First publ. in: The Journal of Experimental Education ; 75 (2006), 1. - pp.

5–29
DOI 10.3200/JEXE.75.1.5-29

The Domain Specificity of Academic

Emotional Experiences

THOMAS GOETZ
ANNE C. FRENZEL
REINHARD PEKRUN
University of Munich

NATHAN C. HALL
University of California-Irvine

ABSTRACT. The authors analyzed the domain specificity of emotions and focused on
experiences of enjoyment, anxiety, and boredom in the domains of mathematics, Latin,
German, and English. On the basis of assumptions in R. Pekrun's (2000; in press) con·
trol·value theory and findings of pilot studies, the authors hypothesized the existence
of a largely domain-specific organization of emotional experiences. The sample con-
sisted of 721 students from grades 7 to 10. Confirmatory multitrait-multimethod fac-
tor analysis of the 2-faceted dataset (emotions and domains) corroborated assumptions
of domain specificity. Furthermore, using multilevel analysis, the authors found that
emotions were significantly more domain-specific than students' grades, with enjoy-
ment being the most domain-specific of the three emotions under investigation. The
authors discuss implications for future research and practice.
Key words: achievement, anxiety, boredom, domain specificity, emotions, enjoyment

IS IT POSSIBLE TO INFER FROM STUDENTS' ENJOYMENT in mathemat-

ics classes that they will also enjoy other classes, such as science or English? The
answer to this question is relevant from not only theoretical but also research-
and practice-oriented perspectives, and not solely for enjoyment but for every
emotional experience. For example, is representing the construct of test an~iety
(Zeidner, 1998) as a highly generalized construct empirically justifiable? That is,

Address correspondence to: Thomas Goetz. Department of Psychology. University of

Munich. Leopoldstl: 13. D-80802 Munich. Germany. E-mail: [email protected]
muenchen.de

Konstanzer Online-Publikations-System (KOPS)

URL: http://nbn-resolving.de/urn:nbn:de:bsz:352-137535
6

does a student show a similar level of test anxiety across all academic domains,
or does the intensity of the students' anxiety prove to be domain-specific, mak-
ing it more appropriate to speak of domain-specific test anxieties such as math-
ematics test anxiety or English test anxiety? Does a single, domain-specific as-
sessment of emotions suffice as a method of acquiring knowledge on emotional
experiences across an entire array of domains? Can a mathematics teacher who
notices that a student is often bored in math classes conclude that the student will
also be bored in other domains?
These questions represent a critical area of investigation for theory and re-
search on academic emotions. Although research examining emotional experi -
ences related to learning and achievement is lacking, in recent years there has
been a considerable increase in theoretical and empirical contributions to this
field. This increased interest is reflected in two recent special issues on emotions
and learning in the Educational Psychologist (Schutz & Lanehart, 2002) and in
Learning and Instruction (Efklides & Volet, 2005). Despite this proliferation of
research on academic emotions, few researchers have focused on the domain
specificity of students' emotional experiences. Although this topic has already
been investigated with respect to numerous constructs other than emotions (e.g.,
academic self-concept, achievement value; Bong, 2001), little empirical work ex-
ists on the domain specificity of academic emotions.

Theoretical Assumptions

One theoretical assumption consistently demonstrated by empirical research is

that a number of psychosocial constructs other than emotions are organized in do-
main-specific ways. In particular, findings on the domain-specific organization of
academic self-concepts are relatively clearcut (Marsh, 1993; Marsh & Yeung,
1996). Many of those studies are based on Marsh's Internal/External Frame of
Reference model ([lIE]; Marsh, 1986). In the lIE model, it is assumed that there
are relatively strong relations between achievement scores in different domains
(e.g., mathematics and languages), but that subjective beliefs and intraindividual
comparison processes, regarding the domain-specific organization of individual
abilities, lead to weak relations between self-concepts of abilities in these domains
(e.g., "IfI get a good grade in math, then I can't be especially good at German").
To test the assumptions of the lIE model, researchers investigate at least two
different domains in a given study. More specifically, mathematics and language-
oriented domains are the focus of most of these studies. Moller and Koller (2004)
reported results from 34 studies on the lIE model and found a weak mean corre-
lation (d = .10, range = -.13 to .22) between mathematic and verbal academic self-
concepts. Besides academic self-concept, other variables have also been investi-
gated with respect to their domain-specific organization, including self-efficacy,
causal attributions, task values, and achievement goals (Abu-Hilal & Bahri, 2000;
 7

Bong, 1998; Bong & Skaalvik, 2003; Marsh, 1984, 1986; Moller & Koller, 200 I;
Stevenson & Newman, 1986).
It is reasonable to assume that the mechanisms underlying the domain speci-
ficity of such constructs as self-concepts, task values, and achievement goals are
also at work for emotional experiences. Therefore, emotional experiences might
be organized in a largely domain-specific manner. That is, the mechanisms ad-
dressed by Marsh's (1986) liE model could operate for students' emotions as
well. More specifically, Pekrun's (2000; in press) control-value theory of acade-
mic emotions represents a potentially valuable conceptual framework in which to
examine the assumptions of the lIE model and apply available evidence on the
domain specificity of motivational constructs to the field of emotions. Integrat-
ing assumptions from expectancy value (Pekrun , 1992; Turner & Schallert, 200 I)
and attributional (Weiner, 1985) approaches to achievement-related emotions,
Pekrun postulates that control and value appraisals relating to academic learning
and achievement are of primary importance for students' emotions (Pekrun, 2000
in press; Pekrun, Goetz, Titz, & Perry,. 2002a). Control-related appraisals in-
clude self-directed beliefs underlying the subjective controllability of learning
and its outcomes, with self-concepts of abilities being one type of relevant belief.
Value appraisals include subjective task values and are closely related to stu-
dents' achievement goals. Considering self-concepts, task values, and achieve-
ment goals are known to be organized in a largely domain-specific manner,
Pekrun's control-value theory implies that resulting emotions should also be or-
ganized in domain-specific ways.

Empirical Studies

To date, empirical research on the domain specificity of students' emotional

experiences is lacking. This lack is particularly true for the domain specificity of
academic emotions other than anxiety. Within research on academic anxiety, two
studies have explicitly addressed the domain specificity of this emotion (Marsh
1988a; Marsh & Yeung, 1996). In addition, we found three studies that provide
indirect evidence in support of the domain specificity of academic anxiety, al-
though domain specificity was not explicitly stated as a prim~ry research ques-
tion (Everson, Tobias, Hartman, & Gourgey, 1993; Gottfried, 1982; Hembree,
1990). With the exception of this research on academic anxiety, little empirical
work on the domain-specific nature of academic emotions has been conducted.
Thus, two preliminary investigations that more specifically addressed the domain
specificity of other academic emotional experiences; these form the empirical
framework on which the present study is based (Goetz, 2004; Pekrun, Goetz,
Titz, & Perry, 2002b).
Concerning studies on the domain specificity of anxiety, Marsh (1988a) con-
centrated his investigation on experiences of anxiety during instruction and
8

found a disattenuated correlation of rd = .04 between anxiety in mathematics and

English lessons (four items each, parallelized scales; N = 14,825, grade 10;
analysis of data from the High School and Beyond [HSB] Study; National Cen-
ter for Educational Statistics [NCES], 1986). Marsh and Yeung (1996) examined
anxiety in four academic domains (mathematics, sciences, social studies, and
English) by analyzing data from the National Educational Longitudinal Survey
of 1988 (NELS88 database, Ingles et aI., 1992; N = 24,599 eighth graders). Using
confirmatory factor analyses, the authors demonstrated that academic anxiety in
school-aged children is organized in a domain-specific way. They found very
weak intercorrelations between domain-specific anxiety ratings, with the
strongest relationships found between anxiety in conceptually similar academic
domains (e.g., mathematics and science). Moreover, anxiety showed a greater de-
gree of domain specificity than academic achievement in these domains (with
grades as well as standardized test scores).
Everson et al. (1993) investigated the relations between subjective difficulty of
a particular subject and test anxiety in that domain among first-year college stu-
dents (average age = 21 years). They found highly distinctive individual levels of
test anxiety in English, mathematics, physics, and social studies. Analyzing the
relationship between intrinsic motivation and anxiety, Gottfried (1982) investi-
gated similar variables among 141 fourth and seventh graders in four academic
domains (reading, mathematics, social studies, and sciences). Significant nega-
tive correlations were found between anxiety and intrinsic motivation within do-
mains (e.g., intrinsic motivation and anxiety involving mathematics) but not be-
tween domains (e.g., intrinsic motivation in mathematics and anxiety related to
reading). In a meta-analysis of 51 studies using the Mathematics Anxiety Rating
Scale ([MARS]; Richardson & Suinn, 1972), Hembree (1990) reported a mean
correlation of r = -.06 between mathematics anxiety and verbal performance, as
well as a correlation coefficient of r = -.34 between the MARS scores and math-
ematical performance. This finding points to the domain specificity of anxiety
based on the premise that, if anxiety were organized in a domain-transcending
manner, then higher MARS scores would predict more anxiety in verbal do-
mains, which would translate into significantly poorer verbal performance than
evidenced by Hembree's nonsignificant correlation.
Nevertheless, all of these studies pertained to anxiety. We conducted two pilot
investigations to also obtain preliminary evidence concerning the domain speci-
ficity of academic emotions other than anxiety. Pekrun and colleagues (2002b)
used Pekrun's (2000; in press) control-value theory of emotions to derive as-
sumptions involving the relative domain specificity of students' academic emo-
tions. Between-domain correlations in a sample of 697 middle school students
were nonsignificant and near zero for academic enjoyment in mathematics, lan-
guages (German and English), music, and sports, and low for anxiety (.13 < r <
.26). Goetz (2004; N = 699, Grades 7 and 9) also found very weak correlations be-
 9

tween enjoyment in the domains of mathematics, German, English, music, and

physical education. The corresponding correlations for anxiety were somewhat
stronger. Taken together, the results of these two studies indicate that academic en-
joyment and anxiety may be conceptually organized in a largely domain-specific
manner, yet they may vary in their degree of domain specificity.
In sum, the available evidence suggests a domain-specific organization of stu-
dents' anxiety. Empirical evidence on the domain-specific organization of emotions
other than anxiety, however, is still largely lacking. Therefore, in the present study,
we further explored the domain-specific nature of students' academic emotional
experiences and more specifically, examined the relative degree of domain speci-
ficity in different academic emotions.

Research Questions and Hypotheses

We aimed to examine whether students' emotional experiences-namely ex-

periences of academic enjoyment, anxiety, and boredom-are organized in a do-
main-specific manner. Because theoretical assumptions and the findings of our
pilot studies suggest a domain-specific structure of academic emotions, we also
anticipated a domain-specific organization of emotional experiences in the pre-
sent study.
In addition to the analysis of the internal structures of emotional experiences
(within-network), we wanted to investigate the external relations (between-
network) of academic enjoyment, anxiety, and boredom. We hypothesized sub-
stantial relations between emotional experiences and achievement in similar
school subjects (e.g., English and German; convergent validity of emotion mea-
sures) and a relative lack of correlations between scores in dissimilar school sub-
jects (e.g., mathematics and English) supporting the divergent validity of the
emotions measures.
Furthermore, we also examined the extent to which enjoyment, anxiety, and
boredom differ in their respective degrees of domain specificity. Based on the
findings of the pilot research (Goetz, 2004; Pekrun et aI., 2002b), we hypothe-
sized that enjoyment would show a greater degree of domain specificity than
with anxiety. Because there is a lack of research findings on the extent to which
boredom is experienced as domain specific, the current study was exploratory in
nature with respect to this specific academic emotion.

Method

Sample and Data Collection

Our sample consisted of 721 students (54% female) from 32 classes within
five schools (7th grade: n = 224; 8th grade: n = 154; 9th grade: n = 187; 10th
 10

grade: n = 156; schools 1-5: n = 189,217,92, 138, and 85, respectively). Data
collection lasted 45 min per class and was conducted by trained testing person-
nel using fully standardized student questionnaires. The average age of the par-
ticipants was 14.42 years (SD = 1.29); no significant gender difference existed,
t(719) = .99, p = .32. Participation in the study was voluntary and required
parental consent.

Selection of Emotions and Academic Domains

We used two selection criteria in choosing which emotions to assess. First, we

wanted to assess emotions that were conceptually distinct. Numerous models for
categorizing emotions based on underlying dimensions, such as activation, va-
lence, intensity, and duration are available. We based our selection of emotions
on Watson and Tellegen's (1985) circumplex model, which categorizes emotions
according to aspects relevant in the context of learning and achievement, name-
ly activation and valence. Second, we searched in the research literature for emo-
tions that are particularly important in academic settings (see Goetz, 2004;
Pekrun et aI., 2002a). As a result of these two selection strategies, we chose to
assess the following three emotions that have been found to be salient in an aca-
demic environment and could be distinguished based on activation and valence:
enjoyment (positive and activating), anxiety (negative and activating), and bore-
dom (negative and deactivating). With respect to the academic domains in which
each of the three emotions were assessed, we selected major school subjects
taught in all five of the schools and in each grade level: mathematics, Latin, Ger-
man, and English.

Measurement of Discrete Emotions in Various Academic Domains

Single items developed within the Project for the Analysis of Learning and
Achievement in Mathematics study (PALMA; Goetz, 2004; Pekrun, Goetz, &
Frenzel, 2006) were used to measure enjoyment, anxiety, and boredom in the do-
mains of mathematics, Latin, German, and English (12 items in total). The four
enjoyment items were "How much do you enjoy [domain]?" The four anxiety
items were "How much anxiety do you experience in [domain]?" The four bore-
dom items were "How much do you feel bored in [domain]?" Response format
consisted of a 5-point Likert scale ranging from I (not at all) to 5 (very much).
The Appendix shows the means and standard deviations as well as intercorrela-
tions of all single emotion items.
We also administered three Latin-related multi-item emotion scales assessing
enjoyment, anxiety, and boredom in this study to test the validity of the single-
item measures. The three multi-item scales were based on the PALMA scales and
modified to pertain to the Latin domain. Response format for these three scales
 II

consisted of a 5-point Likert scale ranging from I (strongly disagree) to 5

(strongly agree). The scales showed the following statistical characteristics: (X =
.88, M = 24.97, SD = 7.23 for enjoyment (nine items); (X = .91, M = 35.40, SD =
11.95 for anxiety (15 items); and (X = .91, M = 15.15, SD =6.42 for boredom (six
items). Using linear structural equation modeling (USREL 8.53; Joreskog &
Sorbom, 2002), we calculated the sizes of effects from the three Latin-related
single-item scales (enjoyment, anxiety, and boredom in Latin) to the corre-
sponding Latin-specific latent variables (also enjoyment, anxiety, and boredom
in Latin). All three path coefficients from the Latin-related single-item scales
(manifest variables) to the corresponding multi-item scales (latent variables)
were relatively high (.86 for enjoyment, .66 for anxiety, and .80 for boredom).
Overall, these high path coefficients indicate that the measurement of emotions
using single-item measures in the current study was sufficiently valid.

Achievement

We obtained achievement data for mathematics, Latin, German, and English

from students' self-reported final grades for the preceding school year. In the
German school system, grades vary between I (very good) and 6 (insufficient).
Given that high values represent poor achievement outcomes is counterintuitive,
we inverted the grades measure in subsequent analyses. Thus, high values on the
inverted grades variable represent good achievement outcomes. Concerning the
validity of students' self-reported final grades, a recent study by Dickhaeuser and
Plenter (2005) showed very strong correlations (r = .90) between self-reported
and actual academic performance using a German sample of 866 students in
grades 7 and 8. Furthermore, Dickhaeuser and PI enter also showed that the ac-
curacy of the self-reported performance did not depend on students' gender or
achievement level. Thus, one can assume that self-reported final grades assessed
in the present study were sufficiently accurate indicators of students' actual
grades. Means, standard deviations, and intercorrelations of the achievement
scores (inverted grades) are presented in the Appendix.

Data Analyses-Domain Specificity of Emotional Experiences

The set of 12 items reflects an underlying two-facet design in which the emo-
tion facet has three levels (enjoyment, anxiety, boredom) and the school subject
facet has four levels (mathematics, Latin, German, English). As such, the dual-
faceted emotion data collected in the present study can be viewed as multi-
trait- multimethod (MTMM) data, and confirmatory factor analysis (CFA) can be
used to address our research question (Byrne, 1998; Marsh, 1988b; Widaman,
1985). A central advantage of analyzing MTMM data with CFA is that method
effects can be taken into account when analyzing the traits. Method effects are
 12

often regarded as unwanted disruptive factors to be controlled for (Marsh &

Yeung, 1996); however, they can in fact reveal important information concerning
the structure of the data (Marsh, Byrne, & Craven, 1993). Because structural as-
pects played a pivotal role in the emotion measures employed in the current
study, we analyzed our data in line with the analysis of the MTMM data.
To this end, we adapted an approach suggested by Marsh and Yeung (1996), in
which linear structural equation modeling (USREL 8.53; Joreskog & Sorbom,
2002) is used for testing a set of CFA-MTMM models competitively (see also
Marsh, 1989). We tested a total of five models (A to E) depicted in Figure I . In
Model A, one latent general factor is defined. Because all 12 items of the two-
facet design (3 emotion x 4 domains) refer to one single factor, we termed this
general factor Domain-Transcending Emotionality. In Model B, we defined three
emotion-specific latent factors (experiences of enjoyment, anxiety, and boredom)
with each domain-specific item for each emotion predicted by the latent factor for
that emotion (e.g., all four enjoyment items were predicted by the latent enjoy-
ment factor). We labeled each of the three factors of Model B a Domain-General
Discrete Emotion representing the total of experiences of each of the three emo-
tions across mathematics, Latin, German, and English. Finally, Model C included
four domain-specific latent emotion factors (emotional experiences in mathemat-
ics, Latin, German, and English). That is, each of the three domain-specific items
(e.g., enjoyment in mathematics, anxiety in mathematics, boredom in mathemat-
ics) was predicted by the latent factor representing the given domain, (e.g., math-
ematics). We labeled the four factors of Model C Domain-Specific Emotionality
that is, the total of students' experiences of enjoyment, anxiety, and boredom in
each of the four domains.
We did not assess a model in which emotions and domains were defined as
traits simultaneously because previous researchers have indicated that such mod-
els typically result in unstable solutions (Marsh, Byrne, & Craven, 1993). How-
ever, several researchers have shown that correlated uniqueness (CU, a combina-
tion of specific and error variance) models are suitable for this kind of analysis
(Byrne, 1998; Marsh & Grayson, 1995).One can assume that CU-MTMM mod-
els better fit the data than do models where method effects are not considered
(Byrne; Marsh, Byrne, & Craven). In both Models Band C, only one facet of the
two-faceted data structure was taken into account in assessing the corresponding
latent factors (Model B: Discrete Emotion factor; Model C: Domain factor). Con-
sequently, both models could be modeled as CU-MTMM models, in which the
other complementary factor would be accounted for as a method effect. However,
both CU Models (CU-B, CU-C) had a similar structure that is, the effects of the
emotion and domain factors were considered simultaneously in each model.
Because Model C was most applicable to our present research question con-
cerning the strength of relationships between the emotional experiences in the
different subject areas (i.e., the degree of domain specificity), we revised this
 13

model to include correlations between the discrete emotions in the different sub-
ject areas to control for correlated uniqueness among the similar emotion mea-
sures. Once again, we were more interested in the strength of relations involving
the domain specificity of academic emotions (Model C) than the domain-gener-
al nature of discrete emotional experiences (Model B).
For building the CU-MTMM Model D, we defined the four domain-specific
factors of Model C as Traits (domains = traits) and considered the emotion-spe-
cific item components method effects. Thus, Model D is a modified version of
Model C and represents a CU-MTMM model with CU between items referring
to the same emotional experience (18 CUs in total: 4 items per emotion and, con-
sequently, 6 CUs per emotion). That is, CUs between items with identical emo-
tion-related item components were permitted in Model D (4 x "How much do
you enjoy [domain]?;" 4 x "How much anxiety do you experience in [domain]?;"
4 x "How much are you bored in [domain] ?"). In terms of the MTMM approach,
no method factors were specified per se, but we considered their potential.
We integrated achievement outcomes in the four subject areas into the analy-
sis as external criteria because, in addition to the internal structure (within-net-
work) of Model D, the external validity (between-network) of the latent factors
of this Model was to be tested. We generated Model E for that purpose and it rep-
resented an extension of Model D (see Figure I). In this model, we integrated an
achievement indicator (i.e., grades) for each of the four domains as a manifest
variable in addition to the four domain-specific factors of Model D, with the error
variance of each achievement score fixed to zero.
Testing was based on a comparison of the chi-square statistics and selected fit
indices (Nonnormed Fit Index-NNFI, Tucker-Lewis Index-TLI; Tucker &
Lewis, 1973; Comparative Fit Index-CFI, Bentler, 1990; root mean square error
of approximation [RMSEA] Steiger & Lind, 1980). In line with Marsh and
Yeung's (1996) methodological approach and in view of our large sample, the
chi-square test was not integrated in our analysis as this test has been shown to
be overly sensitive to sample size (e.g., Marsh, Balla, & McDonald, 1988).
Both the NNFI and the CFI are incremental indices with goodness of fit scaled
from 0 to 1. The zero point represents a null model in which all the measured
variables are assumed to be uncorrelated and the 1.0 point reflects a perfect fit.
NNFIs and CFIs greater than .90 indicate an acceptable model fit, but for our pur-
poses, a comparison of the fit indices for competing models was also important.
CFI and NNFI differ from each other in that the CFI is insensitive to model com-
plexity, whereas the NNFI corrects the fit according to the model complexity
with more complex models typically resulting in poorer fit levels (for a more
exact description of NNFI and CFI see Byrne, 1998). The RMSEA takes the error
of approximation in the population into account and is also sensitive to model
complexity. RMSEAs less than .06 indicate good fit, and values as high as .08
represent reasonable error of approximation in the population.
 +>-

B.

ee
 D.

E.

~
To [§J
~o
."
FIGURE 1. Structural model for the testing of domain specificity of emotional experiences. The 12 items measure the experi-
ence of En: enjoyment, An: anxiety, Bo: boredom in the domains of M: mathematics, L: Latin, G: German, E: English. Model
A: General factor model (Domain-Transcending Emotionality). Model B: three emotion-specific factors (Domain-General
Discrete Emotions). Model C: four domain-specific factors (Domain-Specific Emotionality). Model D: four domain-specific
factors with correlated uniquenesses of emotion ratings (Correlated Uniqueness Model [CU-Model]; that is, assumption of
domain-specific emotionality with consideration of the effects of the emotion-specific factors). Model E: identically equal to
Model D (CU-Model), but achievement scores were integrated additionally as manifest variables.
Ul
 Iii

Data Analyses-Domain Specificity of Discrete Emotions

Because the confirmatory factor analyses provided no information concerning

the extent to which enjoyment, anxiety, and boredom may vary in their domain-
specific organization, we conducted a multilevel analysis procedure to answer
that question (hierarchical linear modeling [HLM 6.01]; Raudenbush, Bryk,
Cheong, & Congdon, 2001). To examine the extent of domain specificity in the
three emotions under investigation in relation to the external criterion we includ-
ed achievement outcomes in each of the four domains in our analyses. First, a
Level I dataset was created with four variables, namely enjoyment, anxiety,
boredom, and performance. One row in this dataset represented the scores for
these four variables for I participant in one domain. Thus, every participant had
four rows of data with each row corresponding to one of the four domains as-
sessed (mathematics, Latin, German, and English). The level I file thus included
a total of 2,884 level I rows (721 [sample size] x 4 [domains)).
To generate the level 2 dataset, we aggregated the data for each participant
across the four domains for each of the four variables. For example, for enjoy-
ment, we calculated the mean value of enjoyment in mathematics, Latin, Ger-
man, and English for each participant. We applied the same aggregation proce-
dure for anxiety, boredom, and performance. The level 2 file thus consisted of
721 units (sample size). The multilevel analytical approach allowed us to model
the level I and level 2 datasets simultaneously (Raudenbush & Bryk, 2002).
Hence, the total variance for enjoyment, anxiety, and boredom, as well as for the
achievement scores, can each be split into intersubject variance (interindividual
variances on each of the mean scores for enjoyment, anxiety, boredom, and per-
formance) and intrasubject variance (intraindividual variance for enjoyment,
anxiety, boredom, and achievement across the four domains):
(1)

For enjoyment, for example, Var(Yij)En represents the total variance of enjoy-
ment across the four domains i) to i4 (i) = mathematics, i2 = Latin, i3 = German ,
and i4 = English) and all 721 students}) to j721 (N = 721). This total variance of
the independent variables (Yij) can be split into an intrasubject variance (Var(rij)
= ( 2) and an intersubject variance (Var(uOj) ='too). We can calculate the intraclass
correlation (p) using these two proportions of variance. The intraclass correlation
reflects the proportion of intersubject variance 'too relative to the total variance
Var(Yij) given Var(Yij) = 'too + 0 2 :
(2)

Thus, the intraclass correlation can be interpreted as an indicator of the extent

of domain specificity in each of the constructs assessed: The smaller the intra-
class correlation, the higher the proportion of the intrasubject variance in the total
 17

variance, and, consequently, the more domain-specifically the construct is orga-

nized. Conversely, a high intraclass correlation would show that the total vari-
ance of the emotional experience might be primarily the result of differences be-
tween individuals (a large proportion of intersubject variance relative to the total
variance), which would suggest a domain-general organization of emotional ex-
periences. The present approach makes it possible to directly compare the intra-
class correlations of enjoyment, anxiety, boredom, and performance and thus the
extent of domain-specific organization for each of these constructs. To see
whether intraclass correlations of our four constructs under investigation differed
significantly from each other, we calculated 99% confidence intervals for each
construct. We calculated these confidence intervals using a formula suggested by
Fisher (1970), that was recently discussed by Ukoumunne (2002), as an adequate
way of calculating confidence intervals for intraclass correlation coefficients in
two-level hierarchical models.

Results

Domain Specificity of Emotional Experiences

The chi-square statistics and fit indices (NNFI, CFI, RMSEA) showed unac-
ceptable model fit for both the general factor model (Model A) and the two mod-
els that assume emotion- or domain-specific factors (Models Band C; Table 1).
Nevertheless, Model C showed a substantially better fit than did Model B, which
in turn showed a slightly better fit than did Model A. The better fit of Model C
compared with Model B indicates that differentiating the two-faceted data struc-
ture along domains better fits the data than does differentiating it along emotions.
Model D showed a very good fit to the data (see Table 1). This means that, when
taking method effects into account, it makes sense to assess students' emotional
experiences based on four domain-specific latent factors.
Model E was an extended version of Model D in which, in addition to the four
domain-specific factors, the four achievement scores were also included in the
analyses (CU-MTMM model + achievement). The statistical characteristics of
this model were acceptable (see Table I). Table 2 shows the intercorrelations of
the domain-specific factors, the intercorrelations of the achievement scores, as
well as the correlations of the domain-specific factors with the achievement
scores. Intercorrelations between the domain-specific factors (Table 2, upper left
section) were generally very weak but were most strongly pronounced between
English and German (.33), followed by Latin and German (.25), and Latin and
mathematics (.20). The intercorrelations between the four domain-specific fac-
tors of Model C are not presented because of poor model fit. However, the inter-
correlations of the domain-specific factors of Model C showed the same pattern,
with correlation values being slightly higher than those in Model E (four domain-
 18

TABLE 1. Multitrait-multimethod Models: Results of Confirmatory Factor

Analyses

Models X2 df NNFI CFI RMSEA

(A) General factor 1629.36 54 . 11 .27 .242

(8) 3 emotion-specific
factors" 1434.79 51 .20 .38 .233
(C) 4 domain-specific
factors b 920.44 48 .56 .68 .191
(D) Model (C) emotion ratings
as correlated uniquenessc 72.47 30 .96 .98 .053
(E) Model (D) + acheivement
scores b 199.98 62 .91 .96 .067

Note. N == 721 . NNFI == Non-Normed Fit Index ; CFI == Comparative Fit Index; RMSEA ==
Root Mean Square Error of Approximation. X2 values for null models used to define the
NNFI and CFI == 2323.38 (df== 66) for models A to D and 3350.58 (df== 120) for Model E.
"Emotion-specific factors are enjoyment, anxiety, and boredom. bDomain-specific factors
and achievement scores refer to mathematics, Latin, German, and English. Cln models D and
E, correlated uniqueness were permitted.

specific factors, correlated uniquenesses). This indicates that the true relations of
domain-specific emotional experiences are overestimated if method effects are
not taken into account.
The intercorrelations between the achievement scores of the four domains
were average in size (should be the range: r = [.37, .56] alternatively it could be:
range = .37 to .56; Mdn = .50; see Table 2, lower right area). We found a consis-
tent pattern in the correlations between the four domain factors and the achieve-
ment scores (Table 2, lower left section). That is, we found the highest correla-
tions within subject domains (mathematics = .48, Latin = .32, German = .28, and
English = .30). We found the strongest correlations in the mathematics domain.
Aside from the four within-domain correlations, there are no noteworthy corre-
lations between the four domain-specific factors and achievement scores (the
range: Irl = [.01; .15] or alternatively: range = .01 to .15; Mdn [Irl] = .07).

Degree of Domain Specificity in Discrete Emotions

Table 3 presents the results of the multilevel analyses. As for the fixed effects,
our results showed that across the four domains (mathematics, Latin, German,
English) the overall mean was highest for enjoyment (Yoo = 2.98), followed by
boredom (Yoo = 2.60), and anxiety (Yoo = 2.05). The estimated mean value for
achievement across the four domains was Yoo = 4.05.
TABLE 2. Intercorrelations of Domain-Specific Factors, Correlations of Domain-Specific Factors With Achievement Scores,
and Intercorrelations of Achievement Scores

Model E
(domain-specific factors and achievement scores.
correlated uniqueness of emotion ratings)
Domain-specific factors Achievement scores
Variable Math Latin German English Math Latin German English

Domain-specific factors
Mathematics 1.00
Latin .20 1.00
German .06 .25 1.00
English .00 .14 .33 1.00
Achievement scores
Mathematics .48 .10 - .03 - .04 1.00
Latin .08 .32 - .01 .09 .52 1.00
German .05 .13 .28 .1 5 .37 .47 1.00
English .01 .10 .00 .30 .42 .56 .52 1.00

Note. Domain-specific factors and performance scores refer to the academic domains of mathematics, Latin, German, and English.

>0
 tv
o

TABLE 3. Results of the Multi-level Analysis of the Extent of Doml!in Specificity of Anxiety, Boredom, and Achievement

Random effects
Fixed effects Level 2 effect (UOi) Level 1 effect (rij) Intra-class correlation
Average student Variance between Variance within 99% confidence
Traits mean (Yoo) SE students ('too) df X2 p students (0-2) P interval of p

Enjoyment 2.98 .02 . 148 709 1073.94 <.001 1.061 .123 .084 - .161
Anxiety 2.05 .03 .405 709 2018 . 15 <.001 .812 .333 .293 - .372
Boredom 2.60 .03 .380 709 1738.64 < .001 .966 .282 .243 - .322
Achievement 4.05 .03 .417 709 2851.46 < .001 .541 .448 .409 - .486

Note. Intra-class correlation: p = Too I (Too + (12). Students means and variations for enjoyment, anxiety, boredom and achievement we.re calculated
from the respective scores in the academic domains of mathematics, Latin, German, and English. N = 721.
 21

The analysis of the random effects showed that the levels of reported enjoy-
ment, anxiety, and boredom, as well as performance, differed significantly across
students (each p < .001). The total variance Var(Yij) = 1.209 for enjoyment, 1.2 17
for anxiety, 1.346 for boredom, and .931 for achievement. When comparing the
variances of emotions and achievement, it is important to note that we obtained
the three emotions using a 5-point Likert scale, whereas we assessed academic
achievement (grades) on a 6-point scale. The intraclass correlation (r), used here
as an indicator of the extent of domain specificity, was .448 for achievement,
.333 for anxiety, .282 for boredom, and .123 for enjoyment. The 99% confidence
intervals for the intraclass correlations suggest that these four constructs, includ-
ed in our study can be located at three statistically distinguishable levels of do-
main specificity. At the first level, there is achievement, which had the highest in-
traclass correlation (and, thus, a considerable degree of domain-transcendence) ,
at the second level there is anxiety and boredom with moderate intraclass corre-
lations (and, thus, a moderate degree of domain specificity), and at the third level
there is enjoyment with the lowest intraclass correlation, which can therefore be
considered the most domain-specific construct in the present study.

Discussion

Domain Specificity of Emotional Experiences

The poor fit of Models A, B, and C to the data indicates that it is not recom-
mendable to group the two-facet items to one factor representing domain-general
emotionality (Model A), to three factors representing domain-transcending emo-
tions (Model B), or to four factors representing domain-specific emotionality
(Model C). The poor fit of Model A clearly shows that the assumption of a general
factor of domain-general emotionality, from whose strength one could infer the in-
tensity of discrete emotional experiences in different domains, is not tenable. Fur-
thermore, and central to the core research question in this study, is the poor fit of
Model B, which shows that it does not make empirical sense to build emotion-
specific factors. This means that one cannot presume that there are latent factors rep-
resenting enjoyment, anxiety, and boredom from which one could infer the strength
of the respective emotional experience in different subject areas. This finding clearly
indicates a domain-specific organization of emotional experiences. The relatively
poor fit of Model C to the data also shows that the assumption of latent factors con-
cerning emotionality predicting discrete emotions across various domains such as
mathematics, Latin, German, and English cannot be empirically confirmed. This
lack of confilmation indicates that one cannot assume the existence of a latent fac-
tor domain-specific emotionality, from which one could draw conclusions regard-
ing the intensities of discrete emotional experiences within a given domain.
 22

Although Models A, B, and C demonstrate an unacceptable fit to the data,

Model C is substantially superior to Models A and B. A comparison of Models
Band C, which assume the existence of three or four factors, respectively, indi-
cates that a structure that groups the three different emotions from the same aca-
demic domain (Model C) is superior to a structure that groups the items reflect-
ing four different academic domains according to the same discrete emotion
(Model B). This finding strengthens the assumption of a domain-specific organi-
zation of emotional experiences in that it is more empirically justifiable to group
items that refer to different discrete emotions but one specific domain rather than
to group items that refer to one discrete emotion but different domains . In other
words, this finding shows that the relations between different emotional experi-
ences within a domain are stronger than those between experiences of one spe-
cific emotion across different domains.
Along with assessing the four domain factors in Model D (CU model), we con-
sidered the potential effects of the three emotion factors by permitting correlated
uniquenesses (i.e., we took into account the existence of method effects). The
good fit of this model makes it clear that it is necessary to evaluate both domain-
specific and emotion-specific factors simultaneously. In contrast to Model C, the
intercorrelations of the factors in Model D are in better congruence with the data
because method effects were taken into consideration in this model. Moreover,
our findings showed that the relations between the domain-specific factors were
overestimated if method effects were disregarded.
The very weak correlations between the four domain factors also clearly indi-
cate a domain-specific organization of emotional experiences. A comparison
with the intercorrelations of the four achievement scores reveals that, despite a
similar correlation pattern, those are much stronger than the correlations between
the factors reflecting domain-specific emotionality. This result indicates that
emotions are more domain-specifically organized than is academic achievement.
In addition , these findings showed weak correlations between the same emotion-
al experiences in different domains that were regarded as relatively similar in na-
ture, most notably between German and English, German and Latin, and mathe-
matics and Latin. Although relationships between the same emotional
experiences in the pure language-based subject areas (e.g., English, German) and
mathematics are virtually zero, there are small correlations between emotional
experiences in Latin and mathematics. In terms of the experienced emotions,
Latin appears to have ties to both the mathematics and language-based do-
mains---especially to German language.
The pattern of correlations between the four domain factors and the perfor-
mance scores of Model E (between-network) indicate high convergent and dis-
criminant validity of these factors, and yet strengthen the assumption of domain-
specific emotional experiences. Using MTMM terminology (Marsh, 1988a),
substantial correlations between domain-specific factors and achievement in
 23

matching school subjects provided strong support for the convergent validity of
the factors , and the relative lack of correlations between scores in nonmatching
school subjects supported the divergent validity of the factors. However, even the
very weak correlations across domains reflect, in part, the between-domain rela-
tionships of the domain-specific factors (e.g., positive correlations between per-
formance in German and emotional experiences in English). This reflection is an
additional indicator of the high validity of the domain factor. It is notable that the
within-domain correlations between the emotion factors and achievement also
vary across subject areas, with the correlations for the mathematics domain being
highest. This finding also indicates that the strength of the relationship between
emotional experiences and performance is domain-specific.
In sum, our results are in line with our theoretical assumptions and clearly in-
dicate domain specificity of emotional experiences. Both the findings from the
competitive model testing and analysis of the model's within-network, as well as
results of the convergent and discriminant validity of the domain-specific factors
can be regarded as clear and consistent empirical support for our hypotheses.

Extent of Domain Specificity of Discrete Emotions

The findings of the HLM analyses show that enjoyment is the most intensively
experienced emotion, followed by boredom and anxiety. As an indicator for the
extent of domain specificity, we looked at the intraclass correlation p. The small-
er the p value, the greater the domain specificity in the given construct. Achieve-
ment showed the significantly lowest degree of domain-specific organization.
With regard to the emotions, we observed a notable difference in the extent of do-
main specificity, with enjoyment being significantly more domain-specific than
anxiety and boredom, which in turn did not significantly differ from each other in
the extent of their domain specificity. That is, enjoyment proved to be the most do-
main-specific of the four constructs in that, even though the total variance of the
three emotions across students and domains was approximately the same, the pro-
portions of variance attributable to the differences between domains was relative-
ly strong for all the three emotions, and particularly for enjoyment.
Even though we argue for a stronger consideration of the domain specificity of
emotions, it may be worth mentioning that our findings do indicate that, for anxi-
ety and boredom, there is a very weak factor representing "domain-general emo-
tionality" (e.g., Model A). Across the domains, experiences of anxiety and bore-
dom appear to vary less than experiences of enjoyment. To the knowledge of the
authors, the question of situation specificity remains unexplored in the research lit-
erature with respect to boredom. Anxiety has been frequently discussed as an idio-
syncratic emotion; a genetically transmitted habitualized disposition in which
physiological processes, related to the subsystems of the limbic system (especially
the amygdala), playa central role (Zeidner, 1998). Our findings suggest that anxi-
24

ety is in fact a more domain-transcending emotion than is enjoyment. Furthermore,

our results show that boredom is also less domain-specific than enjoyment. How-
ever, even for anxiety and boredom, when compared with academic performance,
we conclude that the extent of domain specificity is strong enough to preclude fur-
ther discussion of academic anxiety and boredom as domain-general constructs.

Limitations

There are some limitations of our study which might have implications for fu-
ture research on the domain specificity of academic emotional experiences. Con-
cerning the domains; investigated in this study, we concentrated on mathematics,
Latin, German, and English, that is, on mathematics and three language classes.
It would be interesting to integrate science in further analysis to see how it is re-
lated to mathematics. Further, it would be worth investigating emotions experi-
enced not only in core subjects but also in subsidiary school subjects (e.g., music
and art education). Thus, because our study focused only on four subject areas,
further research on the extent of domain specificity for emotions assessed in
other school subjects is warranted.
Concerning the emotions, we investigated three emotions which have been
found to be salient in an academic environment and could be distinguished based
on activation and valence, namely the emotions of enjoyment, anxiety, and bore-
dom. However, our results are limited to these emotions. It would be worth in-
vestigating the domain specificity of other emotions which can be assumed to be
important for students' psychological well-being in the classroom (e.g., enthusi-
asm, shame, guilt, and disappointment).
Finally, although our results indicate that the measurement of emotions using
single-item measures in the current study was sufficiently valid (see Method sec-
tion), it might be assumed that single-item scales assess enjoyment, anxiety, and
boredom in a more global way than do multi-item scales which focus on specif-
ic facets of these emotions. Therefore, our results may have overestimated the re-
lationships of emotional experiences across the various subjects as compared to
results based on multi-item scales. However, this is merely a conjecture that can-
not be drawn from our results and should be investigated in more detail. Future
studies on the domain specificity of emotions using multi-item scales and ac-
counting for scale reliability are encouraged. Further, the use of multi-item scales
would enable an even more detailed level of analysis (see Bong, 200 I, for ana-
lyzing domain specificity of motivational constructs with multi-item scales).

Conclusions

CFA-MTMM and HLM analyses proved to be appropriate for analyzing the

two-faceted data structure of the present study. Results of these analyses provide
 25

critical empirical support for our assertion that emotional experiences are orga-
nized primarily in a domain-specific manner and that different emotions vary in
their extent of domain specificity. Thus, these analyses are recommended for other
studies utilizing a two-faceted design for investigating domain specificity of relat-
ed constructs (e.g., other emotions, self-concepts, or aspects of self-regulated learn-
ing). Through the use of single-item scales, such studies can be conducted in a
timely and cost-efficient manner. Nevertheless, single-item scales raise the ques-
tion about the reliability of these scales.
A number of conclusions can be drawn from our findings on the domain speci-
ficity of emotional experiences. First, it appears to be more appropriate to speak
of domain-specific rather than domain-transcending emotional experiences (e.g.,
enjoyment in mathematics instead of school-related enjoyment or enjoyment in
class).
Second, from a research perspective, it appears to be appropriate to investigate
students' emotions using domain-specific approaches. Within test anxiety research,
an area that has been investigated most intensively in the achievement-related con-
text, the majority of studies and intervention programs were conceptualized under
the implicit assumption of a domain-transcending organization of emotional expe-
riences (Zeidner, 1998). An exception is research on math-related anxiety (e.g.,
Miller & Mitchell, 1994). For testing models that refer to emotional experiences
and their relations to other constructs, the present research suggests that domain-
specific designs may be most appropriate (e.g., for testing the control-value theory
of emotions developed by Pekrun, [2000; in press], or a recently developed model
linking achievement goals and affect by Linnenbrink & Pintrich, [2002]). Conse-
quently, the development of domain-specific instruments for the measurement of
emotional experiences among students is also encouraged. Specifically, for the do-
main of mathematics and for students in grades 5-10, such an instrument already
exists (Academic Emotions Questionnaire-Mathematics, AEQ-M; Pekrun, Goetz,
& Frenzel, 2006) and included scales for the assessment of enjoyment, pride, anx-
iety, anger, hopelessness, boredom, and shame. Adapted versions of this instrument
could be used in future studies assessing academic emotions in different domains
and contexts.
Third, from a practical perspective, these results suggest that it may also be
helpful to generate domain-specific emotion-related counseling and intervention
programs (see Everson et aI., 1993). For example, it seems recommendable that
attributional retraining, an effective treatment method fostering adaptive causal
attributions and consequently positive emotional experiences (Hall, Hladkyj,
Perry, & Ruthig, 2004), and programs aimed at designing emotionally sound in-
struction (Astleitner, 2000; see also Goetz, Frenzel, Pekrun, & Hall, 2005)
should be implemented in a domain-specific way. Although domain-general pro-
grams may also have effects on domain-specific emotions, our results point to the
fact that domain-specific programs might be more efficient.
 26

Finally, the present analyses show that only limited conclusions about a stu-
dent's emotional experiences in a specific domain can be made from his or her
experiences in' another subject area. This finding stands in contrast to several
studies indicating that teachers tend to view individual students' characteristics
as habitual, domain-general attributes and not as domain-specific phenomena
(see Marsh, 1993; Marsh, Smith, & Barnes, 1983; Pohlmann, Moller, & Stre-
blow, 2004). Thus, one can assume that false judgments are often made by teach-
ers with regard to the emotional experiences of their students in different do-
mains. To prevent teachers from making false evaluations and to increase the
effectiveness of school instruction (e.g., adaptive teaching; see Como & Snow,
1986; Snow & Swanson, 1992), the findings of this study suggest that it is of crit-
ical importance for both teachers and educators to also be aware of the domain
specificity of students' emotional experiences.

REFERENCES

Abu-Hilal, M. M., & Bahri, T. M. (2000) . Self-concept: The generalizability of research on the SDQ,
Marsh/Shavelson model and liE frame of reference model to the United Arab Emirates students.
Social Behavior and Personality, 28, 309-322.
Astleitner, H. (2000). Designing emotionally sound instruction: The FEASP-approach. Instructional
Science, 28, 169-198.
Bentler, P. M. (1990). Comparative fit indices in structural models. Psychological Bulletin, 107,
238-246.
Bong, M. (1998). Tests of the internal/external frames of reference model with subject-specific aca-
demic self-efficacy and frame-specific academic self-concepts. loumal of Educational Psycholo-
gy, 90, 102-110.
Bong, M . (200 I). Between- and within-domain relations of academic motivation among middle and
high school students: Self-efficacy, task-value and achievement goals . loumal of Educational Psy-
chology, 93, 23-34.
Bong, M., & Skaalvik, E. M. (2003). Academic self-concept and self-efficacy: How different are they
really? Educational Psychology Review, 15, 1-40.
Byrne, B. M . (1998). Structural equation modeling with L1SREL, PRELlS, and SIMPLlS: Basic con-
cepts, applicatiolls, and programmillg . Mahwah, NJ : Erlbaum.
Corno, L., & Snow, R. E. (1986). Adapting teaching to individual differences among learners. In M .
C. Wittrock (Ed.), Handbook of research on teaching (3rd ed., pp. 605--{i29). New York: Macmil-
lan.
Dickhaeuser, 0., & Plentei', I. (2005). Letzes Halbjahr stand ich zwei. Zur Akkuratheit selbst
berichteter Noten [On the accuracy of self-reported school marks]. Zeitschriji jlir Pddagogische
Psychologie, 19, 219-224.
EfkJides, A., & Volet, S . (Eds.). (2005). Feelings and emotions in the learning process [Special issue].
Learning and Instruction, 15.
Everson, H. T., Tobias, S., Hartman, H., & Gourgey, A. (1993) . Test anxiety and the curriculum: The
subject matters. Anxiety, Stress and Coping: An Internatiollal loumal, 6, 1-8.
Fisher, R. A. (1970). Statistical methods for research workers . New York: Hafner.
Gottfried, A. E. (1982). Relationships between academic intrinsic motivation and anxiety in children
and young adolescents. loumal of School Psychology, 20, 205-215.
Goetz, T. (2004). Emotionales Erleben und selbstreguliertes Lemen bei Schiilem im Fach Mathe-
matik [Emotions and self-regulated learning of students in mathematics] . Munich, Germany: Utz.
Goetz, T., Frenzel, C. A., Pekrun, R., & Hall, N. (2005). Emotional intelligence in the context of
learning and achievement. In R. Schulze & R. D. Roberts (Eds.), Emotional intelligence: An in-
ternational Handbook (pp. 233-253). Cambridge, MA: Hogrefe & Huber.
 27

Hall, N. C ., Hladkyj, S., Perry, R. P., & Ruthig, J. C. (2004). The role of attributional retraining and
elaborative learning in college students' academic development. The Journal of Social Psycholo-
gy, 144,591-612.
Hembree, R. (1990). The nature, effects, and relief of mathematics anxiety. Journalfor Research ill
Mathematics Education, 21, 33-46.
Ingles, S. J., Scott, L. A., Lindmark, J. T., Frankel, M. R, Myers, S. L., & Wu, S. (1992). National
Education Longitudina l Study of 1988. Firstfollow-up: Studellt component datajile user 's manu-
al (Vol. I). Washington, DC: U.S. Department of Education.
Joreskog, K., & Sorbom, D. (2002). LISREL 8.53. Chicago, IL: Scientific Software International.
Linnenbrink, E. A., & Pintrich, P. R. (2002). Achievement goal theory and affect: An asymmetrical
bidirectional model. Educational Psychologist, 37, 69-78.
Marsh, H. W. (1984) . Relations among dimensions of self-attribution, dimensions of self-concept,
and academic achievements . Journal of Educational Psychology, 76, 1291-1308.
Marsh, H. W. (1986). The self-serving effect (bias?) in academic attributions: Its relation to academ-
ic achievement and self-concept. Journal of Educational Psychology, 78, 190-200.
Marsh, H. W. (1988a). The content specificity of math and English anxieties: The high school and be-
yond study. Anxiety Research, I, 137-149.
Marsh, H. W. (l988b). Multitrait-multimethod analyses. In J. P. Keeves (Ed .), Educational research,
methodology, and measurement: An international handbook (pp. 570-580). Oxford, England:
Pergamon.
Marsh, H. W. ( 1989). Confirmatory factor analyses of multi trait-multi method data: Many problems
and a few solutions. Applied Psychological Measuremelll, 13, 335-361.
Marsh, H. W. (1993). Academic self-concept: Theory measurement and research. In J. Suls (Ed .),
Psychological perspectives on the seif(Vol. 4, pp. 59-98). Hillsdale, NJ: Erlbaum.
Marsh, H. W., Balla, J. R., & McDonald, R. P. (1988). Goodness-of-fit indexes in confirmatory fac-
tor' analysis: The effect of sample size. Psychological Bulletin, 103, 391-410.
Marsh, H. W., Byrne, B. w., & Craven, R (1993). Overcoming problems in confirmatory factor
analyses of MTMM data: The correlated uniqueness model and factoria l invariance. Multivariate
Behavioral Research, 27, 489-507.
Marsh, H. w., & Grayson, D. (1995). Latent variable models of multitrait-multimethod data. In R. H.
Hoyle (Ed .), Structural equation modelling: Concepts, issues, and applications (pp. 177-198).
Thousand Oaks, CA: Sage.
Marsh, H. W., Smith, I. A., & Barnes, J. (1983). Multitrait and multi method analysis of the Self-De-
scription Questionnaires: Student-teacher agreement on multidimensional ratings of student self-
concept. American Educational Research Journal, 20, 333-357.
Marsh, H. w., & Yeung, A. S. (1996) . The distinctiveness of affects in specific school subjects: An
application of confirmatory factor analysis with the National Educational Longitudinal Study of
1988. American Educational Research Journal, 33, 665-689.
Miller, L. D., & Mitchell, C. E. (1994). Mathematics anxiety and alternative methods of evaluation.
Journal of Instructional Psychology, 21, 353-358.
Moller, J., & Koller, O. (2001) . Dimensional comparisons: An experimental approach to the Inter-
nal/External frame of reference model. Journal of Educational Psychology, 93, 826-835.
Moller, J., & Koller, O. (2004) . Die Genese akademischer Selbstkonzepte: Effekte dimensionaler und
sozialer Vergleiche [On the development of academic self-concepts: The impact of social and di -
mensional comparisonsJ. Psychologische Rundschau, 55, 19-27.
National Center for Educational Statistics. (1986). High school and beyond, 1980: Sophomore cohort
secondfollow-up (1984). Datajile user's manual. Ann Arbor, MI: Inter-university Consortium for
Political and Social Research.
Pekrun, R. (1992). The impact of emotions on learning and achievement: Towards a theory of cogni-
tive/motivational mediators. Applied Psychology: An International Review, 41, 359-376.
Pekrun, R. (2000). A social-cognitive, control-value theory of achievement emotions. In J. Heck-
hausen (Ed .), Motivational psychology of human development (pp. 143-163). Oxford, England:
Elsevier.
Pekrun, R. (in press). The control-value theory of achievement emotions: Assumptions, corollaries,
and implications for educational research and practice. Educational Psychology Review.
Pekrun, R, Goetz, T., & Frenzel, A. C. (2006) . Academic Emotions Questionnaire-Math ematics
28

(AEQ-M)- Vser's manual. Munich, Germany: University of Munich, Department of Psychology.

Pekrun, R., Goetz, T., Titz, W. , & Perry, R. (2002a). Academic emotions in students' self-regulated
learning and achievement: A program of qualitative and quantitative research. Educational Psy-
chologist, 37, 91 - 105.
Pekrun, R., Goetz, T., Titz, W, & Perry, R. P. (2002b, April). A social cognitive, control-value theo-
ry of achievement emotions: Social antecedents and achievement effects of sit/dents ' domain -relat-
ed emotions. Paper presented at the annual meeting of the American Educational Research Asso-
ciation, New Orleans, LA.
Pohlmann, B., Moller, J., & Streblow, L. (2004). Zur Fremdeinschiitzung von Schiilerselbstkonzepten
durch Lehrer und Mitschiiler [On students' self-concepts inferred by teachers and classmates].
German Journal of Educational Psychology, 18, 157- 169.
Raudenbush, S. W, & Bryk, A. S. (2002) . Hierarchical linear models. Applications and data analy-
sis methods (2nd ed.). Thousand Oaks, CA: Sage.
Raudenbush, S., Bryk, A., Cheong, Y. F. , & Congdon , R. (200 I). HLMTM 5. Hierarchical linear and
nonlinear modeling. Lincolnwood, IL: Scientific Software International.
Richardson, F. c., & Suinn, R. M. (I 972). The Mathematics Anxiety Rating Scale: Psychometric
data . Journal of Counseling Psychology, 19,551 - 554.
Schutz P. A., & Lanehart, S. L. (Eds.). (2002) . Emotions in education [Special issuel . Educational
Psychologist, 37(2).

APPENDIX
TABLE AI. Means (M), Standard Deviations (SD), and Intercorrelations of Single
Items Scales and Achievement Scores

Variable Abbreviations M SD 2 3 4

1. Enjoyment M M_En 3.00 1.17

2. Enjoyment L L_En 3.04 1.24 .16
3. Enjoyment G G_En 3.22 1.01 .04 .21
4. Enjoyment E E_En 2.88 1.01 .07 .13 .25
5. Anxiety M M_An 2.18 1.15 -.57 -.07 .lD .05
6. Anxiety L L_An 2.25 1.24 -.17 - 047 .01 -.05
7. Anxiety G G_An 1.92 .97 - .09 .05 -.27 - .16
8. Anxiety E E_An 1.78 .95 - .03 -.0 1 -.05 -.50
9. Boredom M M_Bo 2.50 1.15 - .58 -.18 - .12 - .07
lD. Boredom L L_BQ 2.57 1.28 -.19 - .75 -. 18 -.14
II. Boredom G G_Bo 2.37 1.05 - .lD -.19 -.57 - .22
12. Boredom E E_Bo 2.80 1.14 -. 11 - .11 -. 18 -.59
13. Performance M M_Ac 4.03 .97 040 .lD - .06 -.01
14. Performance L L_Ac 3.92 1.21 .08 .31 .00 .09
IS. Performance G G_Ac 4.19 .77 .01 .13 .19 .16
16. Performance E E_Ac 4.05 .93 -.02 .05 .00 .31

Note. Means and standard deviations of the single-item scales (measurement of discrete e motional
experiences in various academic domains) are represented along with achievement scores. M =
mathematics, L = Latin, G = German, E = English. p < .05 for Irl ~ .07; p < .0 1 for Irl ~ . 10;
p <.OOI for lrl ~. 12 . N=72l.
 29

Snow, R. E., & Swanson, J. (1992). Instructional psychology: Aptitude, adaptation, and assessment.
Allllual Review of P~ychology, 43, 586-626.
Steiger, J. H. , & Lind, J. C. (1980, June). Statistically based testsfor the lIumber of comlllon fa ctors.
Paper presented at the Psychometric Society Annual Meeting, Iowa City, IA.
Stevenson, H. w., & Newman, R. S. (1986). Long-term prediction of achievement and attitudes in
mathematics and reading. Child Development, 57, 646-659.
Tucker, L. R., & Lewis, C. (1973). A reliability coefficient for maximum likelihood factor analysis.
Psychometrika, 38, 1-10.
Turner, J., & Schallert, D. (200 I). Expectancy-value relationships of shame reactions and shame re-
silience. Journal of Educational Psychology, 93, 320-329.
Ukoumunne, O. C. (2002). A comparison of confidence interval methods for the intraclass correla-
tion coefficient in cluster randomized trials. Statistics ill Medicine, 21, 3757-3774.
Watson, D., & Tellegen, A. (1985). Toward a consensual structure of mood. Psychological Bulletin,
98, 219-235.
Weiner, B. (1985). An attributional theory of achievement motivation and emotion. Psychological Re-
view, 92, 548- 573 .
Widaman, K. F. (1985). Hierarchically nested covariance structure model s for multi trait- multi-
method data. Applied Psychological Measurement, 9, 1-26.
Zeidner, M. (1998). Test anxiety: The state of the art. New York: Plenum.

5 6 7 8 9 10 II 12 13 14 15 16

042
.36 .32
.30 .39 044
Al .18 .13 .05
. 14 .35 -.02 .02 .34
.04 .07 .22 .08 .30 .34
.07 .08 .15 .33 .27 .26 .39
- .29 - .10 .07 .05 - .24 -. 10 - .01 -.01
-.01 -.26 .06 - .10 - .08 - .18 .06 .09 .52
.04 - .10 -. 14 - .07 - .06 - .09 -.12 - .04 .37 047
.08 - .06 -.02 - .23 - .00 - .02 .04 -.03 042 .56 .52