The Nature, Effects, and Relief of Mathematics Anxiety

Author(s): Ray Hembree

Reviewed work(s):
Source: Journal for Research in Mathematics Education, Vol. 21, No. 1 (Jan., 1990), pp. 33-46
Published by: National Council of Teachers of Mathematics
Stable URL: http://www.jstor.org/stable/749455 .
Accessed: 02/10/2012 13:43

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

National Council of Teachers of Mathematics is collaborating with JSTOR to digitize, preserve and extend
access to Journal for Research in Mathematics Education.

http://www.jstor.org
Journalfor Researchin MathematicsEducation
1990, Vol. 21, No. 1, 33-46

THE NATURE,EFFECTS,AND RELIEFOF

MATHEMATICSANXIETY

RAY HEMBREE,Adrian College

Resultsof 151studieswereintegrated to scrutinize

bymeta-analysis theconstruct
mathematics
anxiety.Mathematicsanxietyis relatedtopoorperformanceonmathematicsachievement
tests.
It relates inverselyto positive attitudestowardmathematicsand is bounddirectlyto avoidance
of the subject. Variables that exhibit differential mathematicsanxiety levels include ability,
school grade level, and undergraduatefields of study, with preservice arithmetic teachers
especiallyproneto mathematicsanxiety.Femalesdisplayhigherlevelsthanmales.However,
mathematics anxietyappearsmorestronglylinkedwithpoorperformance andavoidanceof
mathematics inprecollegemalesthanfemales.A varietyof treatments
areeffectivein reducing
mathematics anxiety.Improvedmathematics performance consistentlyaccompaniesvalid
treatment.

Anxiety has fostered strong researchconcerns within the last quartercentury

(Endler& Edwards,1982).The constructis broadlydefinedto be a state of emo-
tion underpinnedby qualities of fear and dread (Lewis, 1970 ). This emotion is
unpleasant,is directedtowardthe future,and is out of all proportionto the threat.
Its special characteristicsare "the feelings of uncertaintyand helplessness in the
face of danger"(May, 1977, p. 205). Anxiety is an omnibus construct,and under
its rubricthere has appeareda host of subconstructsthat relate to discrete situ-
ations. In academics,two of these seem prominent:test anxiety and mathematics
anxiety.
To examine the latteranxiety,it will help to look at the former.From its begin-
ning, the researchof test anxiety has proceededon well-definedtheoreticalpaths.
The effect of test anxiety on performancehas remainedthe focal concern.
On the basis of their responses to a Test Anxiety Questionnaire (Sarason &
Mandler,1952),studentsat Yale Universitywere categorizedas high- or low-test-
anxious. In test after test, the low-anxious studentsoutperformedtheir high-anx-
ious peers in both the scores and their variability.Mandlerand Sarason(1952)in-
terpretedthis differentialperformanceon the basis of learnedpsychologicaldrives.
Two kinds were said to be evoked by the test situation. First are task-directed
drives; these stimulatebehaviorsto reduce the drive by completing the test. Sec-
ond arelearnedanxietydrivesthatstimulatetwo behaviors:1) task-relevantefforts
to finish the test and therebyreduce the anxiety; and 2) self-directed,task-irrele-
vant behaviorssuch as heightenedheartbeat,anticipationsof punishmentand loss
of status or esteem, and strong desires to escape the test situation.Persons with
stronganxiety drives arepromptedby habitto reenacttheirtask-irrelevantbehav-
iors that impairperformance.Low-test-anxiouspersonscan more easily attendto
task-directedefforts thatenhance achievement.
Later researchersbuilt on this behavioralfoundationfor the test anxiety con-
struct. Liebert and Morris (1967) proposed that test anxiety consists of two
34 MathematicsAnxiety

components:emotionality,behavioralin nature,and conscious worryor concern,

a cognitive element. Wine (1971)adopteda purely cognitive orientation.Accord-
ing to her attentionaltheory,test-anxiouspersons divide their attentionbetween
task-relevantefforts and preoccupationswith worry,self-criticism, and somatic
concerns. With less attentionavailable for test-directedactivities, their perform-
ance is depressed.
These theories all conceptualizean interferencemodel of test anxiety,in which
test anxiety disturbsthe recall of priorlearning,therebydegradingperformance.
An alternativedeficitsmodel has been proposed(Tobias,1985) that attributesthe
lower scores of test-anxious students to poor study habits and/or deficient test-
takingskills. Withinthis model, test anxietydoes not cause poor performance;the
reverse is true.An awarenessof poor past performancecauses test anxiety.
Conflictswithin this discussion raise questions. Is test anxiety a cognitive con-
struct,or is it behavioral?What is the causal directionin the relationshipbetween
test anxiety and performance?A synthesis of test anxiety research (Hembree,
1988a)foundthe constructmore behavioralthancognitive in nature.It was seen to
cause poorperformance;hence, the evidence supportedthe interferenceratherthan
the deficits test-anxietymodel.
No such theoreticalbase has been constructedfor the researchof mathematics
anxiety (Reyes, 1984).Early in the 1970s,results of its study began to appear,us-
ing the methods,procedures,and treatmentsalreadyappliedto test anxiety.Most
researchers have viewed the two constructs as highly related. Some describe
mathematicsanxiety as no more than subject-specifictest anxiety (Brush, 1981).
Othersdefine its context more broadly,includinga generaldreadof mathematics,
and of tests in particular(Richardson& Woolfolk, 1980). Often, reviews of test
anxietyresearchcontaina section thatexplicitly regardsmathematicsanxiety (for
example,Tryon,1980).A tacit belief has seemed to prevailthattest-anxietytheory
can be used to supportboth constructs.
Despite its lack of independentidentity,the researchof mathematicsanxietyhas
prospered, spurredby increasing perceptions that the construct threatensboth
achievementand participationin mathematics.These suggestions have national
import;when otherwise capable students avoid the study of mathematics,their
options regardingcareersare reduced,erodingthe country'sresourcebase in sci-
ence and technology. Especially,this avoidancehas been thoughtto apply to fe-
males. Their anxiety towardmathematicsseems consistently higher than that of
males (Betz, 1978), and the construct is charged with being a cause of Ernest's
(1976)findingthatfar fewer women thanmen takepartin high school andcollege
mathematics(Meece, Parsons,Kaczala,Goff, & Futterman,1982;Tobias,1980).
In light of these concerns,it seems wise to challenge the priorassumptionsand
develop a scholarlybase for the construct,assuringthat treatmentstowardits re-
lief coincide with its nature. Existing research can be used for this effort. The
presentreview was thus conceived with comprehensiveintentions:to help build
theory, examine effects, compare prescriptionsfor treatment,and describe the
constructas fully as the mass of researchwill allow.
Ray Hembree 35

ProblemStatement
The purposeof the studywas to integratethe findingsof the researchon mathe-
matics anxiety,regardingits nature,effects, and relief. Methods of meta-analysis
were selected for this synthesis, to describe relationshipsand effects with scale-
invariantmetrics. Focus was suppliedby a drive to reduce or resolve theoretical
issues surroundingthe construct.

ResearchTasks
To performthese objectives, five tasks were defined.The first two regardedthe
natureof mathematicsanxiety.Task 1 set out to identify variables that correlate
with the construct(for example, mathematicsperformance,mathematicsavoid-
ance, and test anxiety). Task2 was meant to identify variablesthatexhibit differ-
ent levels of the construct(for example, gender and school grade level). Task 3
regardedthe relationbetween mathematicsanxiety and mathematicsperformance
on the basis of effect size. Task 4 examined treatmentsto reduce mathematics
anxiety,to (a) comparetheirrelativedegreesof mathematicsanxietyreductionand
(b) determineif the treatmentsaffected performance.
Fulfillmentof these tasks would allow for a probingof issues relatedto theory.
Task5 was definedto addressthe following theoreticalquestions,using the results
of Tasks 1 through4:
1. Is therea causaldirectionin the relationshipbetweenmathematicsanxietyand
mathematicsperformance?
2. Does test anxiety subsumemathematicsanxiety?
3. Are behaviorsrelated to mathematicsanxiety more pronouncedin females
than males?

Collecting the Studies

The identificationof studiesto be used in the meta-analysisbeganwith computer
andmanualsearchesof threedatabases:DissertationAbstracts,PsychologicalAb-
stracts, and the EducationalResources InformationCenter(ERIC).Otherstudies
were found by trackingcitations from study to study.All candidatereportswere
screenedagainstthe following criteria:
1. The study reportprovidedproduct-momentcorrelationcoefficients and their
sample sizes or, in the case of experiments,sufficientdata for effect-size calcula-
tions.
2. Mathematicsanxiety measurementswere made with validatedinstruments.
3. Experimentsused at least two groups,includinga control.
4. Each experimentalgroupcontainedat least 10 subjects(for rigorin the meta-
analytic tests of homogeneity).
This screening delivered a body of 151 studies: 49 journal articles, 23 ERIC
documents,75 doctoraldissertations,and 4 reportsin other sources.A bibliogra-
36 Mathematics Anxiety

phy is presentin Hembree(1988b).The distributionof studiesperformedat each

gradelevel 1 through12 andpostsecondary(P) is shown in Table 1. (Some studies
dealt with more than one grade, so the total exceeds 151.) The data were coded
throughguidelinesprovidedby Hembree(1988a).This effortproducedthe follow-
ing totalsaddressingthe tasksof research:428 correlations,85 effects for variables
that exhibited different levels of mathematicsanxiety, 13 effects involving the
relationshipbetweenmathematicsanxietyand mathematicsperformance,and 115
effects of treatingmathematicsanxiety.Treatmentsrangedin length from 3 to 12
hours or greater,with median 8. Design ratings,which evaluatedesign qualityon
a scale of 1 = poor to 3 = excellent, rangedfrom 1.5 to 2.5, with median2.0.
Table 1
Distribution of Grade Levels in 151 Studies of MathematicsAnxiety
Grades 1 2 3 4 5 6 7 8 9 10 11 12 P
Number of studies 0 0 1 0 1 5 12 14 17 21 18 18 122

Coding the Data

The studies differed across a broad range of propertiesand features such as
school grade level, ability levels, and quality of the researchdesigns. Thus, their
findings seemed likely to vary because of these differing characteristics.Meta-
analysis drawsstrengthfrom its capacityto identify interactionsand relationships
among the propertiesof studies and their outcomes. To preparefor this eventual-
ity, studypropertieswere coded as independentvariables,with outcomes(correla-
tions or experimentalcomparisons)treatedas dependentvariables. Interactions
could then be exploredat the time of dataanalysis.

IndependentVariables
Each study was coded regardingpropertiesthat could change from study to
study,including:
"*
Gradelevel (K-12, postsecondary)
"*
Ability level (low or high where indicated,otherwiseaverage)
"*
Socioeconomic status(low, middle,upperwhereindicated,otherwisecomposite)
"*
Ethnicity(predominantethnic or culturalgroup)
"*
Instrumentused for mathematicsanxiety measurement
"*
Lengthof treatment(numberof hoursacross numberof weeks)
"*
Researchdesign quality (1 = poor to 3 = excellent)

DependentVariables
In correlationalstudies, the product-momentcoefficients were directlythe out-
come variable.To guardagainst skew, all values of r were changed to Fisher's z
(Ferguson,1981, p. 194) for their analysis. In experimentalstudies, the criterion
Ray Hembree 37

was effect size as definedby Glass, McGaw,and Smith (1981,p. 102):

ESX -X
ESs= (1)
The numeratortermsareposttestmeans,with s computedas the pooled standard
deviationof the posttestdata,the squareroot of within-cellvariance,or the square
root of errormean square.
Procedures
The totalcollections of datawere partitionedinto subsetsthatrelatedto the tasks
and subtasksof research.Then each subset was analyzed separately,in the hope
that its mean could be used to integratethe subset.A use of the mean was appro-
priateif the subsetwas consistent,as determinedby a test of homogeneity(Hedges
& Olkin, 1985). Whenevera subset was homogeneous, its mean value, weighted
with respect to sample sizes, was declaredthe correlationor the effect for its re-
searchtask.A 99%confidenceintervaltestedthe null hypothesisthatthe meanwas
not significantlydifferentfromzero, andthe synthesisof thatsubsetwas complete.
Whenevera subset was heterogeneous,two conditionswere implied:(a) the pres-
ence of outlier data, or (b) interactions among the independent and dependent
variables.A searchfor the cause was then performed,to find and exclude the out-
lier data or find and describe interactions,using proceduresdetailed by Hembree
and Dessart (1986).Where a cause could not be found, the mean of the heteroge-
neous groupwas offered as a descriptorfrom which no statisticalinferences were
drawn.

RESULTS
CorrelationalFindings
To gatherinsight into the natureof mathematicsanxiety,Task 1 of the study set
out to identify correlatesof the constructand to learn the extent of each relation-
ship. Tables 2-5 present the findings with regard to mathematicsperformance,
studentattitudestowardmathematics,avoidancebehaviors,andmeasuresof other
anxieties. The data regardingeach correlateare summarizedby 1) providingthe
numberof correlationsfor each relationshipplus the total numbern of students
involved, values of outliers (if any), smallest and largest values of r (excluding
outliers),and the gradesinvolved; 2) providingthe groupmean as the correlation
that, in the presenceof homogeneity,describesthe relationship;and 3) indicating
whetheror not the mean was significantlydifferentfrom zero.

Performance Correlates
Table 2 gives mean correlationsbetween mathematicsanxiety and measuresof
studentperformance.Highermathematicsanxiety was slightly relatedto lower IQ
levels, whereas its relationshipwith verbal ability was so low that it was not of
practical importance. Correlationsbetween mathematics anxiety and aptitude/
38 Mathematics Anxiety

achievementmeasures were inverse across grade levels, so higher mathematics

anxiety consistentlyrelatedto lower mathematicsperformance.In Grades5-12,
the inverse relationwas strongerfor males than females, a differencethat disap-
pearedamongcollege students.Gradesin mathematicscourses seemed depressed
in relationto anxiety by aboutthe same proportionas the students'test scores.

Table 2
Mean Correlations of MathematicsAnxiety and Performance
Correlate of Description of correlational group Meana r
mathematics anxiety n Outliers End values Grade level
IQ test 5(449) - -0.23/0.22 6, P -0.17*
Verbalaptlach 17(1941) -0.27/0.05 9-12, P -0.06*
Math apt/ach by grades
5-12:
Males 6(2794 - -0.46/-0.28 5-12 -0.36*
Females 6(2864 - -0.39/- 0.16 5-12 -0.30*
Both genders 7(5555)
- -0.47/- 0.18 7, 8, 11 (-0.34)
College 58(6137) -0.64/-0.04 P -0.31*
Math apt/ach by subtest
Computation 5(957) - -0.43/-0.10 7, 9-12, P -0.25*
Concepts 4(894 - -0.40/-0.13 7, 9-12, P -0.27*
Problem solving 3(871) - 0.42/-0.15 7, 9-12, P - 0.27*
Abstract reasoning 3(325) - 0.43/- 0.29 P -0.40*
Spatial ability 5(374) -0.34/0.21 P - 0.29*
Grade in math course
High school 4(903) - -0.46/-0.27 9-12 -0.30*
College 17(1624) - -0.57/0.02 P (-0.27)
Note. P = postsecondary. apt/ach = aptitude/achievement.
"Entries in parentheses are mean correlations for heterogeneous data.
*p < .01.

AttitudeCorrelates
Table3 presentsmean correlationsbetweenmathematicsanxietyandattitudinal
constructs. Positive attitudes toward mathematicsconsistently related to lower
mathematicsanxiety,with strong inverse relationsobserved for an enjoymentof
mathematicsand self-confidencein the subject. Relationshipsseemed weaker at
postsecondarylevels. Small correlationswere found between mathematicsanxi-
ety and desire for success and a view of mathematicsas male-oriented.
High-anxious studentsviewed parentsand teachers as somewhat negative to-
wardmathematics.These relationstoo were smallerat postsecondarylevels.
AvoidanceBehaviors
Table4 relatesmathematicsanxietywith tendenciesof studentsto avoid mathe-
matics. High-anxiousstudentstook fewer high school mathematicscourses and
showed less intention in high school and college to take more mathematics.A
significantgenderdifferenceappearedinjuniorandseniorhigh school. Males with
higher levels of mathematicsanxiety appearedless likely than high-anxious fe-
males to take more mathematics.
Ray Hembree 39

Table 3
Mean Correlations of MathematicsAnxiety and Attitude-Related Variables
Correlate of Description of correlational group Mean r
mathematics anxiety n Outliers End values Grade levels
Student attitudes
Enjoyment of math
Grades 5-12 6(3856) -0.36 - 0.76/ - 0.63 5-12 - 0.75*
College 9(1383) - - 0.63/ - 0.38 P -0.47*
Self-confidence in math
Grades 6-11 4(514) - -0.85/-0.76 6-11 -0.82*
College 19(2912) - - 0.84/ - 0.37 P - 0.65*
Self-concept in math 6(3748 -0.44 -0.74/ - 0.61 5-8, 11, P -0.71*
Motivation in math 3(2623 - -0.64/-0.57 8,11 -0.64*
Math as male domain - -0.12/0.31 6-12, P 0.14*
19(3678)
Attitude toward:
Success in math 8( 1703 -0.42 - 0.22/0.08 9-12, P -0.12*
Usefulness of math 24(8889 - -0.70/-0.16 5-12, P -0.37*
Problem solving 12(2709 -0.71/-0.38 7-12, P -0.58*
Math teachers 4(2723) -- -
-0.69/ 0.25 7, 8, 11 - 0.46*
Computers 7(1018) - 0.53/ - 0.23 7-12, P - 0.32*
Self-confidence with
computers 5(478) - 0.58/ - 0.39 7-12, P -0.43*
Student perceptions of
others' attitudes
toward math
Father
Grades 9-12 3(575) --0.45/-0.36 9-12 -)0.39*
College 11(1592) - 0.42/-0.06 P -0.25*
Mother
Grades 9-12 3(575) - -0.44/-0.33 9-12 -0.37*
College 12(1643) - -0.51/0.01 P - 0.23*
Math teacher
Grades 9-12 3(575) - -0.55/-0.45 9-12 -0.49*
College
Males 3(313) -0.54/- 0.43 P - 0.47*
Females 7(1178) - -0.68/-0.35 P -0.41*
Note. P = postsecondary.
""< .01.

Table 4
Mean Correlations of MathematicsAnxiety and Student Avoidance Behaviors
Correlate of Description of correlational group Mean r
mathematics anxiety n Outliers End values Grade levels
Extent of high
school math 28(6358) - -0.44/0 12, P -0.31*
Intent to take
more math
Grades 7-12
Males 3(1272) - - 0.45/ - 0.28 7-12 - 0.35*
Females - 0.35/ - 0.19 7-12 - 0.25*
3(1333
College 8 2225) - 0.63/- 0.05 P - 0.32*
Note. P = postsecondary.
"*p< .01.

CorrelationsAmongAnxieties
Table5 displays correlationsbetween mathematicsanxiety and otheranxieties.
Direct relationships were found regarding a general anxiety proneness and its
components, chronic A-Trait and transitoryA-State as defined by Spielberger
40 Mathematics Anxiety

(1972). Mathematicsanxiety related directly to debilitating test anxiety and in-

versely to the anxiety drive that facilitatesperformanceduringtesting (Alpert&
Haber,1960).
Table 5
Mean Correlations of MathematicsAnxiety and Other Anxiety Measures
Correlate of Description of correlational group Mean r
mathematics anxiety n Outliers End values Grade levels
General anxiety 7(1692) 0.80 0.33/0.50 6, 8-10, P 0.35*
Trait anxiety 11 1941) - 0.24/0.54 P 0.38*
State anxiety 4(815) - 0.31/0.52 P 0.42*
Fear of negative
evaluation 4(257) - 0.40/0.48 P 0.44*
Test anxiety 21(3187) 0.78 0.29/0.73 P 0.52*
Worry component 8(1329) - 0.30/0.69 P 0.45*
Emotionality 8(1329) - 0.29/0.72 P 0.46*
Facilitating TA 7(792) - -0.34/-0.15 P -0.28*
Computer anxiety 8 840) - 0.21/0.58 7-12, P 0.39*
Note. P = postsecondary. TA = test anxiety.
*p < .01.

Results for Differentials

The second task of the study set out to comparethe degrees of anxiety within
variables ("differentials")that exhibit different levels of the construct (for ex-
ample, gender,female versus male). Each comparisonwas drawnin terms of an
effect size; thenrelatedgroupswere analyzedthroughthe proceduresgiven before.
Table6 presentsthe resultsfor ability level, sex, and ethnicity.Mathematicsanxi-
ety seemed somewhathigher in slow and averagestudents,but no differencewas
found between these groupings.Females displayed higher levels than males, es-
pecially in college. Studies describingethnic effects were limited to college stu-
dents.No differenceappearedbetween white andblack students,but the Hispanic
group (in two studies) seemed more anxious thanthe otherethnic groups.
Table 6
Mean Effects for Differentials Regarding MathematicsAnxiety
Differentials Description of effect-size (ES) group Mean ES
and comparisons n Outliers End values Grade levels
Ability level
High vs. average 6(1511) - -0.57/-0.04 6, 7, 9-12, P -0.23*
Low vs. average 3(741) - -0.06/0.32 6, 7, P 0.07
Sex, female vs. male
Precollege 22(6299) - -0.02/0.52 5-12 0.19*
College 47(9209) - -0.24/0.77 P 0.31*
Ethnicity
Black vs. white 5(804) - -0.33/0.30 P -0.01
Hispanicvs. white 2(1489) - 0.82/0.83 P 0.82*
Note. P = postsecondary.
*p < .01.

Is mathematicsanxietyrelatedto school gradelevel? How do its levels compare

between courses and respectingcollege majors?These questions were examined
by descriptivemethods.
Ray Hembree 41

School Grade. Figure 1 shows average mathematicsanxiety levels of females

and males in Grades6-P, basedon 10 428 measurementsof the construct.(No data
were available for the earlier grades.) The levels increased throughjunior high
school, peakednearGrades9-10, andleveled off in upperhigh school and college.

MathematicsAnxiety
Level

220

- 6 250 females
200

180
1804 178 males

160

140

6 7-8 9-10 11-12 P

School GradeLevel
Figure 1. Average mathematicsanxiety levels for GradesK-12 and undergraduate.

Courses and Majors. Table 7 shows levels of mathematicsanxiety for various

college courses and majors, along with the numbersof studies and students in-
volved. High levels appearedin remedial mathematicsand declined with more

Table 7
MathematicsAnxiety Level by College Courses and Majors
Anxiety Anxiety
Course n level' Major n level'
Developmental math 12(836) 236.3 Math/science 5(169) 166.5
Elementary
Remedial algebra 11(1028) 206.1 education 25(1835) 219.2
College algebra 9(578) 201.8 Business 4(194) 187.8
Precalculus 5 436) 180.5 Social sciences 5(161) 190.3
Calculus/analytic Health sciences 2 50) 187.5
geometry 10(730) 152.5 Physical sciences 2(54) 149.4
Math for elementary Humanities 5 174) 198.5
teachers 6(420) 243.0
Elementary statistics 5(435) 185.6
Elementary accounting 3(88) 193.8
"Based on the Mathematics Anxiety Rating Scale (MARS) of Richardson and Suinn (1972).
42 Mathematics
Anxiety

advanced study. Mathematics and science majors were predictably low in the
construct.The highest levels occurredfor studentspreparingto teach in elemen-
tary school.
Effects on Performance
Task 3 of the study set out to determinethe relationshipbetween mathematics
anxietyandperformancein termsof effect size. Thirteenstudiesof college mathe-
matics were found, comparingthe test scores of studentswith high and low levels
of mathematics anxiety. Within these studies, all samples were formed before
obtainingscoreson any measure.The meanof the 13 effects was -0.61, so the low-
anxious studentsconsistently scored better.An effect size representsthe number
of pooled standarddeviations between the scores of the two groups being com-
pared.If a pooled standarddeviation of 12 is assumed for scores on a 100-point
scale, the effect size of -0.61 depicts a differenceof about 7 points between the
high-low categories.
Findingsfor MathematicsAnxiety Treatments
Task4 of the study set out to display the effects of treatingmathematicsanxiety
on the level of anxietyand on performance.These effects were measuredby equa-
tion 1, comparingthe posttreatmentscores of treatedversus untreatedstudents.
The datawere thenpartitionedon two dimensions:posttestmeasure(mathematics
anxietylevel, or test performance);andtype of treatment.Classroominterventions
attemptedto relieve mathematics anxiety within whole classes. Reductions in
mathematics anxiety levels were sought through changes in the curriculumor
throughpsychological interventions.Psychological treatmentswere behavioralor
cognitive in nature.Behavioral modes proposedto relieve 'emotionality'toward
mathematics(feelings of dreadand nervousreactions).Cognitivetreatmentswere
set to relieve expressedconcernsor worryaboutthe subject.Cognitive-behavioral
treatments attended to the worry factor but also provided elements to reduce
emotionality.Each subsetof datawas synthesizedthroughthe proceduresgiven for
computingeffect size. Tables8 and 9 presentthe findings.Meaneffects near?0.8
may be consideredfairlylarge,with sizes near?0.2 consideredsmall (Cohen,1977,
pp. 24-27).
Effects on MathematicsAnxiety
Table8 comparesthe end-of-treatmentmathematicsanxietylevels of treatedand
untreatedsubjects.
ClassroomInterventions.Curricularchanges as a meansof reducingmathemat-
ics anxiety included concentratedefforts to improve the students'achievement,
heuristic versus algorithmic instruction, special classwork in microcomputers,
provisionof special equipment(for example, calculators),and special techniques
for presentingmaterial(tutorial,small-group,and self-paced). Such changes did
not seem effective in reducing mathematicsanxiety. Whole-class psychological
treatmentswere also not effective.
Ray Hembree 43

Table 8
Mean Effects of Treatmenton MathematicsAnxiety
Result by Description of effect-size (ES) group
treatment style n Outliers End values Grade levels Mean ES
Classroom intervention
Curricular
change 17(1045) - -0.46/0.48 10, P -0.04
Psychological 8(581) - -0.38/0.18 9-12, P -0.10
Behavioral
SD and others 18(673) - -2.41/- 0.36 9-12, P - 1.04*
Relaxation training 3(80) - -0.62/-0.41 9-12, P -0.48
Cognitive
Group counseling 3(94) - -0.22/0.17 10-12, P -0.03
Restructuring 14(746) - -1.12/0.05 9-12, P -0.51*
Cognitive-behavioral 10(364) - - 1.83/- 0.46 7-12, P - 1.15*
Note. P = postsecondary. SD = systematic desensitization.
"*< .01.

Out-of-ClassPsychological Treatments.The most commonbehavioraltreatment

mode was systematicdesensitization.This technique,along with anxiety manage-
ment training and conditioned inhibition, were highly successful in reducing
mathematicsanxiety levels. Typically,these techniquesused relaxationtrainingas
a component.However, this trainingalone did not seem effective. The cognitive
treatmentof group discussion was also not effective. Cognitive modificationto
restructurefaulty beliefs and build self-confidence in mathematicsproduced a
moderate reduction in mathematics anxiety. Cognitive restructuringcombined
with systematic desensitizationor relaxationtrainingsucceeded in mathematics
anxiety reductionat a level comparableto systematicdesensitizationalone.
Effects on MathematicsPerformance
Table 9 gives mean effects comparingthe post-treatmenttest performanceof
treatedand untreatedsubjects.Treatmentsthatresultedin significantmathematics
anxiety reductionwere accompaniedby significantincreases in mathematicstest
scores.The largestincreasesreferredto the treatmentsprovidingthe largestmathe-
matics anxiety reduction,that is, to behavioralmethods (except relaxation)and
cognitive-behavioraltreatments.The cognitive modificationsthathademphasized
confidencebuildingproducedboth moderatemathematicsanxiety reductionsand
moderateincreasesin test performance.Therewere no performanceimprovements
from the classroom interventions,relaxationtraining,or groupcounseling.
The behavioraltreatments(except relaxation)andcognitive-behavioralmethods
produceda collective mean improvementof 0.57 in test performance.This value
comparedthe averagescores of studentswith low mathematicsanxiety(the treated
students) and students with high mathematics anxiety (the untreated control
groups).Previous comparisonsbetween high- and low-anxious studentshad dis-
played a mean effect of -0.61 (see the previousdiscussionon effects of mathemat-
ics anxiety on performance).Thus, mathematics anxiety reductions by way of
these methodsappearedto be relatedto betterperformanceapproachingthe level
of studentswith low mathematicsanxiety.
44 Mathematics Anxiety

Table 9
Mean Effects of Treatmenton Mathematics TestPerformance
Result by Description of effect-size (ES) group
Mean ES
treatment style n Outliers End values Grade levels
Classroom intervention
Curriculum-related 6(441) - -0.36/0.19 10, P 0.02
Psychological 9(570) - -0.31/1.01 9-12, P 0.03
Behavioral
SD and others 12(517) 0.19/0.94 9-12, P 0.60*
Relaxation training 2(52) -0.17/0.31 P 0.07
Cognitive
Group counseling 2(110) - -0.37/0.04 P -0.07
Restructuring 7(318) - -0.13/1.21 P 0.32*
Cognitive-behavioral 4(142) 0.14/0.84 P 0.50*
Note. P = postsecondary. SD = systematic desensitization.
*p < 01.

CONCLUSIONSAND DISCUSSION
MathematicsAnxiety and Performance Causality. Does mathematicsanxiety
tend to contributeto poor performance?Does a knowledge of poor past perform-
ance induce the anxiety?Or is the relationshipcircular?
Because of the following evidence, it seems thatmathematicsanxiety depresses
performance:
1. Higher achievement consistently accompanies reduction in mathematics
anxiety.
2. Treatmentcan restorethe performanceof formerlyhigh-anxiousstudentsto
the performancelevel associatedwith low mathematicsanxiety.
There is no compelling evidence that poor performancecauses mathematics
anxiety.The construct'srelationswith IQ and abilityseem small (see Tables2 and
6), and special work to enhance students'competencefailed to reducetheir anxi-
ety levels.
MathematicsAnxietyand TestAnxiety.Does test anxiety subsumemathematics
anxiety?A comparisonof the present findings for mathematicsanxiety with the
results of a similar analysis of test anxiety (Hembree,1988a)shows a numberof
parallelproperties:
1. Mathematicsand test anxietiesboth relateto generalanxiety.
2. The differencesin anxiety level regardingstudentability,gender,andethnic-
ity are similarfor both constructs.
3. Both forms affect performancein similarfashion.
4. The constructsrespond to the same treatmentmodes, with best relief from
behavioral-relatedmethods and little result from the cognitive treatment,group
counseling.
5. Improvedperformancerelatesto the relief of both constructs.
Ray Hembree 45

These findings suggest that researchersof mathematicsanxiety have been rea-

sonably prudentin adoptingtest anxiety's theoreticalbase for mathematicsanxi-
ety. Like test anxiety,mathematicsanxiety seems to be a learnedconditionmore
behavioralthan cognitive in nature.However, the observedmean correlationbe-
tween mathematicsanxiety andtest anxiety was 0.52 (Table5), a moderatevalue.
Correctedfor attenuationusing instrumentreliabilitiesnear0.85 (Tryon,1980),an
r of 0.52 increases to 0.61. The correspondingcoefficient of determinationr2 is
0.37; thus, only 37 percent of one construct's variance is predictable from the
varianceof the other.The remaining63 percentmustbe attributedto othersources,
factorsattendingone constructthatareabsentat the other.Hence, it seems unlikely
that mathematicsanxiety is purely restrictedto testing. Rather,the constructap-
pears to comprise a general fear of contact with mathematics,including classes,
homework,and tests.
Gender-RelatedBehaviors. Across all grades, female students report higher
mathematicsanxiety levels thanmales. However,the higherlevels do not seem to
translateinto more depressedperformanceor to greatermathematicsavoidanceon
the partof female students.Indeed,male studentsin high school exhibit stronger
negativebehaviorsin both these regards.This paradoxmay be explainedalongtwo
lines: 1) Females may be more willing thanmales to admittheiranxiety,in which
case theirhigherlevels are no more thana reflectionof societal mores; 2) females
may cope with anxiety better.Whateverthe cause, at precollege levels mathemat-
ics anxiety effects seem more pronouncedin male thanfemale students.

REFERENCES
Alpert, R., & Haber,R. (1960).Anxiety in academicachievementsituations.Journal of Abnormaland
Social Psychology, 61, 207-215.
Betz, N. (1978).Prevalence,distribution,and correlatesof mathanxiety in college students.Journalof
CounselingPsychology, 25, 441-448.
Brush, L. (1981). Some thoughts for teachers on mathematicsanxiety. ArithmeticTeacher, 29(4),
37-39.
Cohen, J. (1977).Statisticalpower analysisfor the behavioralsciences. New York:Academic Press.
Endler,N., & Edwards,J. (1982).Stressand personality.In L. Goldberger& S. Breznitz(Eds.), Hand-
book of stress: Theoreticaland clinical aspects (pp. 36-48). New York:The Free Press.
Ernest,J. (1976).Mathematicsand sex. AmericanMathematicalMonthly,83, 595-614.
Ferguson,G. (1981).Statisticalanalysis in psychology and education.New York:McGraw-Hill.
Glass, G., McGaw,B., & Smith,M. (1981).Meta-analysisin social research.Beverly Hills, CA: Sage.
Hedges, L., & Olkin, 1. (1985).Statisticalmethodsfor meta-analysis.Orlando,FL: Academic Press.
Hembree,R. (1988a).Correlates,causes, effects, and treatmentof test anxiety. Reviewof Educational
Research, 58, 47-77.
Hembree,R. (1988b).Bibliographyof research on mathematicsanxiety.Adrian,MI: AdrianCollege,
Departmentof Mathematics.(ERIC DocumentReproductionService No. ED 293 695)
Hembree,R., & Dessart,D. (1986).Effects of hand-heldcalculatorsin precollege mathematicseduca-
tion: A meta-analysis.Journalfor Research in MathematicsEducation,17, 83-99.
Lewis, A. (1970).The ambiguousword "anxiety."InternationalJournal of Psychiatry,9, 62-79.
Liebert,R., & Morris,L. (1967).Cognitiveandemotionalcomponentsof test anxiety:A distinctionand
some initial data.Psychological Reports,20, 975-978.
46 MathematicsAnxiety

Mandler,G., & Sarason,S. (1952).A study of anxiety and learning.Journal of Abnormaland Social
Psychology, 47, 166-173.
May, R. (1977).The meaningof anxiety.New York:Norton.
Meece, J., Parsons,J., Kaczala,C., Goff, S., & Futterman,R. (1982).Sex differencesin mathachieve-
ment:Towarda model of academicchoice. Psychological Bulletin, 91, 324-348.
Reyes, L. H. (1984).Affective variablesand mathematicseducation.ElementarySchool Journal, 84,
558-581.
Richardson,F., & Suinn, R. (1972). The MathematicsAnxiety Rating Scale: Psychonometric data.
Journal of CounselingPsychology, 19, 551-554.
Richardson,F., & Woolfolk, C. (1980). Mathematicsanxiety. In I. G. Sarason (Ed.), Test anxiety:
Theory,research, and applications(pp. 271-288). Hillsdale, NJ: Erlbaum.
Sarason,S., & Mandler,G. (1952). Some correlatesof test anxiety. Journal of Abnormaland Social
Psychology, 47, 810-817.
Spielberger, C. (1972). Anxiety as an emotional state. In C. D. Spielberger(Ed.), Anxiety: Current
trends in theoryand research (Vol. 1, pp. 23-49). New York:Academic Press.
Tobias, Sheila. (1980).Mathanxiety:Whatyou can do aboutit. Today's Education,69(3), 26-29.
Tobias, Sigmund. (1985). Test anxiety: Interference,defective skills, and cognitive capacity. Educa-
tional Psychologist, 20, 135-142.
Tryon,G. (1980).The measurementandtreatmentof test anxiety.Reviewof EducationalResearch,50,
353-372.
Wine, J. (1971).Test anxiety and directionof attention.Psychological Bulletin, 76, 92-104.

AUTHOR
RAY HEMBREE, Associate Professor of Mathematics, Adrian College, 110 S. Madison Street,
Adrian,MI 49221