You are on page 1of 11

AN ADAPTED ALTERNATING TREATMENTS DESIGN FOR INSTRUCTIONAL RESEARCH

Author(s): Paul T. Sindelar, Michael S. Rosenberg and Richard J. Wilson


Source: Education and Treatment of Children, Vol. 8, No. 1 (Winter, 1985), pp. 67-76
Published by: West Virginia University Press
Stable URL: http://www.jstor.org/stable/42898888
Accessed: 24-02-2016 13:06 UTC

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 support@jstor.org.

West Virginia University Press is collaborating with JSTOR to digitize, preserve and extend access to Education and
Treatment of Children.

http://www.jstor.org

This content downloaded from 131.172.36.29 on Wed, 24 Feb 2016 13:06:38 UTC
All use subject to JSTOR Terms and Conditions
EDUCATION& TREATMENT
OF CHILDREN VOL.8, NO.1, WINTER,1985

AN ADAPTED ALTERNATING TREATMENTS DESIGN


FOR INSTRUCTIONAL RESEARCH

Paul T. Šindelář
The Pennsylvania State University

Michael S. Rosenberg
Ball State University

Richard J. Wilson
Bowling Green State University

ABSTRACT
Intraditional
instructional twoormore
research, methods
teaching areusedwithseparatebut
equivalent andstatistical
groups ofgroup
analysis performance
permits aboutthe
conclusions
relative
effectiveness
of thetreatments. AppliedBehavior
Traditional Analysis
designs,
however,donotlendthemselvestothecomparisonoftwoormore Inthispaper,
treatments.
anadapted treatments
alternating (AATD)ispresented
design towithdrawal,
asanalternative
andmultiple
baseline,
multiple forinstructional
probedesigns research.TheAATDrequires
thattwoequivalent
setsofinstructional
items with
thateachsetbetaught
bedeveloped, a dif-
ferent andthataffects
methodology, indifferential
aresought rates Theadvan-
ofacquisition.
oftheAATDarediscussed
tagesanddisadvantages alongwith totheoriginal
itsrelationship
ATD.
★★★

Researchersusing applied behavior analysis (ABA) methodologyhave


recentlymade major contributionsto the field of instructionalresearch
(Greer, 1983), an area heretoforedominatedby researchersusing tradi-
tional group designs and statisticalanalyses. Unlike group design and
methodology,applied behavior analysis methodologyfocuses upon the
behaviorof individualstudents.However,the mostcommonlyused ABA
designspresenta major shortcomingto instructional researchers:They do
not readily allow for the comparison of two or more instructional
approaches.
For example, in the withdrawal(or A-B-A-B) design,the effectsof a
singletreatmentare contrastedto the effectsof baselineby alternatingex-
perimentaland baselineconditions.The demonstration of controlin an A-
B-A-B design reliesupon the premisethat the behaviorunder studywill
revertto baselinelevel followingthe initialinterventionphase. In a sense,
thecontrolthattheintervention exertsoverthebehavioris demonstrated at
least twiceand our confidencein attributing changesin behaviorto the in-
troductionof theintervention is strengthenedbydirectreplicationof effect.
Unfortunately, the responsesunderinvestigationin instructionalresearch
are notlikelyto returnto baselinelevelsfollowingan initialteachingcondi-

Pages 67-76

This content downloaded from 131.172.36.29 on Wed, 24 Feb 2016 13:06:38 UTC
All use subject to JSTOR Terms and Conditions
68 WILSON
ROSENBURG,
SINDELAR,

of learnedacademic responsesis the major limita-


tion. This irreversibility
tion of the A-B-A-B designin instructionalresearch.Academic skillsare
unlikefreeoperantresponsesin thatonce a skillis establishedin an initial
teachingcondition,itcannotbe expectedto revertto pre-instructional levels
in subsequentbaselines.
of academicresponses,there
In additionto theissue of theirreversibility
is a second limitationto the use of A-B-A-B designs in instructional
research.Hersenand Barlow (1976) point out thattwo treatments may be
evaluated in an A-B-A-C-A design,in whichbaseline conditions(A) are-
alternatedwithtwo intervention conditions(B and C). This designallows
for the assessmentof the effectsof both interventions over baseline, but
does notallow fortheassessmentof thedifferential of thetwo
effectiveness
interventions. Extendingthe conditionsof thisdesignto A-B-A-C-A-C-A-
B-A does not resolve this limitationand creates a cumbersomeand ex-
cessivelylong experiment.
Multiplebaseline and multipleprobe designsavoid the problemthat ir-
reversibilitypresents in the A-B-A-B design since interventionis not
withdrawn.Instead, functionalcontrol is demonstratedwhen behavior
change occurs withthe sequentialintroductionof an intervention across
severalbaselinesrepresenting separatebehaviors,subjects,or settings.(In
the multiplebaseline design, continuous observationsare made during
baseline; in the multipleprobe design, baseline observationsare made
periodically.)Both designshave been used to demonstratethe effectgsof
single instructionalprocedures(e.g., Kirby,Holburn, & Bushby, 1981).
However,withbothdesigns,thecomparisonof twointerventions willresult
in the confoundingof intervention witheitherbehaviors,subjects,or set-
tings- whichevervariableis used to establishthemultiplebaselines.Thus,
traditionalgroup designmay be preferredto thesetwo designs(as well as
theA-B-A-Bdesign)sincetheformerpermitsthecomparisonof morethan
one experimentalintervention withinthe same experiment.
The purposeof thispaper is to describeand discussand adaptationof the
alternatingtreatmentsdesignthat overcomesthe limitationsof A-B-A-B,
multiplebaseline, and multipleprobe designsand extendsthe abilityof
ABA researchersto address comparativeinstructionalquestions. In fact,
withthisdesign,educationalresearchersare affordedthe best of both the
single-subjectand group designmethodologies:the individualanalysisof
behavior and the abilityto compare two or more instructionalinterven-
tions.What followsare descriptionsof boththeoriginaland adapted alter-
natingtreatmentsdesigns as well as a discussionof the advantages and
limitationsof the adapted designforinstructional resarch.

TreatmentsDesign
Alternating

TreatmentsDesign(ATD) was proposedby Barlow


The termAlternating

This content downloaded from 131.172.36.29 on Wed, 24 Feb 2016 13:06:38 UTC
All use subject to JSTOR Terms and Conditions
DESIGN
TREATMENTS
ADAPTEDALTERNATING 69

and Hayes (1979) because severaldifferent termshave been used to referto


thesame design.For example,Sulzer-Azaroff and Mayer(1977) used multi-
elementdesign,borrowingsomewhatinappropriately fromSidman (1960).
Kazdin and Hartmann(1978) referredto this design as the simultaneous
treatment design(eventhoughthistermrefersto a designin whicha choice
of treatments is continuouslyavailableto thesubject).For clarity'ssake, we
willadopt theterminology proposedby Barlow and Hayes (1979) and refer
to it as theAlternating TreatmentsDesign. In our discussion,however,we
will cite considerationsmade in all threepapers.
In an ATD, a singlebaselineof behavioris followedby an experimental
conditionin which two or more interventions (or an intervention and a
baseline) are rapidlyalternated.In theirdescriptionof the simultaneous
treatmentdesign, Kazdin and Hartmann(1978) noted that a second ex-
perimentalconditionshouldbe implemented once responsesstabilizeunder
theseparatealternating interventionsof thefirstexperimental condition.In
this second condition,the more successfulof the two interventions is im-
plementedin both sessions. Experimentalcontrolis demonstratedby dif-
ferentialrespondingduringthe firstexperimentalconditionand uniform
respondingconsistentwiththeinitialsuccessfulperformance in thesecond.
Kazdin and Hartmannalso emphasizedtheimportanceof counterbalanc-
ing to thelogic of the design:All relevant,nonexperimental variablessuch
as timeof administration, agents,and situationsmustbe controlled.For ex-
ample,if proceduresare administeredat two or moretimesof theday, it is
criticalto avoid confoundingtime of administrationwith interventions.
Such confoundingsare avoided by counterbalancing the administration of
the two interventions across the time periods duringwhich theyare ad-
ministered.In thecase of a morningand afternoonperiodin whichtwo in-
terventionsare compared, it is imperativethat both be administeredas
oftenin the morningas in the afternoon.The same logic applies whentwo
or moreagentsadministertheinterventions and whentheinterventions are
administeredin two or more settings.When agents,times,or situations
serveas theexperimental variable,and two or moreinterventions are com-
pared, it is necessaryto counterbalance the order of this variable as well.
The advantagesof the ATD designare considerable.For one thing,the
ATD allows for the comparisonof two or more interventions in a single
subject.By usingtheATD, a researcheravoids theproblemof sequenceef-
fectsinherentin comparingtwo interventions withan elaborationof theA-
B-A-B design.The designis applicableto behaviorsthatcannotbe expected
to reverse or for complex behaviors that can be broken down into
equivalentsubsetsof responses.Finally,thereis an efficiency to thedesign
whencomparedto A-B-A-B,multiplebaseline,and multipleprobedesigns.
Relativeeffectsof two interventions are readilyand rapidlyapparentand,
whenthe experimentis terminatedprematurely, some usable data are still
obtained.

This content downloaded from 131.172.36.29 on Wed, 24 Feb 2016 13:06:38 UTC
All use subject to JSTOR Terms and Conditions
70 SINDELAR, WILSON
ROSENBURG,

The ATD also presentscertainlimitations.While counterbalancingthe


order of interventions may controlsequentialconfounding,carryoveref-
fectsare a more criticalproblem.Carryovereffectsare definedby Barlow
and Hayes (1979) as the influenceof one intervention on the effectsof
anotherintervention and have been characterizedbytheseauthorsas occur-
ringthrougheithercontrastor induction.Contrastrefersto how one in-
terventioncontrastingwith anothercan produce behavioral change in a
directionopposite fromwhat is expected. Induction,on the otherhand,
refersto a positivetransfer betweenthetwointerventions. In bothcases, the
effectsof the interventions administeredin rapid alternationmay be dif-
ferentfromthe effectsof the interventions whenadministeredalone. (See
Barlow and Hayes forsuggestionsforcontrollingthesephenomena.)
The finallimitationof the ATD involvesthe assurancethatthe two in-
tervention conditionsare discriminable to thesubject.This considerationis
especiallyimportantwhen the interventioninvolves some manipulation
followinga low frequencybehavior (but less importantin instrutional
research).As a resultof the frequentchangesin the administration of the
it
interventions, may be difficult forthe subjectto determine which of two
conditionsare prevailingat a givenpoint in time. Kazdin and Hartmann
(1978) suggest providing instructionsas a means for heighteningthe
discriminability of the two conditions.
The advantagesof the ATD make it an especiallyusefuldesign for in-
structionalresearch.First and foremost,two or more instructionalap-
proachesmaybe compared.Furthermore, thelogicof theATD does not re-
quire that performance revertto preinstructionallevelswhenteachingends.
Consequently, the ATD (and the Multiple Baseline and Multiple Probe
designs) are better suited for instructionalresearch than the A-B-A-B
design.Finally, academic skillsare typicallycomplex behaviors - as in the
cases of solvinglong divisionproblemsor writinganswersto comprehen-
sion questions- and complexbehaviorsare well-suitedto the ATD.

An Adaptationof the ATD for InstructionalResearch

The Adapted ATD (AATD) differsfromthe standardATD in the sense


thateach intervention is associatedwitha unique set of instructionalitems
(althoughall of theitemsare membersof thesame responseclass). The ma-
jor featureof the AATD involvesthe identification of two equivalentand
functionallyindependentinstructionalsets. A baseline in which the
equivalenceof performanceon thetwo setsis demonstrated is followedby
an experimentalconditionin which acquisitionof one set taughtby one
methodis comparedto the acquisitionof othe second set taughtby the se-
cond method.Differencesare demonstratedwhenacquisitionof one set is
morerapidthanacquisitionof theother,and theeffectis consistentacross
subjects,settings,or behaviors.

This content downloaded from 131.172.36.29 on Wed, 24 Feb 2016 13:06:38 UTC
All use subject to JSTOR Terms and Conditions
ADAPTEDALTERNATING
TREATMENTS
DESIGN 71

a
s
a
4J
S
O
L.
a
M
'S
M
£
C/2
T3
fi
es
fi
H
«4M
O
a
#o
*■¿3
"S
•mm
s
a
«<
fh
S
S
.2?
£

This content downloaded from 131.172.36.29 on Wed, 24 Feb 2016 13:06:38 UTC
All use subject to JSTOR Terms and Conditions
72 SINDELAR, WILSON
ROSENBURG,

Mandelker,Sindelar,and O'Shea (1984) used thisadaptaion in a recent


study.Arithmetic factsfromtwo instructional objectives(forexample,+7
and +8 facts)weredividedlogicallyintotwo setswithpresumedequal dif-
ficulty.One set was taughtdirectlyby the teacher; the second througha
seriesof programmedworksheetsthatthechildrencompletedindependent-
ly. Each day followingthe lessons,two measuresof performance wereob-
tained, one on each instructional set. The correct and errorrates for one
child on the two measuresare presentedin Figure 1. The open circlesand
trianglesrepresentcorrectand errorperformanceon the independentseat-
work set; the black circlesand trianglesrepresentperformanceon the set
taughtbytheteacher.The data showan effectin favorof theteacher-ledin-
structionprogram, although replicationis required for making more
generalstatementsabout the two interventions.
The need forcounterbalancing the timeof administration, the teachers,
and thesettingswhenthesevariablesdifferis the same for theAATD as for
the originalATD. In the Mandelker et al. study,only the time of ad-
ministration was of concern,because the teacherand the setting did not
changeat any timethroughoutthe study.The authorscounterbalancedthe
orderof thetwo instructional sessionsso thatstudentsbegan lessonswith
seatworkas oftenas theybegan withteacher-directed instruction.
An importantconsiderationin theAATD is theequivalenceof thetwo in-
structionalsets. The logic of the designrestsupon the constructionof two
sets of itemsthatwill be equally difficultto learn. Matson and Ollendick
(1982) describeda procedurein whichitemsof equal difficulty are random-
ly assignedto stimuluscohorts. For theirrandom stimulus design,thispro-
cedurewould sufficesincethesame treatment is used withvariouscohorts.
However,whentwo or more treatments are compared,the equivalenceof
the instructionalsets mustbe demonstrated.
Three activities are essential to a satisfactorydemonstrationof
equivalence: a logical analysis of the universeof items,considerationof
baselineperformance, and post hoc verification.In Mandelkeret al. (1984)
only logicalanalysisof theitemsand considerationof theinitialbaseline
the
rates were used to determineequivalence. Bursuck, Sindelar, Halle,
Cumblad, and Imlay(1983), in a studyof sightwordacquisition,conducted
an additional post hoc analysis of the equivalenceof the two sets. They
testedmorethan 50 thirdgrade childrenon theirabilityto read the words
fromthe two instructionalsets. The resultsshowed thatthe thirdgraders
read both lists with equal proficiency,thus substantiatingthe logical
analysisand the equivalenceof the student'sinitialperformances.
In the second experimentalconditionadvocated by Kazdin and Hart-
mann(1978), themoreeffective intervention fromthecomparativephase is
implementedin both sessions.This conditionhas limitedusefeulnessin in-
structionalresearch. To illustratethis point, consider an experimentin
which the acquisition of one instructionalset (Set A) associated with

This content downloaded from 131.172.36.29 on Wed, 24 Feb 2016 13:06:38 UTC
All use subject to JSTOR Terms and Conditions
ADAPTEDALTERNATING
TREATMENTS
DESIGN 73

< CD
h- h-
UJ LJ
CO CO
v I -J
V ' < <
' ' 2 2
' o O
'* '' H H
O CJ
I ' 3 3
'' ' cr oc
' h- h-
' co co
' - -
^ '
' ' o 2
LU ' ° O
'
. '' UJ uj
< 'x o o
X Z 2
û_ < <
S S
1 CE CE
O O
Ll_ li.
'' 'I K *
LU LJ

:i

en ''
'
1 v
CL

I
LU I
i I
ÜJ I1
co
< I1

Figure2. Subtlechangein trendthatresultsfromadoptingsingletreatment


in second phase of AATD.

This content downloaded from 131.172.36.29 on Wed, 24 Feb 2016 13:06:38 UTC
All use subject to JSTOR Terms and Conditions
74 WILSON
ROSENBURG,
SINDELAR,

MethodA is demonstrated to be superiorto theacquisitionof a second in-


structionalset (Set B) associated withthe Method B. This outcome is il-
lustratedin a generalwayin theBaselineand Phase 1 conditionsin Figure2.
Had a second experimentalconditionbeen introducedin whichboth sets
weretaughtwithMethodA, performance on Set B itemswould not change
immediately to approximatephase 1 performance on Set A. Rathertherate
of acquisitionof Set B would accelerateto approximatetherateof acquisi-
tionof Set A in phase 1. As illustratedin phase 2 in Figure2, thischangeis
slight.Furthermore, performanceon Set B itemswould remainbelow per-
formanceon Set A items,even thoughthe two sets were acquired at the
same rate in this finalcondition.
The limitationsthataffectthe interpretation of thisAATD overlapwith
the limitationsof the originalATD. Multipletreatmentinterference is a
potentialproblem.Similarly,among the carryovereffects, inductionseems
especiallypronouncedwiththeadapted design.An assessmentof induction
across the instructionalsets would requiresamplingof performanceon a
thirdequivalent set on which no instructionoccurs. This procedureis
describedmore fullybelow.
Of course,the AATD is not the designof choice forall questionsin the
domainof instructional research.TraditionalABA designsare preferred to
the AATD whenevera singletreatmentis investigated, because theyallow
for the demonstrationof functionalcontrolover performance.That is,
changesin performanceare reliablyand repeatedlyassociatedwiththe in-
troductionof theindependentvariable.However,in theAATD, treatments
are introducedonce. Independentvariablesare confoundedwithhistory,
maturation,and otherthreatsto the internalvalidityof the investigation.
Replicationallows foran inductivedemonstration of functionalcontrol
and is an essentialaspectof researchwiththeAATD. For example,Bursuck
et. al. (1983) replicatedtheircomparisonof unisonand orderedresponding
withthreegroups. Their design can be constructedas a combinationof
AATD and multiplebaselinedesigns,providedthatwe acceptan extension
of thebaselinebackwardin time.This assumptionseemsreasonablein this
case because the studentsknewnone of the wordsin the instructional sets
whenthe studybegan. However,replicationis not theonlyway to increase
our confidencein the resultsof studiesusingthe AATD.
For one thing,researcherscan choose to compare (and in all likelihood
do compare) two interventionsfor which functionalcontrol has been
establishedin previousresearch.In thiscase, theissueof functionalcontrol
is secondaryto a considerationof the differential of the two
effectiveness
interventions.Second, certainattributesof academic performanceslend
more power to the AATD than would otherwisebe the case. Specifically,
performance of academicskillsdoes notreadilychangein theabsenceof in-
struction.Consequently,changesin academicperformance thatoccur with
theintroduction of instructioncannotbe explainedconvincintly bymatura-

This content downloaded from 131.172.36.29 on Wed, 24 Feb 2016 13:06:38 UTC
All use subject to JSTOR Terms and Conditions
ADAPTEDALTERNATING
TREATMENTS
DESIGN 75

tion, history,or otherrivalhypotheses.


Nonetheless,the abilityof theAATD to controlforthesethreatscan be
enhancedby addinga third,equally difficultinstructional set on whichno
instructionoccurs. Performanceon thisthirdset could be assessed in the
same wayand at thesame timeas performance on thetwoinstructional sets.
In a sense,baselineis extendedbysamplingperformance on thenoninstruc-
tionalset duringtheintervention phase of theAATD. Differencesbetween
the performanceson the two instructionalsets, on the one hand, and the
noninstructional set, on the other, can strengthen our confidencein at-
tributing theseeffectsto the introductionof the interventions.
Furthermore, by includinga third,noninstructional set, researchersmay
assess the degreeof inductionamong the sets and, hence,theirfunctional
independence.Ideally,performanceon the noninstructional set would not
change when instruction was begun on the items in the two instructional
sets. Otherwise,the comparisonof the two treatments would be clouded.
Manipulationof instructionalvariablescannot be expectedto produce
rapid and dramaticchangesin performance.Because changesare slow to
occur,theefficiency of theATD is lostto a degree.Compoundingthispro-
blemis thesubstantialdegreeof intrasubjectvariability in academicrespon-
ding. Highly variable performancerequireslongerexperimental conditions
beforetheeffectsof thetreatments stabilize.However,theuse of unknown
instructionalitems potentiallyalleviates this problem. Under these cir-
cumstances,baselinesare stable at zero; even highlyvariableperformance
in an intervention conditioncan be discriminated readilyfromstable,zero-
level baselines. The highlyvariable performance,then,presentsproblems
only in the discrimination of treatmenteffects.
In conclusion,all of the advantagesof the originalATD hold for the
adaptationwe have described.In addition,theissue of the discriminability
of intervention conditionsis no longercritical.Instructionoccursindepen-
dent of the children'sbehavior; the interventions (for example, seatwork
versusdirectinstruction)are readilydiscriminableto them.Furthermore,
thedesignhas considerablegenerality;it has been used to studytheacquisi-
tionof suchdiverseskillsas additionand multiplication facts(see also Har-
ris & Sherman, 1973), spelling(Ollendick, Matson, Esveldt-Dawson,&
Shapiro, 1980) as well as readingsightwordvocabularies.Any universeof
itemsthat can be identifiedand divided into two equivalentsets is ap-
propriateto studywiththe AATD.
On a practicalside, the designis simpleenough to be used by teachers
duringtheirday-to-dayinstruction.Littledisruptionin usual instructional
proceduresresultsand, when permanentproductsare used for measure-
ment, the task of obtaining a daily sample of performanceis greatly
simplified.Finally,componentanalysesof instructionalprogramscan be
conductedusingthe AATD, and the effectsof adding additionalinstruc-
tional procedurescan be assessed in muchthe same say.

This content downloaded from 131.172.36.29 on Wed, 24 Feb 2016 13:06:38 UTC
All use subject to JSTOR Terms and Conditions
76 SINDELAR, WILSON
ROSENBURG,

REFERENCES

Barlow, D.H., & Hayes, S.C. (1979). Alternatingtreatments design: One


for
strategy comparing the effects
of two treatments in a singlesubject.
Journalof Applied BehaviorAnalysis,12, 199-210.
Bursuck,W.D., Sindelar,P. T., Halle, J.W., Cumblad, C., & Imlay, D.
(1983, May). The effectsof two variationsof teacherquestioningon stu-
dentbehaviors. Paper presentedat theannual meetingof theAssociation
forBehaviorAnalysis,Milwaukee,Wisconsin.
Greer,R.D. (1983) Contingenciesof thescienceand technologyof teaching
and pre-behavioristicresearch practices in education. Educational
Researcher,72(1), 3-9.
Harris, V.W., & Sherman, J.A. (1973) Use and analysis of the "good
behaviorgame" to reducedisruptiveclassroombehavior.Journalof Ap-
plied BehaviorAnalysis,6, 405-417.
Hersen, M., & Barlow, D.H. (1976) Single case experimentaldesigns:
Strategiesfor studyingbehaviorchange. New York: Pergamon.
Kazdin, A.E., & Hartmann, D. P. (1978) The simultaneoustreatment
design.Behavior Therapy,9, 912-922.
Kirby,K.C., Holborn, S.W., & Bushby,H.T. (1981) Word game bingo: A
behavioraltreatmentpackage forimprovingtextualrespondingto sight
words. Journalof Applied BehaviorAnalysis,14, 317-326.
Mandelker,A.V., Sindelar, P. T., & O'Shea, L.J. (1984) The effectsof
teacher-directed instructionand seatwork on the acquisition of basic
mathfacts. Manuscriptsubmittedforpublication.
Matson,J.L., & Ollendick,T. H. (1982) The randomstimulusdesign.Child
Behavior Therapy,5(4), 69-75.
Ollendick, T.H., Matson, J.L., Esveldt-Dawson,K., & Shapiro, E. S.
(1980) Increasingspellingachievement:An analysis of treatmentpro-
cedures utilizingan alternatingtreatmentsdesign. Journalof Applied
BehaviorAnalysis,13, 645-654.
Sidman, M. (1960) Tactics of scientificresearch:Evaluatingexperimental
data in psychology . New York: Basic Books.
Sulzer-Azaroff, B., & Mayer,G.R. (1977) Applyingbehavior-analysis pro-
cedures with children and youth. New York: Holt, Rinehart and
Winston.

This content downloaded from 131.172.36.29 on Wed, 24 Feb 2016 13:06:38 UTC
All use subject to JSTOR Terms and Conditions

You might also like