Professional Documents
Culture Documents
net/publication/247496732
CITATIONS READS
301 699
1 author:
Bettina Seipp
Rajiv Gandhi Proudyogiki Vishwavidyalaya
4 PUBLICATIONS 328 CITATIONS
SEE PROFILE
All content following this page was uploaded by Bettina Seipp on 18 April 2018.
Bettina Seipp
To cite this article: Bettina Seipp (1991) Anxiety and academic performance: A meta-analysis of
findings, Anxiety Research, 4:1, 27-41, DOI: 10.1080/08917779108248762
The strength of the relationship between anxiety and performance vanes from study to study with
correlations from extreme negative to positive values. In order to reveal the sources of this inconsis-
tency, a series of meta-analyses was conducted using the Schmidt-Hunter algorithm for effect sizes r.
One hundred and twenty-six studies published from 1975 to 1988, based on a total sample of 36,626
subjects, were located after a comprehensive literature search. They include 156 indspendent samples.
An overall analysis with the 156 effect sizes yielded a population effect size of rV=-.21. Further
analyses aimed at exploring moderator variables that would account for the residual variance, but tests
of gender, culture (USA, West Germany and others), and anxiety stability (stateltrait) failed to unveil
the expected moderator impact. However, analyses with the anxiety components worry and emotiona-
lity, kinds of anxiety such as general and test anxiety, and the anxiety measurement point in time yielded
systematic differences: the more cognitively determined and the more specific the anxiety measure, the
closer was its association with academic performance. A closer relationship was also found if anxiety was
measured after the performance situation compared to being measured before.
KEY WORDS: Cognitive theories of anxiety, academic performance, meta-analysis, research integ-
ration
27
28 B. SEIPP
intelligent” (Hornke & Tries, 1979) integration of empirical research results from
conceptual replication studies. It uses statistics from primary studies (means,
correlations, probabilities) to calculate the (weighted) mean or population effect
size (standardized mean difference, average correlation, combined significance).
Hunter, Schmidt, and Jackson (1982) correct the variance of effect sizes across
studies for sampling error by subtracting the estimated error from the observed
variance and thus end up with the corrected or residual variance. The mean effect
size is significantly different from zero if it is larger than twice the corrected
standard deviation. It can be interpreted meaningfully if it is calculated from a
homogeneous set of effect sizes. Homogeneity, according to Hunter et al. (1982), is
given if at least 75% of the observed variance is explained by sampling error.
Downloaded by [Universitaetsbibliothek Dortmund] at 04:01 11 December 2017
METHOD
Data Base
As a result of a systematic computer-assisted (PSYNDEX, ERIC) and an addi-
tional unsystematic literature search, a total of 359 studies were localized for the
time span of 1975 to 1988. Studies, however, were excluded i f
- anxiety was operationalized on physiological grounds,
- performance was operationalized as IQ, anagram-solution, memory, etc. ,or
- the direction and the strength per se of the effect were not calculable e.g., in
the case of multiple regression weights, canonical correlations, etc.).
Measured by these criteria, 126 studies (35.1%) could finally be used for the
analysis. The associated dropout rate is in line with other recent meta-analyses,
where an average of 365 studies were found, but only an average of 106 (33%)
could be used (see Seipp, 1990).
From these 126 independent studies a total of k = 156 effect sizes (two per study
in the case of separate results for males and females, otherwise one) were entered
into the analysis.
Table 1 shows the distribution of studies to various sources of origin.
A total sample size of 36,626 persons with a mean of 235 is investigated. One
hundred and twenty-six effect sizes are originally correlations, 30 means and
standard deviations, t or F values, which are transformed into correlations, too.
Since most of the original studies are correlational, the causal relationship
between anxiety and performance is not always clear; anxiety, however, is treated
as the independent variable (in accordance with the wording of the anxiety
questionnaires). Anxiety is operationalized by the questionnaires used in the
ANXIETY AND PERFORMANCE 29
primary studies, which, in the majority of cases, are intended to measure trait
anxiety. Furthermore, the components “worry” and “emotionality” are mostly
measured together because these two factors were included intuitively in nearly all
of the questionnaires, even before Liebert and Morris (1967) suggested the
existence of two separate factors.
Academic performance, i.e., the results of exams or tests in schools or universit-
ies, is regarded as the dependent variable. Once again, it is defined by the
instruments used to measure it. They are standardized tests, such as subscales from
the Achievement College Test, the Stanford Achievement Test, or informal
measures, such as test grades, Grade Point Average (GPA), and oral examin-
ations.
Downloaded by [Universitaetsbibliothek Dortmund] at 04:01 11 December 2017
Origin Frequency
Country
USA 89
FRG 29
Others 19
Publication
Journal articles 75
Book chapters 6
Books 5
Dissertations 19
Conference papers 21
Data Analysis
The first research question asks for the homogeneity of effect sizes: The stem-and-
leaf display in Figure 1 as well as the histogram of effect sizes in Figure 2 show a
symmetrical unimodal distribution.
This almost normal distribution seems to imply that all effects belong to the same
population and that all deviations from the parameter are due to sampling error.
The graphic displays, however, are not “tests” of homogeneity, but are only first
indicators of homogeneity, which has to be tested by the Hunter et al. (1982)
statistical criteria (see above).
RESULTS ’
Total Sample
The total sample, comprising N=36,626 persons and k=156 effect sizes, yields a
population effect size of T,= - .212.Its 95% credibility interval (which uses the
residual instead of the observed standard deviation) ranges from - .37 to - .07 (see
Table 2).
The reported mean (T,,,=-.21) - even though it is the best estimator of the
relationship - is of only limited interpretational value. It is clearly negative, but its
’ The analyses were performed with a microcomputer program written by Ralf Schwarzer (1989).
30 B. SEIPP
-.9 I
-.8 I
-.7 I
-.6 I 6
-.S 1036678
-.4 I00023367799
-.3 I 0000122223445667778
Downloaded by [Universitaetsbibliothek Dortmund] at 04:01 11 December 2017
-.2 I 00111111122222233333334444466666777888888899999
-.l I 00001111112333333444444466666666666777778888999
-.O I 01123666666779
+.O I 1249
+.l I 237
+.2 I
+.3 I 7
+.4 I
+.5 I
+.6 I
+.7 I
+.8 I
+.9 I
heterogeneity calls for a search for additiona. ,actors connected with study cdsigns,
subjects, kinds of anxiety, etc. They would be considered as real moderators if (a)
“the average effect size will vary from subset to subset, and (b) the corrected
variance will average lower in the subsets than for the data as a whole” (Hunter et
al., 1982, p. 48).
Number
of Effect Sizes
t
50
40
30
Downloaded by [Universitaetsbibliothek Dortmund] at 04:01 11 December 2017
20
10
-
Variable k N rw 3 r s2e s2c s, 7'0 expl. 95%
variance credibility
interval
~ ~~ ~ ~
All 156 36626 -.212 .Wa) .Ma) .ma) .075 41.0 -.36 to
-.07
IT,l>2,=.15 sig., s,>lTwl x.25=.053, heterogeneous
-
Variable k N rw 8, 4, s2c s, 7'0 expl. 95%
variance credibility
interval
This result is one of the most important ones of this meta-analysis: on the highest
level of generalization, gender is not a moderating factor for the anxiety-
performance relationship. The difference between their respective means is only
.03. In spite of that, additional factors are differentially influencing the effects for
the male and female subgroups, which are both heterogeneous.
the respective means are comparable, but the large credibility intervals show that
additional factors must be found here, too.
Variable k N
-rW 2, s2, sc % expl. 95%
S2C
variance credibility
interval
USA
Females 22 ' 2926 -.241 .012 .007 .006 .076 53.9 -.39 to
.09
lPwl > .15 sig., s,>.060, heterogeneous
Males 18 3937 -.167 .008 .004 .004 .062 52.6 -.29 to
-.04
IP, I> .12 sig., s, > .042, heterogeneous
Others
Variable k N -
TW sz, s2, s2c s, YO expl. 95%
variance credibility
interval
~
Variable k N -
rv ?r s2e s2c s, % expl. 95%
variance credibility
interval
~~
Contradictory to expectations, the effect sizes are identical for state and trait
anxiety. This is also true for the credibility intervals. On the other hand, one has to
take into account the large number of effects for trait anxiety (k = 137) as
compared to the few for state anxiety (k = 29). Therefore, a second analysis uses
only those studies which investigate the impact of both state and trait anxiety (see
Table 8).
The absolute values change slightly, but the general statements of no difference
and of heterogeneity are maintained. This is also partly true when a third analysis is
done with 22 random effect sizes for both state and trait anxiety. Still, the effect
sizes remain nearly identical; trait anxiety constitutes a homogeneous subset here ,
ANXIETY AND PERFORMANCE 35
Table 8 Meta-Analysisfor state and trait anxiety measured with the same persons
but this is supposed to be an effect of the small sample sue if compared to the
available set.
General anxiety 53 11680 -.163 .007 .004 .003 .050 63.6 -.26 to
-.07
IPWl>.10sig., s,>.041, heterogeneous
~~~~~~~~~
Test anxiety 114 28424 -233 .009 .004 .005 .070 42.0 -.37 to
- .09
IPW1>.14sig., sc>.058, heterogeneous
The mean effect sizes differ with .07,and the average residual variance is smaller
than for the data as a whole (s-2c=.004<s2,=.006). Even though both subgroups
are not sufficiently characterized on this level of analysis, generalised across all
other possible factors the situation specificity of anxiety is a first-order moderator of
its relationship with performance.
The three findings concerning the anxiety construct -worry/emotionality ,state/
trait anxiety, generaUtest anxiety - are summarized in Figure 3.
36 B. SEIPP
Effect Sizes
-.25
-.20
-.15
Downloaded by [Universitaetsbibliothek Dortmund] at 04:01 11 December 2017
-.lo
-.05
0 construct
State Trait nerd Anxietflest Anxiety Worry Emotionality
Type of anxiety Anxiety content Anxiety component
-
Variable k N 1, 3, sze s“, s, % expl. 95%
variance credibility
interval
Before 31 4498 -.211 .009 .006 .003 .051 70.3 -.32t0
-.12
I i, I > .10 sig., s, > .055, homogeneous
After 9 2056 -.283 .001 .004 -.003 .ooO 100 -
homogeneous
Downloaded by [Universitaetsbibliothek Dortmund] at 04:01 11 December 2017
ES
-.30 1
-.25
-.20
-.15
-.lo
-.05
0
indepen- before after
- time
dent
Figure 4 Effect sizes depending on measurement times.
Taking this result into account, the possibility arises that anxiety is not so much
impairing performance, but rather that it is the achievement outcome that is
decisive for the amount of anxiety experienced. An alternative explanation,
however, would be to conceive the retrospective confirmation of anxiety as a
deliberate “excuse” for poor performance results (see Laux and Glanzmann, 1987).
DISCUSSION
This meta-analysis aimed at (a) establishing the homogeneity or heterogeneity of
38 B. SEIPP
the relationship between anxiety and academic performance as well as its absolute
strength, and (b) determining different effects for distinguishable populations and
facets of the anxiety construct ’.
The overall data set turned out to be highly heterogeneous and yielded a general-
ized effect size of T,= - .212 with a credibility interval ranging from - .07 to -.36.
The relationship is thus clearly negative, which means that high anxiety goes
together with poor performance and vice versa. However, one must keep in mind
that this relationship is much weaker than has often been postulated. One could
think that this might be due to the often claimed nonlinear relationship between
anxiety and performance (see Spence & Taylor, 1951,1953) which is not taken into
account here. On the other hand, academic tasks can always be characterized as
“complex”, and all theories trying to explain the negative relationship between
anxiety and performance claim that in this case a “high enough” intensity of anxiety
is reached (See Carver & Scheier, 1986, 1988; Easterbrook, 1959; Mandler &
Sarason, 1952; Sarason, 1975; Taylor, 1956; Wine, 1971, 1980). Thus, in the
presence of academic achievement, the relationship tends strongly towards linear-
ity, and the weak correlation can be accepted as reflecting the “true” state of
affairs. This weakness is convincingly demonstrated when determining the amount
of variance shared by the two variables. A mean correlation of -.21 yields an
T2,x10O% of 4.5%, which is generally assumed to be a very small amount of
common variance. Two other ways of interpreting the identical correlation do more
justice to the importance of the strength of the anxiety-performance relationship.
The effect size r (correlation) can be transformed into d (standardized mean
difference), which denotes an effect size in units of the standard normal distribu-
tion. In the present case, an T, of -.21 is converted into a d of - -043: High-anxious
subjects differ almost half a standard deviation from low-anxious subjects on a
fictitious achievement scale. If those low-anxious subjects obtain a percentile of 50,
their high-anxious counterparts would arrive at a percentile of 33, i.e., two-thirds of
the low-anxious students score better than the average high-anxious student. The
relationship between anxiety and academic performance can be illustrated by the
Binomial Effect Size Display (BESC) (Rosenthal, 1984) as well, if both variables
are simplified into dichotomies. According to this illustration method, only 39% of
the low-anxious subjects would fail when 61% of the high-anxious subjects fail.
Several moderators were found to mediate the relationship on the highest level
(first-order moderators), having closer relationships with performance:
For many more specific relationships-within the performance construct,for publicationfeatures, and
for the methodologicalquality of the original studies -see R. Schwaner (1989); R. Schwarzer, Seipp,
& C. Schwarzer (1989); Seipp (1990), Seipp & C. Schwarzer, (1987).
ANXIETY AND PERFORMANCE 39
References
Bangert-Drowns, R.L. (1986). Review of developments in meta-analytic method. Psychological
Bulletin, 99, 388-399.
Bangert-Drowns, R.L., & von Saldern, M. (1986). Methoden der Meta-Analyse -- ein Uberblick
[Methods of meta-analysis]. Zeitschrifi f i r Erziehungswissenschafiliche Forschung, 20.3-32.
Carver, C.S., & Scheier, M.F. (1986). Functional and dysfunctional responses to anxiety: The
interaction between expectancies and self-focussed attention. In R. Schwarzer (Ed.), Self-related
cognitiom in anxiety and motivarion. (pp. 111-114). Hillsdale, N.J.: Erlbaum.
Carver, C.S., & Scheier, M.F. (1988). A control-process perspective of anxiety. Anxiety Research, 1,17-
22.
Easterbrook, J.A. (1959). The effect of emotion on cue utilization and the organization of behaviour.
Psychological Review, 63, 183-201.
Fricke, R., & Treinies, G. (1985). Einfiihrung in die Meta-Analyse. [Introduction to meta-analysis].
Bern: Huber.
Glass, G.V. (1976). Primary, secondary, and meta-analysis of research. Educational Researcher, 5,3-8.
Glass, G.V. (197). Integrating findings: The meta-analysis of research. Review of Research in
Education, 5,351-379.
Glass G.V., McGaw, B., & Smith. M.L. (1981). Meta-analysis in social research. Beverly Hills: Sage.
Hedges, L.V. (1986). Issues in meta-analysis. Review ofResearch in Education, 13, 353-403.
Hedges, L.V., & Olkin, I. (1985). Statistical methods for meta-analysis. Orlando, FL: Academic Press.
Hodapp, V. (1989, JunelJuly). An expanded version of the German Test Anxiety Inventory. Paper
presented at the 10th Conference of the Society for Test Anxiety Research. Amsterdam, The
Netherlands.
Hornke, L.F. (1973). Verfahren zur Mittelung von Korrelationen. [Methods for averaging correlations].
Psychologische Beitdge, 15, 87-105.
Hornke, L.F. (1983). Integration empirischer Forschungsergebnisse?[Integration of empirical research
results?] Psychologie in Erziehung und Unterricht, 30, 54-63.
Hornke, L.F., & Tries, J. (1979). Zusammenfassung empirisch-pfidogogischerErgebnisse im S i n e der
Meta-Analyse nach Glass. [Integration of empirical findings in education using Glass’ meta-analysis].
40 B. SEIPP
Dokumentatwn der Beiwtige der 27. Tagung der Arbeirrgemeinschaji f i r Empirisch P&dagogische
Forschung. (pp. 158-171). Bremen, Germany.
Hunter, J.E., Schmidt, F.L., & Jackson, G.B. (1982). Meta-analysis. Cumulating research findings
across studies. Beverly Hills: Sage.
Krohne, H.W. (1976). Theokn zur Angst. [Theories of Anxiety]. Stuttgart, Germany: Kohlhammer.
Laux, L., & Glarumann, P. (1987). A self-presentational view of test anxiety. In R. Schwaner, H.M.
van der Ploeg, & C.D. Spielberger (Eds.), Aduances in test anxiety research, Vol. 5 , (pp. 31-37).
Lisse, The Netherlands: Swets & Zeitlinger.
Liebert, R.M., & Moms, L.W. (1%7). Cognitive and emotional components of test anxiety: A
distinction and some initial data. Psychological Reports, 20, 975-978.
Light, R.J., & Pillemer, D.B. (1984). Summing up. Cambridge: Harvard University Press.
Light, R.J., & Smith, P.V. (1971). Accumulating evidence: Procedures for resolving contradictions
among different research studies. Haruard Educational Review, 41,429-471.
Downloaded by [Universitaetsbibliothek Dortmund] at 04:01 11 December 2017
Mandler, G., & Sarason, S.B. (1952). A study of anxiety and learning. Journal of Abnormal and Social
Psychology, 47, 166-173.
McDaniel, M.A., Hirsh, H.R., Schmidt, F.L., Raju, N.S., & Hunter, J.E. (1986). Interpreting the
results of meta-analytic research: A comment on Schmitt, Gooding, Noe, & Hirsch (1984). Personnel
Psychology, 39, 141-148.
Ploeg, H.M. van der, Schwaner, R., & Spielberger, C.D. (Eds.) (1983). A d u m e s in test anxiety
research, Vol. 2. Lisse, The NetherlanddHillsdale, NJ: Swets & ZeitlingerErlbaum.
Ploeg, H.M. van der, Schwaner, R., & Spielberger, C.D. (Eds.) (1984). Aduances in rest anxiety
research, Vol. 3. Lisse, The NetherlanddHillsdale, NJ: Swets & Zeitlinger/Erlbaum.
Ploeg, H.M. van der, Schwaner, R., & Spielberger, C.D. (Eds.) (1985). Aduances in test anxiety
research,, Vol. 4. Lisse, The Netherlands: Swets & Zeitlinger.
Rosenthal, R. (1984). Meta-analytic procedures for social research. Beverly Hills: Sage.
Sarason, I.G. (1975). Anxiety and self-preoccupation. In I.G. Sarason & C.D. Spielberger (Eds.),
Sfress and anxiety, Vol. 2 (pp. 27-44). Washington, D.C.: Hemisphere.
Schwarzer, C., & Kim, M.-J. (1984). Adaptation of the Korean form of the Test Anxiety Inventory: A
research note. In H.M. van der Ploeg, R. Schwarzer, & C.D. Spielberger (Eds.), Aduances in rest
anxiety research, Vol. 3 (pp. 277-285). Lisse, The NetherlanddHillsdale: Swets & ZeitlingerErIbaum.
Schwarzer, R. (1984). Worry and emotionality as separate components in test anxiety. International
Review of Applied Psychology, 33,205-220.
Schwarzer, R. (1987). Stress, Angst und Hiylosigkeit (2nd edition). [Stress, anxiety, and helplessness].
Stuttgart, Germany: Kohlhammer.
Schwarzer, R. (1989). Meta-Analysis Programs. Manual. Freie Universitht Berlin.
Schwarzer, R., Ploeg, H.M. van der, & Spielberger, C.D. (Eds.) (1982). Aduances in test anxiety
research, Vol. 1. Lisse, The Netherlands: Swets & Zeitlinger.
Schwarzer, R., Ploeg, H.M. van der, & Spielberger, C.D. (Eds.) (1987). Advances in test anxiety
research Vol. 5 . Lisse, The Netherlands: Swets & Zeitlinger.
Schwarzer, R., & Quast, H.-H. (1985). Multidimensionality of the anxiety experience: Evidence for
additional components. In H.M. van der Ploeg, R. Schwarzer, & C.D. Spielberger (Eds.)., Advances
in test anxiety research, Vol. 4 (pp. 3-14). Lisse, The Netherlands: Swets & Zeitlinger.
Schwaner, R., Seipp, B., & Schwarzer, C. (1989). Mathematics performance and anxiety: A meta-
analysis. In R.Schwaner, H.M. van der Ploeg, & C.D. Spielberger (Eds.), Aduances in test anxiety
research, Vol. 6 (pp. 109-119). Lisse, The Netherlands: Swets & Zeitlinger.
Seipp, B. (1990). Angst und Leistung in Schule und Hochschule: Eine Me&-Analyse. [Anxiety and
achievement: A meta-analysis]. Frankfurt: Peter Lang.
Seipp, B., & Schwarzer, C. (1987). Anxiety and mathematics achievement: A meta-analysis. In C.
Schwarzer 8c B. Seipp (Eds.), Trends in European educational research. @p. 89-99). Braunschweig:
Braunschweiger Studien zur Erziehungs- und Sozialarbeitswissenschaft.
Spence, K.W., &Taylor, J.A. (1951). Anxiety and strength of the US as determiners of the amount of
eyelid conditioning. Journal of Experimental Psychology, 42, 183-188.
Spence, K.W., & Taylor, J.A. (1953). The relation of conditional response strength to anxiety in
normal, neurotic and psychotic subjects. Journal of Experimental Psychology, 45,265-272.
Spielberger, C.D. (1%). Theory and research on anxiety. In C.D. Spielberger (Ed.), Anxiety and
behaviour. (pp. 3-20). New York: Academic Press.
Spielberger, C.D. (1983). Manual for the State-Trait Anxiety Inuentory (STAI). Palo Alto, CA:
Consulting Psychologists Press.
Taylor, J.A. (1956). Drive theory and manifest anxiety. Psychological Bulletin, 53,303-320.
ANXIETY AND PERFORMANCE 41