the No-AR condition, χ2(1) = .66, p > .05, or between Time
1 and Time 3, χ2(1) = 1.19, p > .05. Attrition in the AR and No-AR
conditions was comparable for each initial-performance (IP) group
between Time 1 and Time 2: Low IP: χ2(1) = .32, p > .05; Average IP:
674 PERRY ET AL.

χ2(1) = 2.34, p > .05; High IP: χ2(1) = 2.34, p > .05; and between Time
1 and Time 3: Low IP: χ2(1) = .53, p > .05; Average IP: χ2(1) = .15, p >
.05; High IP: χ2(1) = 2.99, p > .05.

Procedure

Students from several sections of an Introductory Psychology

course volunteered for a three-part study in which treatment condi-
tions (AR, No-AR) were randomly assigned to experimental ses-
sions prior to the start of the study. Students selected one of several
session times to complete a Time 1 questionnaire without knowing
which treatment condition (AR or No-AR) would occur in a given
session. Only one experimental condition was administered during
a specific session, and each session was comprised of students from
different sections of the course who selected a session time based
on availability. During the AR treatment session, students first com-
pleted the Time 1 questionnaire, and then received AR. In the No-
AR condition, students responded to the Time 1 questionnaire and
left immediately after completing it.
Time 1 questionnaire/treatment sessions (October) were sched-
uled only after students received feedback on the first test in Intro-
ductory Psychology at the start of the first semester. Sessions for the
Time 2 questionnaire (March) containing the attribution and emo-
tion measures were conducted midway through the second semes-
ter. After the course was completed, Time 3 (June) test results and
final course grades were obtained from course instructors, and cu-
mulative GPAs were provided by the Office of Institutional Records
for consenting students.

Independent Variables and Covariates

Initial-Performance Groups (Time 1). Three initial-performance

groups were created to reflect typical classifications in university
courses (low, average, high) using students’ scores on their first
class test in October (M = 67.96, SD = 14.45, range = 28-98). The low
initial-performance group had test scores less than 60% (M = 49.60,
SD = 7.46, n = 127; letter grades = F to D); the average initial-perfor-
mance group had scores ranging from 60% to 79.9% (M = 69.76, SD
= 5.57, n = 221; letter grades = C to B); the high initial-performance
group had scores ranging from 80% to 100% (M = 86.03, SD = 4.78,
n = 107; letter grades = B+ to A+).
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 675

Though grouping cutoffs in classrooms and other achievement

settings can vary, the means in our study were reasonably represen-
tative (low M ~ 50%; average M ~ 70%; high M ~ 86%), and reflected
letter grade groupings used in many college classrooms (e.g., low
= F or D; average = B; high = A). The ecological reality of college
classrooms is that letter grades are determined by such cutoffs, as
for example, when 69.1% = C+, and 70% = B. In addition, receiving
a B instead of a C+ on a test often has notable psychological mean-
ing to a student, even though the actual difference between 70% and
69.1% may be inconsequential statistically.
Attributional Retraining (Time 1). The AR treatment was adminis-
tered in sessions of 30-50 students after students received feedback
on their first class test. In the AR sessions (n = 200), the treatment
was introduced following the Time 1 questionnaire; in the no-treat-
ment (No-AR) sessions (n = 259), students responded to the Time 1
questionnaire and then left the session. The AR treatment was made
up of three separate components administered during a one-hour
session: causal search activation, attribution induction, and attribu-
tion consolidation.
The causal search activation component involved a task designed to
initiate attributional thinking about the causes of success and fail-
ure and coincided with feedback on the first tests in their courses.
The task required students to respond to 10 items that assessed their
attributions for failure and success in their courses (e.g., If I study
in appropriate ways, then I will be able to learn the material in my
courses) 1 (strongly disagree), 10 (strongly agree). Students also
completed two items requiring them to estimate how much their
ability and effort influenced their performance in their psychology
course, 1 (not at all), 10 = (very much so).
Following the activation component, the attribution induction com-
ponent constituted a 10-minute videotape designed to encourage
controllable causal attributions as explanations of unsatisfactory
performance (e.g., Hall et al., 2006; Perry & Struthers, 1994; Stru-
thers & Perry, 1996). The videotape depicts two students discussing
how poor performance can be changed and how their performance
can subsequently improve. The dialogue focused on controllable at-
tributions which were summarized at the end by a male professor
who noted that students’ explanations of their university experienc-
es can have direct consequences for performance. He emphasized
that, by viewing failure in terms of controllable causes, students can
676 PERRY ET AL.

modify their academic behaviors and thereby improve future per-

formance.1
The third AR component, attribution consolidation, was comprised
of two tasks intended to crystallize the content of the AR videotape
by having students think about the causes of success and failure.
The first consolidation task was a GRE-type aptitude test made up
of verbal analogy, quantitative, and sentence completion sections,
having 10, 5, and 10 items respectively. Students had five minutes
to complete each section, after which they rated their performance
on the test and their perceived success. The aptitude test was in-
tentionally difficult to heighten students’ attention on performance
and thereby highlight the content of the videotape. The second con-
solidation task involved a discussion of the videotape by the experi-
menter who highlighted adaptive and maladaptive attributions. On
leaving the session, students received a one-page AR handout in-
tended to reinforce the videotape by summarizing its attributional
content. It did so by listing maladaptive attributions on the left side
of the page and adaptive attributions on the right side. Students
were encouraged to keep the handout accessible in their course
notes or near their desk as a reminder.
Thus, the attribution consolidation component involved a GRE
test and a group discussion to encourage students to actively seek
explanations of success and failure (causal search). The attributions
most readily available in memory were expected to be those pre-
sented in the AR videotape. The one-page handout was intended to
reinforce the attributions introduced in the AR videotape through
a visual reminder. In following the AR videotape, these activities
were intended to prime, reinforce, and consolidate the attributions
contained in the AR treatment, so that they were retained after the
students returned to their classrooms and study areas.
1. The ability attribution was not included in our AR treatment for several reasons.
First, controllable and unstable attributions are modifiable and consequently
more sensitive indicators of AR effectiveness. Ability is typically regarded as more
uncontrollable and stable than effort and strategy, and thus is less modifiable (but
see Dweck, 1999, entity vs. incremental; Weiner, 1983). Moreover, compared to other
attributions, attributions to low ability are more likely to elicit self-protective strategies
linked to self-concept, creating ego-defensive strategies and hence measurement
problems. Following failure, self-protective strategies and the self-serving bias (Heider,
1958) can inhibit an attribution to low ability because of its negative consequences for
self-worth (Covington, 1997), self-efficacy (Bandura, 1997), or self-esteem (Rosenberg,
1965). In having an external- rather than internal-locus, professor quality and test
difficulty were assumed to be more amenable to measurement as uncontrollable
attributions than ability which introduces self-worth complications.
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 677

Age (Time 1 Covariate). Students indicated their age on a 10-point

scale (1 = 17-18, 2 = 19-20, 10 = older than 45) in the Time 1 question-
naire. First-year students were selected whose ages ranged from 17
to 30 years, the modal response being 17-18 years (n = 305). Because
the study deals with individuals confronting novel learning condi-
tions, first-year college students provide a suitable target popula-
tion. However, some first-year students may have more life experi-
ence in dealing with novel learning situations due to age, and hence
differ from younger, less experienced students. For this reason, a
restricted sample was selected in which the majority of students
were between 17-22 years. We used 31 years and older as a cut off
to exclude extreme outliers, and age as a covariate to further control
any confounding effects due to age.
Self-Reported High School Grade (Time 1 Covariate). High school
grade (HSG) was obtained from the Time 1 questionnaire by ask-
ing students to indicate their overall average performance in their
last year of high school (1 = 50% or less, 2 = 51-55%, ... 9 = 86-90%,
10 = 91-100%; M = 7.15 or approximately 80%, SD = 1.68, range =
2-9). HSG was used to control for pre-existing aptitude/academic
differences instead of SAT and ACT scores because the latter are
not assessed in Canadian high schools. Self-reported HSG was used
as a proxy for actual high school achievement based on previous
studies showing a strong relation between self-report HSG and ac-
tual HSG, e.g., r(670) = .84 (Hall et al., 2007), r(464) = .84 (Perry,
Hladkyj, Pekrun, Clifton, & Chipperfield, 2005). Self-reported HSG
in our study correlated with final course grades, r(456) = .40, and
1st -year GPA, r(459) = .54., the latter consistent with Perry et al.’s
(2005) findings for 1st- , 2nd-, and 3rd-year GPAs, rs(345) = .52, .52,
and .53 respectively.

Dependent Measures

Causal Attributions (Time 2). The impact of the AR treatment on

four causal attributions was assessed in the second semester Time
2 questionnaire. Students responded to the following question, “To
what extent does each of the following factors influence your per-
formance in your Introductory Psychology course?” (1 (not at all,
10 (very much so): strategy (M = 7.06, SD = 1.93, range = 1-10); ef-
fort (M = 8.18, SD = 1.92, range = 1-10); professor quality (M = 7.24,
SD = 2.00, range = 1-10); test difficulty (M = 7.38, SD = 1.97, range =
678 PERRY ET AL.

1-10). Strategy and effort represented controllable attributions, and

professor quality and test difficulty uncontrollable attributions, and
are the most common attributions used in university classrooms
(Perry et al., 2008). The controllability of these attributions can vary
depending on the phenomenology of the student, but is generally
considered to be as specified herein (Van Overwalle, 1989; Weiner,
1985, 2006). The inclusion of professor quality and test difficulty at-
tributions specified critical contextual factors associated with learn-
ing environments in new achievement settings.
Achievement Emotions (Time 2). The Time 2 questionnaire also as-
sessed the impact of AR on five attribution-related emotions and
one outcome-related emotion (Weiner, 1985) using 10-point scales,
1 (not at all), 10 (very much so). Students rated the extent to which
they experienced each emotion in relation to their course perfor-
mance: hope (M = 6.85, SD = 1.90, range = 1-10); pride (M = 5.64,
SD = 2.48, range = 1-10); shame (M = 3.02, SD = 2.56, range = 1-10);
guilt (M = 3.73, SD = 2.62, range = 1-10); helplessness (M = 2.93, SD =
2.16, range = 1-10). Helplessness was viewed as a less severe vari-
ant of hopelessness related to the stability dimension in Weiner’s
taxonomy. An outcome-dependent emotion, worry, was included
to represent a general performance-related emotional reaction (M =
4.72, SD = 2.62, range = 1-10).
Test 2 Performance (Time 3). Representing a course-specific measure
of performance, a class test was administered in Semester 2 roughly
one week after students completed the Time 2 questionnaire (M =
69.54, SD = 15.68, range = 28-100).
Course Percentage (Time 3). Representing a second domain-specific
measure of performance, students’ course grade (cumulative per-
centage) was based on all tests completed in the course following
the AR treatment in Semester 1 (M = 68.31, SD = 12.78, range =
35.25-97.75). The first class test (Test 1) was excluded from course
percentage because it was used to define the initial-performance
groups (low, average, high), and because it preceded AR.
Grade-Point-Average (Time 3). First-year GPA represented a general
measure of performance obtained from institutional records several
months after the second semester ended. The GPA measure was
comprised of the average grade received in all first- and second-
semester courses in a student’s first year. GPA is recorded as a nu-
merical value having letter grade equivalents as follows: 0 = F, 1 =
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 679

D, 2 = C, 2.5 = C+, 3 = B, 3.5 = B+, 4 = A, 4.5 = A+ (M = 2.72, SD =

0.90, range = 0.37-4.50).

Results

An AR (No-AR, AR) by initial test performance (low, average, high)

2 X 3 factorial design was used for analysis purposes, with age and
self-reported high school grade included as covariates. Students
were classified into initial test performance groups based on the
first class test (Test 1). The dependent variables consisted of causal
attributions and emotions, measured on the Time 2 questionnaire,
and Time 3 domain-specific and general performance outcomes re-
lated to Test 2 scores, final course grades (percentage), and GPA (see
Table 1).
ANCOVAs and MANCOVAs were used depending on the type
of dependent variables involved (performance, attributions, emo-
tions). We chose ANCOVA for the performance measures (Test 2,
course percentage, GPA) because these outcomes were not deemed
to have sufficient theoretical integrity, as compared to the attribu-
tion and affect measures, to warrant the use of MANCOVA. We em-
ployed .01 rather than .05 as the p-value to protect against Type-1
errors arising from computing multiple ANCOVAs in the absence
of an omnibus MANCOVA.
Our hypotheses involved simple main effects for the three
achievement measures in which AR versus No-AR differences
were assessed separately for each initial performance group with
a priori, one-tailed t-tests (p < .01). A stringent rejection rate and
Welch’s t-tests were used to correct for cell size imbalances and to
protect against violations of the homogeneity of variance assump-
tion. We used a conventional ANCOVA in order to include age and
high school grades as covariates, and tested our hypotheses with
simple a priori t-tests which do not require the interaction effect to
be significant (Keppel & Wickens, 2004; Stevens, 2002; Tabachnick
& Fidell, 2007). Interaction hypotheses may be worthy of further
investigation in which AR is predicted to have a stronger effect on
achievement for one initial performance group compared to anoth-
er. However, this approach would require a level of specification
not warranted by prior research, and hence the omnibus interaction
was not probed.
 TABLE 1. Means and Standard Deviations for the Variables
Low performancea Average performancea High performancea
Measureb No-AR AR No-AR AR No-AR AR
Strategy
M (SD) 6.11 (2.20) 7.11 (1.79) 6.84 (2.05) 7.31 (1.66) 7.18 (1.79) 8.17 (1.61)
Adj. M (SE) 6.10 (.32) 7.11 (.26) 6.84 (.20) 7.32 (.21) 7.19 (.24) 8.18 (.40)
n 36 53 94 83 67 24
Effort
M (SD) 7.14 (2.59) 7.96 (1.92) 8.03 (1.95) 8.51 (1.46) 8.39 (1.82) 9.04 (1.71)
Adj. M(SE) 7.22 (.32) 8.01 (.26) 8.06 (.20) 8.48 (.21) 8.35 (.24) 8.92 (.40)
n 36 53 94 83 67 24
Professor quality
M (SD) 6.42 (2.06) 7.25 (2.02) 7.49 (1.81) 7.31 (2.03) 7.21 (2.06) 7.17 (2.14)
Adj. M(SE) 6.41 (.34) 7.24 (.28) 7.49 (.21) 7.32 (.22) 7.21 (.25) 7.18 (.42)
n 36 53 94 83 67 24
Test difficulty
M (SD) 6.83 (2.27) 7.28 (1.73) 7.61 (1.97) 7.49 (2.04) 7.03 (1.90) 7.83 (1.74)
Adj. M(SE) 6.77 (.33) 7.25 (.27) 7.57 (.21) 7.51 (.22) 7.10 (.25) 7.93 (.41)
n 36 53 94 83 67 24
Hope
M (SD) 6.49 (2.00) 6.25 (1.78) 6.73 (1.78) 6.95 (1.87) 7.09 (1.96) 7.83 (1.83)
Adj. M(SE) 6.49 (.32) 6.25 (.26) 6.75 (.19) 6.97 (.21) 7.05 (.23) 7.83 (.39)
n 35 53 94 83 67 24
680 PERRY ET AL.
Pride
M (SD) 4.60 (2.38) 4.75 (2.73) 5.19 (2.35) 5.70 (2.33) 6.93 (2.06) 7.21 (1.98)
Adj. M(SE) 4.46 (.40) 4.67 (.32) 5.14 (.24) 5.77 (.26) 6.97 (.29) 7.44 (.49)
n 35 53 94 83 67 24
Shame
M (SD) 3.92 (2.72) 3.74 (2.96) 3.23 (2.66) 3.12 (2.68) 1.80 (1.38) 2.17 (1.74)
Adj. M(SE) 3.91 (.42) 3.74 (.34) 3.22 (.26) 3.10 (.27) 1.85 (.31) 2.17 (.52)
n 36 53 94 83 66 24
Guilt
M (SD) 4.44 (2.78) 4.08 (2.83) 4.02 (2.64) 3.91 (2.63) 2.55 (1.92) 3.13 (2.42)
Adj. M(SE) 4.44 (.43) 4.07 (.36) 4.00 (.27) 3.89 (.28) 2.61 (3.2) 3.13 (.54)
n 36 52 94 81 67 24
Helpless
M (SD) 4.67 (2.41) 3.42 (2.37) 2.98 (2.07) 2.82 (2.11) 2.22 (1.89) 2.25 (1.48)
Adj. M(SE) 4.11 (.36) 3.38 (.29) 2.94 (.22) 2.82 (.23) 2.30 (.26) 2.33 (.44)
n 36 53 94 83 67 24
Worry
M (SD) 6.17 (2.62) 5.87 (2.79) 4.94 (2.49) 4.55 (2.38) 3.13 (1.94) 3.83 (2.62)
Adj. M(SE) 6.11 (.41) 5.83 (.34) 4.89 (.25) 4.54 (.27) 3.24 (.30) 3.91 (.51)
n 36 52 94 83 67 24
Pre-AR Test 1
M (SD) 49.22 (8.09) 49.93 (7.03) 70.12 (5.66) 69.30 (5.46) 86.40 (5.00) 85.04 (4.09)
Adj. M(SE) 50.23 (.81) 50.52 (.70) 70.53 (.53) 68.95 (.59) 85.54 (.67) 83.36 (1.16)
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE

n 55 71 123 98 79 27
681
 TABLE 1. (continued)
Low performancea Average performancea High performancea
b
Measure No-AR AR No-AR AR No-AR AR
Post-AR Test 2
M (SD) 53.48 (13.42) 66.82 (15.66) 64.15 (12.43) 76.66 (13.29) 77.85 (11.55) 82.79 (12.67)
Adj. M(SE) 54.65 (1.84) 67.21 (1.57) 64.58 (1.57) 76.56 (1.34) 76.77 (1.51) 81.18 (2.59)
n 52 70 116 95 78 27
Course percent (%)
M (SD) 55.88 (10.29) 60.28 (11.86) 66.35 (9.26) 71.66 (10.80) 78.99 (8.29) 81.87 (8.55)
Adj. M(SE) 57.52 (1.34) 61.12 (1.14) 66.97 (.88) 71.17 (.99) 77.58 (1.12) 79.29 (1.93)
n 55 74 123 98 79 27
GPA
M (SD) 1.81 (.81) 2.15 (.71) 2.57 (.79) 3.05 (.62) 3.35 (.60) 3.72 (.70)
Adj. M(SE) 2.02 (.09) 2.27 (.08) 2.65 (.06) 2.98 (.07) 3.19 (.07) 3.38 (.13)
n 54 72 121 97 79 26
Notes. aInitial-performance (Pre-AR Test 1). bUnadjusted and adjusted means for high school grade and age.
682 PERRY ET AL.
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 683

Preliminary analyses were conducted to identify pre-existing dif-

ferences between AR conditions (No-AR vs. AR) on the first class
test (Test 1) prior to the administration of the treatment in Semester
1. A treatment (No-AR, AR) by initial-performance (low, average,
high) 2 X 3 ANCOVA revealed no differences between the treatment
conditions on pre-AR Test 1, F(1,445) = 3.37, p = .07. Gender (female,
male) and first language (English-No, English-Yes) were entered as
covariates in ANCOVAs for Test 2, course percentage, and GPA out-
comes. Gender and first language produced no significant effects
on the three performance measures, and consequently were not in-
cluded in subsequent analyses.

Academic Performance ANCOVAs

Post-AR Class Test (Test 2). A 2 x 3 ANCOVA revealed an initial-per-

formance main effect on Test 2 results, F(2,430) = 40.79, p < .01, ηp2 =
.16, in which high-performance students (M = 78.98, SD = 15.72) did
better than average-performance students (M = 70.57, SD = 13.03),
t(177) = 4.73, p < .01, d = .60, who in turn outperformed low-perfor-
mance students (M = 60.93, SD = 13.56), t(224) = 6.15, p < .01, d = .73
(see Tables 1, 2, and 3 for means, SDs, ANCOVA effects).
More importantly, an AR main effect, F(1,430) = 46.88, p < .01,
ηp2 = .10, indicated that the AR group (M = 74.98, SD = 13.71) per-
formed better than the No-AR group (M = 65.33, SD = 15.12). As dis-
played in Figure 1, simple main effects (Welch’s) t-tests reveal that
low- and average-performance AR students did better than their
No-AR counterparts: t(109) = 5.19, p < .01, d = .96; t(201) = 6.61, p <
.01, d = .92 respectively. In contrast, high-performance AR students
did not do better than No-AR students. Of note, the AR treatment
boosted the Test 2 scores of low-and average-performance students
by roughly 12%, twice that of high-performance students (see Fig-
ure 2, left-hand bar graphs). These AR effects on Test 2 outcomes are
consistent with AR effects on other class tests in Semesters 1 and 2
of the course.2
Course Percentage (%). A 2 x 3 ANCOVA for course percentage
revealed an initial-performance main effect, F(2,448) = 82.71, p <
.01, ηp2 = .27, in which high-performance students (M = 78.44, SD
= 11.79) outperformed average-performance students (M = 69.07,
SD = 9.74), t(175) = 7.10, p < .01, d = .90, who in turn did better than
 TABLE 2. F Table of AR and Initial-Performance (IP) Main Effects and Interactions on Academic Performance
High school grade Attributional Initial-performance AR X IP (Pre-AR
Age covariate covariate retraining (AR) (Pre-AR Test 1) Test 1)
Measure MSE df MS F MS F MS F MS F MS F
Post-AR Test 2 167.76 430 2298.59 13.70** 1037.94 6.19* 7864.86 46.88** 6842.36 40.79** 489.00 2.92†
Course percent 93.61 448 1237.40 13.22** 2484.75 26.54** 870.10 9.30** 7742.70 82.71** 45.57 0.49
GPA 0.41 441 5.29 12.89** 41.11 100.15** 5.70 13.89** 28.85 70.29** 0.14 0.34
Note. Numerator df = 1 for all F tests on age, high school grade and AR. Numerator df = 2 for Post-AR Test 2 performance and AR X initial-performance interaction. †p =
.055; *p< .05; **p < .01.
684 PERRY ET AL.
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 685

Figure 1. AR x Initial-performance (IP, Pre-AR Test 1) on Post-AR Test 2.

Note. Means are for performance on Test 2 only, but reflect the same
pattern for course percentage and GPA outcomes. They are adjusted for
high school grade and age.

low-performance students (M = 59.32, SD = 10.15), t(246) = 8.79, p

< .01, d = .99.
Similar to Test 2 results, an AR main effect, F(1,448) = 9.30, p < .01,
ηp2 = .02, revealed that AR students (M = 70.53, SD = 11.41) did bet-
ter than their No-AR counterparts (M = 67.36, SD = 10.24). Simple
main effects t-tests indicated that AR increased overall course per-
centage for the low-performance students, t(127) = 2.05, p = .02, d =
.37, and for average-performance students, t(207) = 3.17, p < .01, d =
.43, but not for high-performance students (Figure 2).
Grade-Point-Average (GPA). A 2 x 3 ANCOVA for first-year GPA
(two semesters) revealed an initial-performance main effect, F(2,
441) = 70.29, p < .01, ηp2 = .24, in which high-performance students
(M = 3.28, SD = .79) attained better GPAs than average-performance
students (M = 2.81, SD = .65), t(174) = 5.31, p < .01, d = .68, who in
turn had better GPAs than low-performance students (M = 2.14, SD
= .67), t(342) = 9.12, p < .01, d = 1.02.
2. Test 2 was selected as a post-AR performance measure because it immediately
followed the Time 2 questionnaire assessing attributions and emotions in Semester 2.
Comparable ANCOVA results emerge on a course test immediately following the AR
treatment in Semester 1: an AR treatment effect, F(1,444) = 11.79, p = .001, η2 = .03; an
initial-performance effect, F(2,444) = 62.29, p < .001, η2 = .22; and a nonsignificant AR x
initial-performance interaction.
686 PERRY ET AL.

TABLE 3. Welch’s t-tests for the Effect of Attributional Retraining Across

Initial-Performance (IP) Groupsa
AR vs.
No-AR N Mean SE df t d
Post-AR Test 2
Low IP No-AR 52 54.65 1.84 109 5.19*** .96
AR 70 67.21 1.57
Average IP No-AR 116 64.58 1.22 201 6.61*** .92
AR 95 76.56 1.34
High IP No-AR 78 76.77 1.51 44 1.47 .33
AR 27 81.18 2.59
Course percent
Low IP No-AR 55 57.52 1.34 115 2.05* .37
AR 74 61.12 1.14
Average IP No-AR 123 66.97 .88 207 3.17*** .43
AR 98 71.17 .99
High IP No-AR 79 77.58 1.12 44 0.77 .17
AR 27 79.29 1.93
GPA
Low IP No-AR 54 2.02 .09 115 2.08** .39
AR 72 2.27 .08
Average IP No-AR 121 2.65 .06 201 3.58*** .51
AR 97 2.98 .07
High IP No-AR 79 3.19 .07 40 1.29 .30
AR 26 3.38 .13
Note. Means adjusted for high school grade and age. All t-tests were one-tailed. aInitial-performance
(Pre-AR Test 1). *p < .02, **p < .01, ***p < .001.

As with the Test 2 and course percentage results, an AR main ef-

fect, F(1, 441) = 13.89, p < .=01, ηp2 = .03, showed that the AR group
(M = 2.88, SD = .75) had higher GPAs than the No-AR group (M =
2.62, SD = .69). Simple main effects t-tests indicated that, for low-
and average-performance students, the AR groups had higher
GPAs than the No-AR groups: t(115) = 2.08, p < .01, d = .39; t(216)
= 3.73, p < .01, d = .51 respectively. For high-performance students,
the AR groups did not differ from the No-AR groups in their GPAs
(see Figure 2).
In sum, the AR treatment had consistent effects across course-
specific (Test 2, course percentage) and general performance (GPA)
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 687

Figure 2. AR Effects on Achievement Outcomes for Initial-Performance

(Test 1) Groups.
Note. Values represent effect sizes for the three performance measures
for comparison purposes. They are adjusted for high school grade and
age.

measures. AR generated sizable gains in students whose Test 1 per-

formance was low or average, the very students for whom improve-
ments were most desirable. The magnitudes of these AR effects are
substantial and strikingly reliable across all three performance mea-
sures (Figures 1 & 2). The larger AR effect on Test 2 performance
was likely due to Test 2 being more proximal to the AR treatment
than course percentage, and more content-specific than the tests
and assignments comprising course performance.

Causal Attributions and Emotions MANCOVAs

Treatment (No-AR, AR) by initial-performance (low, average, high)

2 x 3 Wilks’ lambda MANCOVAs were used to assess whether AR
and initial-performance related to causal attributions and emotions
later in the academic year (Time 2 questionnaire). Separate MAN-
COVAs were computed for causal attributions and for emotions,
with age and high school grades included as covariates as in the
ANCOVA analyses (all effects assessed at p < .01).
MANCOVA is well suited to examining complex attributional
thinking and emotion diversity because it is intended to assess
multiple dependent measures simultaneously (Marascuilo & Levin,
1982; Stevens, 2002). This is an improvement over univariate ANO-
688 PERRY ET AL.

TABLE 4. Discriminant Function Analyses for Causal Attributions (Panel A)

and Emotions (Panel B)
Attributional retraining main effect on student attributions
Panel A Structure r z weights
Strategy .961 .927
Effort .686 .123
Test difficulty .292 .168
Professor quality .077 -.325
Initial-performance main effect on student emotions
Panel B
Worry -.877 -.727
Pride .796 .480
Shame -.620 -.020
Helpless -.534 -.020
Guilt -.516 .174
Hope .391 .120

VA procedures because they cannot test multiple dependent vari-

ables, thus limiting interpretation of results involving several mea-
sures. For each significant MANCOVA effect, Discriminant Func-
tion Analysis (DFA) was used as a follow up procedure to identify
the specific structure of the dependent measures arising from the
effect (Tabachnick & Fidell, 2007).
DFA produces a structure or function that specifies unit load-
ings for each measure making up a MANCOVA effect. Structure
coefficients (structure r) denote the simple correlations between
the variables and the function (similar to a factor analysis loading),
whereas structure weights specify the unique partial contribution
of each variable to the discriminant function (similar to a standard-
ized regression coefficient). The unit loadings are weightings for
each measure in the discriminant function ranging from +1 to -1,
and are used to interpret the emphasis given to each measure rela-
tive to others in the overall combination. For the AR and initial per-
formance variables, DFA enabled complex attributional thinking
to be depicted by the emphasis (weightings) students gave to the
four causal attributions (strategy, effort, professor quality, test diffi-
culty). Similarly, emotion diversity was inferred from the emphasis
(weightings) students gave to the six achievement emotions (hope,
pride, shame, guilt, helplessness, worry).
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 689

Causal Attributions. A 2 x 3 MANCOVA revealed an AR effect

on attributions, F(4,346) = 3.55, p < .01, ηp2 = .04 (Wilk’s Lamda =
.961), but no initial-performance effect, F(8,692) = 2.18, p > .01, ηp2 =
.03 (Wilk’s Lamda = .951), or AR x initial-performance interaction,
F(8,692) = 1.00, p > .01, ηp2 = .01 (Wilk’s Lamda = .977). The AR ef-
fect was subjected to a DFA to clarify the structure of the multiple
attributions students used to explain their performance (see Table
4, Panel A). The structure correlations revealed that the controllable
attributions (effort, strategy) were related to the discriminant func-
tion to a sizable degree, whereas the uncontrollable attributions (test
difficulty, professor quality) were related to a much smaller degree.
For the controllable attributions, the standardized weights, in-
dicating each variable’s partial contribution to the discriminant
function, showed strategy related substantially, and positively to the
overall discriminant function (z = .927), whereas effort contributed
little to it (z = .123). For the uncontrollable attributions, professor
quality related moderately negatively to the structure of the dis-
criminant function (z = -.325), and test difficulty contributed very
little to it (z = .168). The resulting structure of this attributional func-
tion is characterized by a strong emphasis on a controllable attribu-
tion (strategy) and a moderate de-emphasis on an uncontrollable
attribution (professor quality).3
Group centroids indicated that the two treatment groups differed
with respect to the overall structure of the attributional function
(No-AR = -.138; AR = .172). Group centroids are interpreted much
like group means, but refer to the function (structure) as a whole, as
defined by the attribution (item) loadings, rather than to a single de-
pendent variable. Based on the highest structure loadings, the attri-
butional function suggests that AR encouraged students to empha-
size strategy in explaining performance and to downplay professor
quality. This pattern is noteworthy because AR fostered a form of
attributional thinking in which a controllable attribution (strategy)
was emphasized in students, while at the same time discounting an
uncontrollable attribution.

3. Because the effort attribution had a significant nonnormal skew (Skewness =

-1.24, Kurtosis = 1.34), a power transformation (i.e., squaring the variable) was used to
reduce the negative skew (M = 70.60, SD = 27.30, Skewness = -.62, Kurtosis = -.59). The
MANCOVA was re-analyzed using the transformed variable and revealed an AR main
effect, F(4,346) = 3.55, p < .01, ηp2 = .04, an initial-performance main effect, F(8,692) =
2.10, p < .05, ηp2 = .03, and no interaction. The results were essentially identical to those
for the untransformed variable which were retained to simplify interpretation.
690 PERRY ET AL.

Achievement Emotions. A 2 X 3 MANCOVA indicated an initial-

performance effect on the six emotions, F(12,674) = 5.07, p < . 01,
ηp2 = .08 (Wilk’s Lamda = .841), but no AR effect, F(6,337) = .830, p
> .01, ηp2 = .02 (Wilk’s Lamda = .985), or AR x initial-performance
interaction, F(12,674) = .832, p > .01, ηp2 = .01 (Wilk’s Lamda = .971).4
A DFA computed for the initial-performance effect showed that a
single emotion function accounted for most of the variance (91%).
The structure correlations (Table 4, Panel B) revealed that the dis-
criminant function positively correlated with positive emotions
(pride, hope), and netgatively correlated with negative emotions
(worry, shame, helplessness, guilt). The standardized weights,
which show each variable’s unique combination to discriminant
function, revealed a more parsimonious pattern of results. Specifi-
cally, the discriminant function had a moderate, positive loading
on pride (.480) and had a strong, negative loading on worry (-.727).
In Weiner’s (1985, 1995) theory, pride is an attribution-dependent
emotion linked to the locus dimension, and worry is an outcome-
dependent emotion not linked to any attribution dimension.
The discriminant function depicts a complex emotion combina-
tion in which the moderate emphasis on a positive emotion (pride)
that would likely foster motivation and the strong de-emphasis on a
negative emotion (worry), likely limiting a depletion of motivation
resources. The centroids for the three performance groups (low =
-.616; average = -.045; high = .669) reveal logically coherent emotion
differences between the three groups consistent with their initial
learning experiences (Test 1). High-performance students had more
pride in their performance and worried less than average-perfor-
mance students who reported more pride and less worry than low-
performance students. This means that the low initial-performance
students had the least pride and the most worry regarding their
course outcomes by the end of the academic year.

4. Shame was also found to be significantly nonnormal (Skewness = 1.24, Kurtosis =

.45) and a root transformation (i.e., square root of the variable) was used to reduce the
positive skew (M = 1.60, SD = .67, Skewness = .84, Kurtosis = -.56). The MANCOVA
for students’ emotions was re-analyzed using the transformed variable, and revealed
an initial-performance main effect, F(8,680) = 5.39, p < .001, ηp2 = .06, however neither
the AR main nor interaction effects were significant. Results were virtually identical to
those found using the untransformed variable, and the original results were again used
to simplify interpretation.
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 691

Discussion

For individuals who struggle in achievement settings, the remedia-

tion of failure is critical; and more so for those who fail repeatedly.
Attributional retraining (AR) is designed to assist failure-prone in-
dividuals under such circumstances (Haynes et al., 2009; Perry et
al., 1993). Our study examined this issue in relation to students’ ini-
tial learning experiences in new achievement settings, represented
by school-to-university transitions. In brief, the AR treatment facili-
tated adaptive attributional thinking in all students, and increased
course-specific and global performance outcomes in students whose
initial learning experiences were less than satisfactory.
Our study advances existing field research on AR treatments with
regards to complex attributional thinking by showing that students
use multiple attributions to explain performance outcomes and by
delineating the structure of their attributional thinking. Following
the AR treatment, students accentuated a controllable attribution
(strategy) in explaining their course performance, while simultane-
ously downplaying an uncontrollable attribution (professor quali-
ty). From a motivation viewpoint, this mind-set is adaptive because
a strategy attribution implies that performance is controllable and
modifiable. De-emphasizing a major uncontrollable cause of perfor-
mance (professor quality) further strengthens students’ perceived
influence over their new learning conditions. The results are notable
because the longitudinal design permitted treatment effects to be
detected several months after the administration of AR.
Our study also adds to the existing research by demonstrating
that the AR treatment had several performance benefits for vulner-
able students who enter a new achievement setting. First, the AR
treatment enabled students whose performance was low or average
on an initial class test to do markedly better than their No-AR coun-
terparts in terms of course-specific achievement outcomes. Their
performance increased by roughly 12% on a class test (Test 2), and
by one letter grade in course percentage, substantial gains in both
statistical and practical terms. These results are important because
these performance outcomes were observed in Semester 2, several
months after AR was given in Semester 1, and because the perfor-
mance gains were consistent across different testing formats that
make up the final grade, such as multiple-choice, short answer and
essay tests, term papers, oral presentations, and so on.
692 PERRY ET AL.

Second, the AR treatment improved the global achievement of

low- and average-performance students, measured in terms of
overall GPAs, in keeping with results for course-specific outcomes
(Test 2, final grades). This GPA increase is notable because the first-
year students were enrolled in 10 one-semester courses in the hu-
manities, social sciences, and basic sciences each year. Thus, AR
improved performance in all first-year courses by as much as one
letter grade, despite differences between courses in student abili-
ties, lecture content, instruction methods, types of assessment, class
size, and class structure. The AR effect sizes are particularly impres-
sive for low- and average-performance students which are almost
three times larger on Test 2, and twice as large for course percent-
age and GPA, than their high-performance counterparts (Figure 2).
These moderate-to-large effect sizes (Cohen, 1988) would be highly
meaningful to students from the viewpoint of grades and overall
academic development.
Finally, the sizable treatment effects observed for average-perfor-
mance students clearly show that AR is effective not just for failure-
prone students, but for a broader range of students than previously
thought, involving the majority of students whose performance is
average. The AR treatment did not increase achievement outcomes
for high-performance students possibly because their pre-existing
attributional mind-sets corresponded to that advocated in the treat-
ment. Alternately, AR may not have increased their performance
because there is an upper limit as to the effectiveness of AR treat-
ments.
Our study also provided a unique opportunity to examine AR
treatment effects on achievement-related emotions; however, our
findings were inconclusive. Although the AR treatment was expect-
ed to influence achievement emotions along with complex attribu-
tional thinking, the results suggest that AR treatments generally, or
ours specifically, have little impact on emotions. This conclusion
may be premature because of the dearth of empirical evidence on
the effects of AR treatments on emotions. One possible explanation
is that the content of AR treatments typically focuses on attribu-
tions and not on emotions, making the elicitation of dimensionally-
related emotions dependent on the activation of attributions. This
distal elicitation of attribution-emotion linkages may fail without
including emotions in the AR treatment itself, suggesting possibili-
ties about the development of future AR treatments that include
both attributions and their dimensionally-related emotions.
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 693

Second, the emotions were assessed with single-item measures in

contrast to approaches that use multi-item scales to examine achieve-
ment related emotions. For example, Pekrun et al. (2002) have devel-
oped multi-item scales that assess the cognitive, affective, physio-
logical, and behavioral components of achievement emotions. Third,
the emotions assessed may not have been the most salient emotions
in the achievement settings in our study compared to other emotions
such as enjoyment, boredom, test anxiety, etc. (Fredrickson, 2001;
Pekrun, 2000). Finally, because emotions are influenced by many fac-
tors in a given situation, they are over-determined and may not be
readily influenced by simple AR treatments.

The Content of Complex Attributional Thinking

The AR treatment in our study encouraged a form of attributional

thinking in which a controllable cause of poor performance (strat-
egy) was accentuated and an uncontrollable cause discounted (pro-
fessor quality). This pattern portrays a mind-set more conducive
to motivation than the undifferentiated combinations of causes in
novel learning conditions identified by Perry et al. (2008). During
school-to-university transitions, it appears that first-year students
generally use an indeterminant combination of internal, unstable,
controllable attributions (effort, strategy), along with external, sta-
ble, uncontrollable attributions (test difficulty, professor quality).
Our AR treatment simplified this mixed mind-set by encouraging
attributional thinking which enables students to concentrate their
cognitive resources and efforts on a more focused, controllable solu-
tion to poor performance.
The attributional thinking induced by our AR treatment also dif-
fers from the maladaptive mind-sets of relinquished and devalued
control students in Perry et al.’s study. Relinquished control stu-
dents focused on four uncontrollable causes for poor performance
(ability, test difficulty, luck, professor quality), and devalued control
students sought to discount two controllable causes (effort, strat-
egy). In contrast, our AR treatment encouraged students to attri-
bute failure to bad strategy and to disregard poor teaching, thereby
emphasizing the controllability of performance. Unfortunately, this
form of attributional thinking can be a double-edged sword because
it puts the responsibility for success entirely on the student and re-
moves a convenient self-protective cause of failure (poor teaching).
694 PERRY ET AL.

Finally, the mind-set resulting from our AR treatment is advanta-

geous because attributing failure to bad strategy rather than lack of
effort avoids negative implications related to a person’s self-worth.
Failing because of insufficient effort incurs society’s admonition in
that laziness is anathema to being a worthy person, whereas failing
because of bad strategy has less negative implications for one’s self
worth. Taken together, our findings suggest that the AR treatment
induced an attributional mind-set that is more functional for moti-
vation than the maladaptive mind-sets spontaneously generated by
first-year students during school-to-university transitions.

Initial Academic Experiences

in New Achievement Settings

Our study focuses on the transition to new achievement settings

and shows that initial learning experiences related to a first test in-
fluence students’ emotional well-being later in the academic year.
High-performance students felt more pride and less worry than av-
erage-performance students who, in turn, exhibited this emotional
pattern more so than low-performance students. Thus, students’
initial learning experiences were uniquely related to their emotional
well-being several months later, even when prior learning experi-
ences in high school were statistically controlled.
As expected, the same initial learning experiences also related to
both specific and global performance outcomes (class tests, course
grades, GPAs) with high school grades also controlled statistically.
High-performance students did better than average-performance
students who did better than low-performance students on all
achievement measures. These achievement outcomes were con-
sistent with the emotion differences between initial performance
groups. The positive emotion pattern of high-performance students
(more pride, less worry) corresponded to better achievement, com-
pared to the more negative emotion patterns of average- and low-
performance students (less pride, more worry) and poorer achieve-
ment.
These distinct emotion patterns associated with the initial-per-
formance groups point to the importance of early academic experi-
ences for students’ emotional development in the transition to new
achievement settings. Combined with course specific and global
achievement, the emotion results indicate that initial performance
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 695

in school-to-university transitions is a significant issue, and sepa-

rate from previous high school learning experiences. Typically, re-
searchers give attention to high school grades in their analyses in
the form of independent variables or covariates, but rarely consider
the importance of initial performance.

Strengths and Limitations

The present study incorporated several conceptual, methodologi-

cal, and statistical improvements ignored in previous AR field stud-
ies. A critical conceptual improvement was to examine attributions
associated with the learning environment to redress the omission of
contextual factors in past AR studies. Causes related to instruction
(professor quality) and to assessment (test difficulty), in combina-
tion with the usual student factors, enabled a more comprehensive
analysis of complex attributional thinking than in previous AR
studies.
Another improvement was to assess AR treatment effects using
course-specific (class tests, course grade) and global performance
measures (GPA). Class tests are indicators of only one type of per-
formance at a single moment in time, and course grade is a broader
indicator consisting of multiple assessments having different for-
mats and administered over an extended period of time. Broader
still is cumulative GPA which is based on many different types of
instructors and courses having a variety of academic tasks, activi-
ties, and assessments. These varied performance measures provide
a comprehensive picture of the breadth and reliability of the AR
treatment effects involving diverse achievement outcomes.
Methodological improvements included a longitudinal design
that captured the temporal sequencing of variables, in contrast to
AR field studies that have used post-treatment only designs and
short-term performance measures. Our study employed more dis-
tal attribution, emotion, and achievement measures months after
AR administration. As well, MANCOVA and discriminant func-
tion analyses afforded a detailed representation of how multiple
attributions and emotions are combined into complex patterns.
Moreover, using high school grades as a covariate in the ANCOVAs
and MANCOVAs controlled for pre-existing differences between
students, and assigning AR treatments to experimental conditions
prior to administration reduced confounds associated with subject
696 PERRY ET AL.

selection. Finally, the comparison of the AR treatment conditions

(AR vs. No-AR) in terms of Time 2 questionnaire and GPA measures
enabled attrition biases associated with treatment conditions to be
ruled out.
Previous AR field studies failed to incorporate these conceptual,
methodological, and statistical features together in the same study.
Nevertheless, several problems with the study remain, notably our
single AR treatment provides little insight into multiple treatment
sessions. A powerful, single AR administration may be sufficient,
but several induction sessions may produce stronger treatment ef-
fects. The sparse evidence available from laboratory experiments
involving multiple AR sessions shows that little is gained beyond a
single administration (Menec et al., 1994). However, the absence of
more systematic evidence, particularly from field studies, suggests
that more research is needed in this regard.
Second, the absence of AR effects on emotion outcomes, in con-
trast to attribution outcomes, is perplexing in view of the clear
attribution-emotion linkages specified in Weiner’s theory. This at-
tribution-emotion anomaly may be better addressed in subsequent
AR studies using path analyses in which both attributions and emo-
tions are considered jointly or sequentially as mediators of AR ef-
fects on performance. Third, little can be said about the AR effects
observed in this study compared to other types of AR treatments
because few studies have systematically compared AR treatments
(cf., Haynes, Daniels, Stupnisky, Perry, & Hladkyj, 2008).
Finally, no other treatment interventions were included in our
study, such as a skills-training treatment (i.e., note-taking), nor were
filler materials used in the no-treatment conditions for comparison
purposes. In our AR condition, the treatment was separated from
the assessment of its effects by more than 4 months, and in the No-
AR condition, a questionnaire served as filler material. Laboratory
studies with strong experimental designs have used videotaped
lectures, aptitude tests, questionnaires, etc., in No-AR conditions to
counter demand characteristics, and to separate AR administration
from the assessment of its effects, given the short time durations
involved in laboratory sessions (e.g., Menec et al., 1994; Perry &
Penner, 1990).
In longitudinal AR field studies, some have used filler material,
but others have not in view of the significant amount of time that
typically elapses between the treatment and the assessment of the
dependent variables, in contrast to the short time period between
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 697

these events in laboratory studies. However, combined with strong

experimental studies conducted in laboratory settings, our quasi-
experimental design provided consistent results in support of AR
treatments that, nonetheless, could be enhanced in future AR field
studies using true experimental designs.

Conclusions

Contextual factors such as instruction methods and testing tech-

niques contribute substantially to the learning process in achieve-
ment settings, but have been ignored in AR field studies and only
briefly examined in laboratory research. However, unfamiliar con-
textual factors in new achievement settings can contribute to forms
of dysfunctional attribution thinking that can undermine motivation
and performance, as seen in the high failure rates in the first year
of university. Our study contributes uniquely to existing research
by pointing to the benefits of AR as a viable motivation treatment
in new achievement settings. Our AR treatment fostered adaptive
attribution thinking and sizable performance gains across both do-
main specific and general achievement measures. By participating
in a brief AR treatment, low- and average-performance students
improved their achievement on a class test by roughly 1½ letter
grades, and their course percentages and overall GPAs by almost
one letter grade. These findings represent ecologically meaningful
performance gains, and highlight the significance of introducing an
attribution change treatment early in the transition to new achieve-
ment settings.

References
Abramson, L. Y., Garber, J., & Seligman, M. (1978). Learned helplessness in humans:
Critique and reformulation. Journal of Abnormal Psychology, 87, 49-74.
Abramson, L. Y., Seligman, M. E. P., & Teasdale, J. D. (1978). Learned helplessness
in humans: Critique and reformulation. Journal of Abnormal Psychology, 87,
49-74.
Andrews, G. R., & Debus, R. L. (1978). Persistence and the causal perception of
failure: Modifying cognitive attributions. Journal of Educational Psychol-
ogy,70,154-166.
Bandura, A. (1997). Self-efficacy: The exercise of control. New York: W. H. Free-
man.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences. Hillsdale, NJ; Er-
lbaum.
698 PERRY ET AL.

Covington, M. V. (1997). A motivational analysis of academic life in college. In Perry,

R. P., and Smart, J. C. (Eds.) Effective teaching in higher education: Research and
practice (pp. 61-100). Netherlands: Springer.
Dweck, C. S. (1999). Self-theories: Their role in motivation, personality, and development.
New York: Psychology Press.
Fredrickson, B. L. (2001). The role of positive emotions in positive psychology: The
broaden-and-build theory of positive emotions. American Psychologist, 56,
218-226.
Glass, D. C., & Singer, J. (1971). Urban stress. New York: Academic Press.
Hall, N. C., Hladkyj, S., Perry, R. P., & Ruthig, J. C. (2004). The role of attributional
retraining and elaborative learning in college students’ academic develop-
ment. Journal of Social Psychology, 144, 591-612.
Hall, N. C., Perry, R. P., Chipperfield, J. G., Clifton, R. A., & Haynes, T. L. (2006).
Enhancing primary and secondary control in achievement settings through
writing-based attributional retraining. Journal of Social and Clinical Psychology,
25, 361-391.
Hall, N. C., Perry, R. P., Goetz, T., Ruthig, J. C., Stupnisky, R. H., & Newall, N. E.
(2007). Attributional retraining and elaborative learning: Improving academ-
ic development through writing-based interventions. Learning and Individual
Differences, 17, 280-290.
Haynes, T. L., Perry, R. P., Stupnisky, R. H., & Daniels, L., M. (2009). A review of
Attributional Retraining treatments: Fostering engagement and persistence
in vulnerable college students. In J. Smart (Ed.), Higher education: Handbook of
theory and research (Vol. 24, pp. 229-275). The Netherlands: Springer Publish-
ers.
Haynes, T. L., Daniels, L. M., Stupnisky, R. H., Perry, R. P., & Hladkyj, S. (2008). The
effect of Attributional Retraining on mastery and performance motivation
among first-year college students. Basic and Applied Social Psychology, 30(3),
198-207.
Haynes, T. L., Ruthig, J. C., Perry, R. P., Stupnisky, R. H., & Hall, N. C. (2006). Reduc-
ing the academic risks of over-optimism: The longitudinal effects of attribu-
tional retraining on cognition and achievement. Research in Higher Education,
47, 755-779.
Heider, F. (1958). The psychology of interpersonal relations. New York: Wiley.
Heller, K. A., & Ziegler, A. (1996). Gender differences in mathematics and the natu-
ral sciences: Can attributional retraining improve the performance of gifted
females? Gifted Child Quarterly, 40, 200-210.
Keppel, G., & Wickens, T. (2004). Design and analysis: A researcher’s handbook (4th ed.).
Upper Saddle River, NJ: Pearson Education Inc.
Marascuilo, L. A., & Levin, J. R. (1982). Multivariate statistics in the social sciences: A
researcher’s guide. Monterey, CA: Brooks/Cole.
Menec, V. H., Perry, R. P., Struthers, C. W., Schonwetter, D. J., Hechter, F. J., & Eich-
holz, B. L. (1994). Assisting at-risk college students with attributional retrain-
ing and effective teaching. Journal of Applied Social Psychology, 24, 675-701.
Pekrun, R. (2000). Social cognitive, control-value theory of achievement emotions.
In J. Heckhausen (Ed.), Motivational psychology of human development. Oxford,
UK: Elsevier Science.
Pekrun, R., Frenzel, A., Goetz, T., & Perry, R. P. (2007). The control-value theory of
achievement emotions: An integrative approach to emotions in education. In
ATTRIBUTIONAL RETRAINING AND ACADEMIC PERFORMANCE 699

P. Schutz & R. Pekrun (Eds.), Emotions in education. San Diego, CA: Academic
Press.
Pekrun, R., Goetz, T., Titz, W., & Perry, R. P. (2002). Academic emotions in students’
self-regulated learning and achievement: A program of qualitative and quan-
titative research. Educational Psychologist, 37, 91-106.
Perry, R. P. (1991). Perceived control in college students: Implications for instruction
in higher education. In J. Smart (Ed.), Higher education: Handbook of theory and
research (Vol. 7, pp. 1-56). New York: Agathon Press.
Perry, R P. (2003). Perceived (academic) control and causal thinking in achievement
settings: Markers and mediators. Canadian Psychologist, 44(4), 312-331.
Perry, R. P., & Hall, N. C. (2009). Attributional retraining. In E. M. Anderman & L. H.
Anderman (Eds.), Psychology of classroom learning: An encyclopedia. Farming-
ton Hills, MI: Thomson Gale.
Perry, R. P., Hall, N. C., & Ruthig, J. C. (2005). Perceived (academic) control and
scholastic attainment in higher education. In J. C. Smart (Ed.), Higher educa-
tion: Handbook of theory and research (Vol. 20, pp. 363-436), The Netherlands:
Springer.
Perry, R. P., Hechter, F. J., Menec, V. H., & Weinberg, L. E. (1993). Enhancing achieve-
ment motivation and performance in college students: An attributional re-
training perspective. Research in Higher Education, 34, 687-723.
Perry, R. P., Hladkyj, S., Pekrun, R. H., Clifton, R. A., & Chipperfield, J. G. (2005).
Perceived academic control and failure in college students: A three-year study
of scholastic attainment. Research in Higher Education, 46, 535-569.
Perry, R. P., Hladkyj, S., Pekrun, R. H., & Pelletier, S. T. (2001). Academic control and
action control in the achievement of college students: A longitudinal field
study. Journal of Educational Psychology, 93, 776-789.
Perry, R. P., & Penner, K. S. (1990). Enhancing academic achievement in college stu-
dents through attributional retraining and instruction. Journal of Educational
Psychology, 82, 262-271.
Perry, R. P., & Struthers, C. W. (1994, April). Attributional retraining in the college class-
room: Some causes for optimism. Paper presented at the American Educational
Research Association annual meeting, New Orleans, LA.
Perry, R. P., Stupnisky, R. H., Daniels, L. M., & Haynes, T. L. (2008). Attributional
(explanatory) thinking about failure in new achievement settings. Special Is-
sue: Attribution and academic judgments: Two sociocognitive approaches. European
Journal of Psychology of Education, 23, 459-475.
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton
University Press.
Ruthig, J. C., Perry, R. P., Hall, N. C., & Hladkyj, S. (2004). Optimism and attribu-
tional retaining: Longitudinal effects on academic achievement, test anxiety,
and voluntary withdrawal in college students. Journal of Applied Social Psy-
chology, 34, 709-730.
Schulz, R. (1976). The effects of control and predictability on the physical and psy-
chological well being of the institutionalized aged. Journal of Personality and
Social Psychology, 33, 563-573.
Seligman, M. E. P. (1975). Helplessness: On depression, development, and death. San
Francisco, CA: Freeman.
Skinner, E. A. (1996). A guide to constructs of control. Journal of Personality and Social
Psychology, 71, 549-570.
700 PERRY ET AL.

Stevens, J. P. (2002). Applied multivariate statistics for the social sciences (4th ed.). Mah-
wah, NJ: Lawrence Erlbaum.
Struthers, C. W., & Perry, R. P. (1996). Attributional style, attributional retraining,
and inoculation against motivational deficits. Social Psychology of Education,
1, 171-187.
Tabachnick, B. G., & Fidell, L. S. (2007). Using multivariate statistics (5th ed.). Boston:
Allyn and Bacon.
Van Overwalle, F. (1989). Structure of freshmen’s causal attributions for exam per-
formance. Journal of Educational Psychology, 81, 400-407.
Van Overwalle, F., & De Metsenaere, M. (1990). The effects of attribution-based in-
tervention and study strategy training on academic achievement in college
freshmen. British Journal of Educational Psychology, 60, 299-311.
Weiner, B. (1972). Attribution theory, achievement motivation, and the education
process. Review of Educational Research, 42, 203-215.
Weiner, B. (1979). A theory of motivation for some classroom experiences. Journal of
Educational Psychology, 71, 3-25.
Weiner, B. (1983). Some methodological pitfalls in attributional research. Journal of
Educational Psychology, 75, 530-543.
Weiner, B. (1985). An attributional theory of achievement motivation and emotion.
Psychological Review, 92, 548-573.
Weiner, B. (1995). Judgements of responsibility: A foundation for a theory of social conduct.
New York: Guilford Press.
Weiner, B. (2000). Intrapersonal and interpersonal theories of motivation from an
attributional perspective. Educational Psychology Review, 12, 1-14.
Weiner, B. (2006). Social motivation, justice, and the moral emotions: An attributional ap-
proach. Mahwah, NJ: Lawrence Erlbaum Associates.
Wilson, T. D., Damiani, M., & Shelton, N. (2002). Improving the academic perfor-
mance of college students with brief attributional interventions. In Aronson,
J. (Ed), Improving academic achievement: Impact of psychological factors on educa-
tion (pp. 89-108), San Diego, CA: Academic Press.
Wilson, D. W., & Linville, P. W. (1982). Improving the academic performance of col-
lege freshmen: Attribution therapy revisited. Journal of Personality and Social
Psychology, 42, 367-376.
Wortman, C. B., & Brehm, J. W. (1975). Responses to uncontrollable outcomes: An
integration of reactance theory and the learned helplessness model. In L.
Berkowitz (Ed.), Advances in experimental social psychology (Vol. 8, pp. 277-336).
New York: Academic Press.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.