Emotional arousal does not enhance association-memory

22 Pages·2012·0.54 MB·English

by  Christopher R. Madan

JournalofMemoryandLanguage66(2012)695–716
ContentslistsavailableatSciVerseScienceDirect
Journal of Memory and Language
journal homepage: www.elsevier.com/locate/jml
Emotional arousal does not enhance association-memory
Christopher R. Madana,⇑, Jeremy B. Caplana,b, Christine S.M. Laua,c, Esther Fujiwarab,c
aDepartmentofPsychology,UniversityofAlberta,Edmonton,AB,Canada
bDepartmentofCentreforNeuroscience,UniversityofAlberta,Edmonton,AB,Canada
cDepartmentofPsychiatry,UniversityofAlberta,Edmonton,AB,Canada
a r t i c l e i n f o a b s t r a c t
Articlehistory: Emotionally arousing information is remembered better than neutral information. This
Received4July2011 enhancementeffecthasbeenshownformemoryforitems.Incontrast,studiesofassocia-
revisionreceived30March2012 tion-memoryhavefoundbothimpairmentsandenhancementsofassociation-memoryby
Availableonline28April2012
arousal. We aimed to resolve these conflicting results by using a cued-recall paradigm
combinedwithamodel-baseddataanalysismethod(Madan,Glaholt,&Caplan,2010)that
Keywords: simultaneouslyobtainsseparateestimatesofarousaleffectsonmemoryforassociations
Arousal
and memory for items. Participants studied sequentially presented words in pairs that
Emotion
werepure(NEGATIVE–NEGATIVEorNEUTRAL–NEUTRAL)ormixed(NEGATIVE–NEUTRAL
Association-memory
or NEUTRAL–NEGATIVE). Cued recall tests had NEUTRAL or NEGATIVE probes and
Taboo
Paired-associatelearning NEUTRALorNEGATIVEtargets.Wefoundimpairedmemoryforassociationsinvolvingneg-
Cuedrecall ative words despite enhanced item-memory (more retrievable targets). A category-list
controlconditionexplainedawaytheitem-memoryenhancementbutcouldnotexplain
theimpairmentofassociation-memoryduetoarousal.Asecondexperimentwithidentical
structurebutusinghigher-arousingtaboowordsrevealedincreasedcuedrecalloftaboo
thanneutralwords.However,thiswasexclusivelymediatedbyitem-memoryeffectswith
neitherenhancementnorimpairmentofassociation-memory.Thus,cuedrecallwaslower
forpurenegativepairsandhigherforpuretaboopairs,butourmodelingapproachdeter-
mined a different locus of action for these memory impairing or increasing effects:
Althoughitemmemorywasincreasedbyarousal,association-memory wasimpairedby
negativewordsandunaffectedbytaboowords.Ourresultssuggestthatpreviousresults
reportinganenhancementofassociation-memoryduetoarousalmayhaveinsteadbeen
solelydrivenbyenhanceditem-memory.
(cid:2)2012ElsevierInc.Allrightsreserved.
1.Introduction associationitself.Single-wordpropertiescanactatanyof
theselevels.Werecentlydemonstratedthiswiththeword
Cuedrecallisoftenusedasatestofassociation-memory. properties of imageability and word frequency (Madan,
However,cuedrecallrequiresonenotonlytorememberthe Glaholt,&Caplan,2010).Briefly,imageabilityisameasure
relationshipbetweenitems(association-memory)butalso ofhowconduciveawordistomentalimagery(e.g.,high-
the items themselves (item-memory). Thus, verbal cued imageability: BIKE; low-imageability: CLAIM), whereas
recallisinfluencedbyitem-memoryfortheprobeandtarget word frequency is a measure of the probability of occur-
wordsin additionto(oreveninsteadof) memoryforthe rences of a word (e.g., high-frequency: AREA; low-fre-
quency: MICA). We found that in both manipulations,
cued recallaccuracywas betterforpurehighpairs(pairs
⇑ Correspondingauthor.Address:DepartmentofPsychology,Biological in which both words were high-imageability or high-
SciencesBuilding,UniversityofAlberta,Edmonton,Alberta,CanadaT6G frequency, respectively) than for pure low pairs (pairs in
2E9.Fax:[phone]. whichbothwordswerelow-imageabilityorlow-frequency,
E-mailaddress:[email protected](C.R.Madan).
0749-596X/$-seefrontmatter(cid:2)2012ElsevierInc.Allrightsreserved.
http://dx.doi.org/10.1016/j.jml.2012.04.001
696 C.R.Madanetal./JournalofMemoryandLanguage66(2012)695–716
respectively).However,in‘‘mixed’’pairs,whichconsistedof & Shimamura, 2001; Kensinger & Corkin, 2003, but see
onehighitemandonelowitem,wefounddifferenteffects Davidson,McFarland,&Glisky,2006).Ifthetwoelements
foreachwordproperty.Intheimageabilitymanipulation, do not coincide, association-memory appears to be re-
accuracy for the mixed pairs was symmetrical regardless ducedduetoarousal(e.g.,peripheralneutrallinedrawings
whetherhigh/lowwordswereprobesortargetsandfellbe- in the presence of central arousing scene pictures;
tweenaccuracyforthepurehighandpurelowpairs.Inthe Touryan,Marian,&Shimamura,2007).Thesefindingshave
wordfrequencymanipulation,accuracyforthemixedpairs beenderivedfromparadigmswithanumberofpotentially
wasasymmetric,withbetterperformancewhenthecued problematiccharacteristics.Almostallpreviousstudiesin
recall target was a high-frequency word than if it was a thisareausedincidentalencodinginstructionsanddissim-
low-frequencyword.However,evenbyincludingallpossi- ilar items to-be-bound in memory. Under these circum-
bletypesofpairs,cuedrecallaccuracyisstilldependenton stances, several ill-controlled factors could account for a
bothitem-andassociation-memoryprocesses.Tosystem- differential formation of associations with arousing com-
atically disentangle the influences of item- and associa- pared to non-arousing items in memory. For example,
tion-memory producing these patterns of results across when presenting an arousing and neutral item together,
manipulations, we then applied a model-based approach. attentionwilllikelybedrawntothearousingitemifthere
Thisapproachsimultaneouslyobtainsformalestimatesof isnoexplicitinstructiontoencodetheassociationbetween
the influence of single-item properties (i.e., imageability thetwo.Thismaytheninfluencelatermemoryfortheneu-
and word frequency) on item- and association-memory. tralitem.Suchimpairmentmaynotreflectimpairedbind-
By including all possible probe and target combinations ing of elements in memory per se, but rather an effect of
(i.e., high–high, high–low, low–high, low–low), we were unequalattentiontothetwocomponents,whichmaythen
abletoobtainestimatesofthecuedrecalltarget’sretriev- impairencodingoftherelationbetweenthem.Anunequal
ability(modelparameter:t),theprobeword’seffectiveness attentiondistributioncouldbefurtheramplifiedbyusing
(modelparameter:p),andstrengthofthememoryforthe dissimilar assemblies of central arousing and peripheral
relationship between the two items (model parameters: neutralelements.Fourpublishedstudiesavoidedsomeof
r ,reflectinganydifferenceinmemorybetweenhigh–high these ambiguities by using intentional encoding instruc-
1
andmixedpairs,andr ,reflectinganydifferenceinmemory tions with pairs of items that were of the same kind
2
betweenmixedandlow–lowpairs.)Bothrelationshipmod- (word–word pairs). To test memory for associations, two
elparametersdependonlyonpairtypebutnotuponwhich of them used associative recognition at retrieval (Onoda,
itemwasusedastheprobeortargetincuedrecall.Thus,one Okamoto, & Yamawaki, 2009; Pierce & Kensinger, 2011),
could imagine that memorizing pairs of dissimilar items andtwousedcuedrecall(Guillet&Arndt,2009;Zimmer-
(i.e., mixed pairs) is more challenging than memorizing man & Kelley, 2010). In Onoda et al. (2009), participants
pairsofsimilaritems.Ourmodelcantestthen,ifandhow learned pairs of words that were either both negative or
the itemmanipulation(e.g., high/lowimageability) inter- both neutral. Subsequently, on associative recognition
actswithsuchbasicdifferencesbetweenpureandmixed tests,wordswereeitherpresentedwiththeiroriginalasso-
pairrecall.Foreachoftheseparameters,avaluegreaterthan ciate (‘‘intact’’), or with a word from another pair (‘‘rear-
onesignifiesthattheprocessisenhancedduetothemanip- ranged’’). Participants were more accurate on neutral
ulation,whereasavaluebelowonesignifiesanimpairment thannegativepairs.PierceandKensinger(2011)presented
ofthatprocessduetothemanipulation.(Formoredetailon participants with pairs of words that consisted of either
ourmodelingapproach,seepage17.)Despitefindingnodif- two negative words, two neutral words, or two positive
ferencebetweenwordfrequencyandimageabilityincued words. When tested immediately, memory for negative
recall performance in pure pairs, with our modeling ap- pairswasworsethanforneutralpairs,aswellasforposi-
proach we found that imageability primarily enhanced tivepairs(Pierce&Kensinger,2011Experiment1).Whena
association-memory(r andr >1)andthatwordfrequency longerstudy–testdelaywasused,memorywasenhanced
1 2
primarilyenhancedtargetretrievability(t>1;alsoreported fornegativecomparedtoneutral/positivepairs(seePierce
byCriss,Aue,&Smith,2011).Wethereforeextendtheuseof &Kensinger,2011Experiment2).
this cued recall paradigm and modeling approach to an Arguably,associativerecognitionisamoredirecttestof
importanttopicofmemoryresearch:theinfluenceofarou- association-memory than cued recall, and that it is unaf-
salonassociation-memory. fected by item-specific information. However, recent evi-
Previousresearchhasshownthatemotionalarousalen- dence has shown that associative recognition is in fact
hances memory for single items or events, a finding that alsosusceptibletomanipulationsofitemstrength(Buchler,
has been replicated with many different paradigms and Light, & Reder, 2008; Criss & Shiffrin, 2005; Kelley &
materials, including arousing words and pictures, as well Wixted, 2001). Thus, it is possible that prior findings of
as more realistic events like flashbulb memories (e.g., an impairment of association-memory due to arousal
Berntsen&Rubin,2004;Bradley,Greenwald,Petry,&Lang, mayinsteadhavebeenprimarilydrivenbyarousaleffects
1992;Brown&Kulik,1977;Christianson,1992;Mather& on item-memory, even in associative recognition. In this
Sutherland, 2011). Previous studies of arousal effects on case we will expect that more arousing pairs will have
association-memory have used a variety of paradigms in higherfalsealarmratesthanneutralpairs,ratherthanare-
which two elements (typically an arousing item paired ducedhitrate.Indeed,PierceandKensinger(2011)found
withaneutralitem)hadtobeboundtogetherinmemory. thatnegativepairshadhigherfalsealarmratesthanneu-
If those elements coincide, association-memory is often tral pairs. False alarm rates were also higher for positive
enhanced (e.g., font color of a presented word; Doerksen pairs than for neutral pairs, but this difference was not
C.R.Madanetal./JournalofMemoryandLanguage66(2012)695–716 697
statisticallysignificant.Nonetheless,hitrateswereequiva- could distract the participant momentarily and make it
lentacrossallthreepairtypes(unfortunately,Onodaetal., hardertosearchforthespecificcorrespondingassociation
2009,didnotreporthitratesandfalsealarmratesintheir inquestion,similartoworking-memoryimpairmentsdue
associativerecognitionresults,butonlyanaccuracymea- toarousal;Matheretal.,2006).Thenetresultoftwosuch
surethatcollapsedthetworates).Takentogether,neither opposingeffectscould producea reductionincued recall
associative recognition nor cued recall is a direct test of (as Zimmerman & Kelley, 2010 observed), but could still
association-memory; they can both be influenced by be unrelated to association-memory per se. Importantly,
item-memoryeffects.Byapplyingthesameexperimental byonlyusingasubsetofallpossiblecombinationsofpair
design we have used in Madan et al. (2010), here using arrangements, it is impossible to know if differences in
manipulationsofarousal,weareabletosystematicallydis- memoryeffectsobservedincuedrecallaredrivenbyitem-
entangle the effects of arousal on item- and association- or association-memory effects. Furthermore, it can be
memoryoncuedrecallperformance. especially difficult to compare across cued recall studies
Usingcuedrecallattestinsteadofassociativerecogni- if they use a different subset of the possible pair types:
tion, Guillet and Arndt (2009) conducted a paired-associ- GuilletandArndt(2009)onlymanipulatedtheproperties
ate study where words were presented in central and of the probe item, while always using a neutral target
peripheral locations during study (Experiments 2A–2C). item.ZimmermanandKelley(2010)insteadmanipulated
The central word was always used as the cued recall thepairasawhole,wherepairswerecomposedofeither
probe, and was either a neutral, negative, or taboo word. twoarousingitemsortwoneutralitems,butneveroneof
Theperipheralwordwasalwayslaterusedasthecuedre- each.
call target, and was always neutral. Guillet and Arndt InExperiment1,weuseda2(cid:2)2designwithallpossi-
(2009) found better cued recall performance with taboo blecombinationsofmoderatelyarousingnegative(N)and
probe words, as compared to neutral or negative probe neutral(n)wordsascuedrecallprobesandtargets(i.e.,NN,
words.Thiswasinterpretedasanenhancementofassoci- Nn, nN, nn). We expected that by being able to better ac-
ation-memory due to arousal. In contrast, Zimmerman count for item-memory effects (model parameters for
and Kelley (2010) had participants study pairs of words probeeffectiveness,p,andtargetretrievability,t)ourmod-
thatconsistedofeithertwonegativewordsortwoneutral el-basedanalysiswouldprovidemoreconclusiveevidence
words. Compared to neutral pairs, they found enhanced regardingtheinfluenceofarousalonassociation-memory
free recall (testing item-memory) for single negative than has been previously demonstrated. Thus, we had
words,butimpairedcuedrecall(aimedattestingassocia- threeplausiblealternativehypotheses:(a)arousalmayen-
tion-memory) for negative pairs. Participants also made hance association-memory (as estimated by the model
judgements of learning (JOLs) during study, and signifi- parameters for the strength of the relationship between
cantly overestimated their ability to remember the nega- the items, r and r ), (b) arousal may impair association-
1 2
tive pairs relative to overt cued recall performance, memory,orthat(c)previouslyreportedresultsregarding
whiletheirJOLsforneutralpairsweremoreprecise.Thus, theeffectsofarousalonassociation-memorymayinstead
thefindingsofZimmermanandKelley(2010)suggestthat be driven solely by item-memory effects (estimated by p
associations between two negative items are impaired, andt).
and that this may be due to a subjective overconfidence We also included a control group that studied words
in memory when subjects encounter such pairs. This composed of categorized neutral (C) and random neutral
may possibly have caused subjects in Zimmerman and (n) words to test whether effects of negative arousing
Kelley (2010) to apply less effort when learning negative wordscouldbeexplainedintermsofcategoryeffects.Fi-
pairs thanneutral pairs. nally,inExperiment2,weexaminedhighlyarousingword
Weproposethatthesetwocuedrecallstudies(Guillet (taboowords[T])tofurthertesttheinfluenceofincreased
& Arndt, 2009; Zimmerman & Kelley, 2010) may have arousalonassociation-memory.
come to opposite conclusions because they did not use Additionally, all four of the studies discussed earlier
all possible pair types of arousing/neutral items, and the usedsimultaneouspresentationofbothwordsinthepair,
cued recall measures may have confounded effects of which may have resulted in ill-controlled differential
item- and association memory. To illustrate, Guillet and attention to each one of the two words during encoding,
Arndt(2009)foundenhancedcuedrecallofaneutralword which in turn may have caused differential recall rather
thathadbeenpairedwithatabooword,comparedtohav- thanarousaloftheword(s)perse.Toavoidthispotential
ingbeenpairedneutralword.Itispossiblethattheresults confound, we instead presented the words of each pair
ofGuilletandArndt(2009)couldalsobeproducediftaboo sequentially to allow participants to equally and fullyat-
words are simply better probe items (e.g., preferentially tendboth.
processed) than neutral words, without truly increasing Basedontheresultsofthesepreviousstudies,weout-
memoryfortheassociationperse.Incontrast,Zimmerman line our three alternative hypotheses regarding the influ-
and Kelley (2010) reported impaired cued recall when enceofemotionalarousaloncuedrecallasfollows:
bothwordswerenegative(versusbothwordsbeingneu-
tral). This reduced recall of negative pairs could exclu- 1.1.Association-memoryenhancementhypothesis
sively result from non-association-memory mechanisms.
Single arousing words could be easier to recall as targets With an intentional encoding instruction and sequen-
in cuedrecallcomparedtoneutralwords,buttheycould tial presentation of the two words in each pair, partici-
be worse probes (for example, an arousing probe word pants’ ability to fully attend to both words in each pair
698 C.R.Madanetal./JournalofMemoryandLanguage66(2012)695–716
equallyandtodeliberatelyformmemoryfortheassocia- found stronger memory impairment for words immedi-
tionshouldbemaximal.Ifarousalincreasestheformation atelypreceding,ratherthanfollowing,emotionaloddballs
ofassociativememories,asproposedbyGuilletandArndt (‘emotion-induced retrograde amnesia’; see also Strange,
(2009),thatmemoryfortheassociationshouldbegreater Hurlemann, & Dolan, 2003). Thus, by including both test
whenthepair involvesoneor morearousingwordsthan directionswearealsosensitivetopossibletemporalasym-
whenapairinvolvesonlyneutralwords:bothr andr >1. metriesduringencoding.
1 2
1.2.Association-memoryimpairmenthypothesis 3.Methods
Alternatively, if participants apply less effort when PriortoExperiment1,weconductedanormingstudy.
formingassociationsthatinvolvearousingwords(assug- This preliminary study was necessary as no published
gested by the results of Zimmerman & Kelley, 2010) dis- database contained sufficient normative data at the time
tracting or otherwise damaging effects of arousal on the of commencement of our experiments. We additionally
formation of association-memory cannot be eliminated used this norming study to obtain similarity judgements,
by sequential presentation alone. Therefore, our alterna- whichwereessentialformatchingournegativeandcate-
tivepredictionisthatmemoryfortheassociationwillbe gorizedneutrallists.
reducedwhenapairinvolvesoneormorearousingwords
thanwhenapairinvolvedonlyneutralwords:bothr and
1 3.1.NormingstudyforExperiment1
r <1.
2
3.1.1.Participants
1.3.Item-memoryhypothesis A total of twelve participants participated in our nor-
ming study. Five participants who took part in the study
Emotionally arousing items are remembered better receivedpartialcredittowardsanundergraduatepsychol-
than neutral items. If arousing items are more easily re- ogycourseandsevenstudentsreceivedmonetarypayment
trievedthanneutralitems,theresultswesummarizedear- of$12fortheirparticipation.Noneoftheparticipantsfrom
lierthatwerebasedonsubsetsofpossiblepairtypesmay thisnormingstudytookpartinthememorytaskofExper-
insteadhavebeendrivensolelybythemodulationofprobe iments1or2.
effectiveness and/or target retrievability due to arousal.
The prediction is that memory for associations will be
3.1.2.Methods
unaffected by arousal (both r and r =1) in our model),
1 2 Participantsratedalargerpoolofwordsforarousaland
but instead may be driven by an impairment of probe
valenceusingtwoSelf-AssessmentManikins(SAM;Brad-
effectivenessduetoarousal(p<1)alongwithanenhance-
ley&Lang,1994)aswellasthesimilaritybetweenselected
mentoftargetretrievability(t>1).
pairsofwords.Specifically,participantsweregivenpencil-
and-paperquestionnaireswith160wordstobejudgedon
2.Experiment1 arousalandvalenceaswellas240wordpairstobejudged
on semantic similarity. To avoid fatigue, tasks alternated
Wetestedcuedrecallofwordpairscomprisedofmod- between blocks of 40 arousal/valence judgments and 80
eratelyarousing,negativewords(N)andneutral(n)words. semanticsimilarityratings.
Previous research has suggested that item-memory for Twelve word lists were created: (a) Four lists of 24
negativewordsmaybeenhancedbecausenegativewords moderately arousing, negative words were preselected
haveaclosersemanticcohesivenessthan‘randomneutral’ based on norms from the Affective Norms for English
words(Buchanan,Etzel,Adolphs,&Tranel,2006;Talmi& Words (ANEW; Bradley & Lang, 1999). According to this
Moscovitch, 2004). If the enhancement of item-memory reference,eachwordhasanassignedvalencevalue(scored
duetoarousalcanbelargelyexplainedbysemanticcohe- onascaleof1[verynegative]to9[verypositive]),andan
siveness, the same may hold for association-memory. To arousal value (scored on a scale of 1 [not arousing] to 9
ruleoutthispossibility,wealsoinvestigatedmemoryper- [highlyarousing]).(b)Foreachoffourcategories‘driving’,
formancewithcategorizedneutral(C)andrandomneutral ‘school’,‘house’,and‘business,’onelistofcategorizedneu-
n)wordsinaseparategroupofparticipants. tralwordswascreated,resultinginfourcategorizedlists.
Toidentifyanyasymmetriesduetosequentialpresen- Thecategories‘driving’and‘school’weretakenfromTalmi
tation,weprobedpairsbothintheusual,‘‘forward’’direc- and Moscovitch (2004) and ‘house’ and ‘business’ were
tion(giventhefirstitem,recalltheseconditem)andinthe self-generated,eachcontaining48words.(c)Fourlistsof
‘‘backward’’ direction (given the second item, recall the 48randomneutralwordseachwerecreated.
first item). To illustrate with an example, a negative (N) For the arousal/valence ratings, lists were counterbal-
word would be the target in neutral–negative (nN) pairs anced such that each participant rated a quarter of the
tested in the forward direction as well as in negative– word pool. Thus, across four participants, each word was
neutral(Nn)pairstestedinthebackwarddirection;however, ratedonceforarousalandvalence.Asournormingstudy
the arousing word in each pair would be encoded in a had twelve participants, each word was rated by exactly
different order due to the sequential presentation. Order threeparticipants.
effectsmayplayaroleintheinfluenceofarousalonmem- Duetothelargenumberofpossiblesemanticsimilarity
ory. For example, Miu, Heilman, Opre, and Miclea (2005) judgments, each participant was randomly assigned 240
C.R.Madanetal./JournalofMemoryandLanguage66(2012)695–716 699
similarity judgments: 12 (negative) or 24 (categorized Table1
neutral or random neutral) word pairs from each of the Arousal, valence, and similarity ratings are subjective ratings from the
norming task in Experiment 1. Familiarity, imageability, and frequency
twelve lists. This (incomplete) set of similarity ratings
ratings were obtained from the MRC Psycholingustic Database (Wilson,
wasusedtoquantifythesimilaritybetweenagivenword 1988).Meanratings areshownwithstandard deviation inparentheses.
andtherestofthelist. Meansinarowwiththesamesuperscriptarenotsignificantlydifferentat
Inter-rater reliability for was reasonably high for both p<.05.
arousal [Cronbach’s a=.78] and valence [a=.86] ratings.
Negative Category Neutral
Inter-raterreliabilitycouldnotbecalculatedforsemantic
Arousal 5.20(0.91)a 3.68(0.82)b 3.75(0.58)b
similarity judgements, since each participant was given Valence 2.23(1.00)a 4.20(0.69)b 4.22(0.54)b
differentwordpairings.
Similarity 4.78(0.63)a 4.75(0.61)a 2.92(0.58)b
Familiarity 520(48)c 531(44)c 528(46)c
3.1.3.Word-selectioncriteria Imageability 490(69)c 492(79)c 505(104)c
To equate arousal levels among negative lists and en- Frequency 44.5(50.5)c 58.2(73.7)c 54.6(52.1)c
sure higher arousal for negative lists than for the neutral Length 5.92(1.60)c 6.36(1.80)c 5.80(1.56)c
lists,weremovedwordsfromournegativelistsbasedon
our obtained arousal ratings. Likewise, to ensure that and to be comfortable typing. Participants gave written
words within all categorized neutral and negative lists informedconsentpriortobeginningthestudy,whichwas
wereperceivedasequallyrelated,weremovedwordsthat approvedbyaUniversityofAlbertaResearchEthicsBoard.
werelessrelatedtotherestofthewordsin thelist.This
ensured a narrow range of similarity ratings within each
3.3.Materials
list. We further removed words from each list to ensure
thatthelistswereequivalentforseveralotherwordprop-
Weusedthethreewordpoolsproducedinthenorming
erties known to affect memory performance: familiarity,
study. Stimuli in the Negative group were negative and
imageability, word frequency, and number of letters
randomneutralwords;stimuliintheCategorygroupwere
(MRCPsycholinguisticDatabase;Wilson,1988).
categorizedandrandomneutralwords.
3.1.4.Finalwordpools
3.4.Procedure
WeproducedthreewordpoolsforuseinExperiment1:
64 negative words, 64 categorized neutral words, and 64
Eachparticipanthadonepracticestudyset(whichwas
randomneutralwords.SeeTable1forwordpropertysta-
notincludedinthedataanalysis)followedbyeightexper-
tisticsforthespecificwordsusedinExperiment1(alsosee
imental study sets. Each study set consisted of three
AppendixA).1
phases: study phase, distractor phase, and cued recall.
With the creation of these three word pools, we pro-
Thesessionconcludedwithafinalfreerecalltask.
ceededwithourmemorystudy.
All stimuli were presented in a white ‘‘Courier New’’
font, which ensured fixed letter width, on a black back-
3.2.Participants
ground,inthecenterofthescreen.Wordswerepresented
sequentially,for3000mseach,plusa50msinter-stimulus
Atotalof240undergraduatestudents(meanage±sd=
intervalwithinpairsanda4000msinter-pairintervaldur-
19.02±1.57; 88 males and 152 females) participated in
ingwhichafixationcross,‘‘+’’,wasdisplayedinthecenter
ourstudyforpartialcreditinanintroductorypsychology
of the screen. During the study phase, participants were
courseattheUniversityofAlberta.Datafromsevenofthese
presentedwitheightwordpairs,askedtostudythepairs,
participants were lost due to machine error. Participants
andtoldthattheirmemoryforthepairswouldbetested
were randomly assigned to either the Negative group
later on. Each study set consisted of two pairs of each of
(N=120)ortheCategorygroup(N=113).Allparticipants
the four pair types: NEGATIVE–NEGATIVE (NN), NEGA-
wererequiredtohavelearnedEnglishbeforetheageofsix
TIVE–NEUTRAL(Nn),NEUTRAL–NEGATIVE(nN),andNEU-
TRAL–NEUTRAL (nn). Pair types in the Category group
1 WealsocalculatedLSAcos(h)asanadditionalmeasureofwithin-pool were: CATEGORIZED–CATEGORIZED (CC), CATEGORIZED–
wordsimilarity(Landauer&Dumais,1997).LSAcos(h)withineachofour NEUTRAL (Cn), NEUTRAL–CATEGORIZED (nC), and NEU-
wordpools(mean±sd)areasfollows:negative(.230±.128),categorized
TRAL–NEUTRAL (nn). NN/CC/nn are ‘pure’ pairs and Nn/
(.153±.145),andneutral(.077±.088).Independent-samplet-tests(withdf
nN/Cn/nCare‘mixed’pairs.Wordpairings,wordmember-
based on the effective number of independent comparisons) of the LSA
cos(h) values suggest that our negative words were more cohesiveas a ship by pair type, order of pairs, and order of pair types
categorythanbothourneutralwords[t(126)=7.92,p<.001,d=1.40]and wereallrandomizedacrossparticipants.FortheCategory
ourcategorizedwords[t(126)=3.19,p<.01,d=.56].TheLSAmeasureof group,withineachstudyset(eightwordpairs),allcatego-
similarity suggests thatnegative words hadmorewithin-pool similarity
rizedneutralwordsbelongedtothesamecategory(e.g.,all
than the categorized words. This contrasts with our subjective ratings,
wherenodifferencewasmeasurablebetweenthenegativeandcategory ‘school’ words). The category used in each study set of
words.Becauseournormingstudywasconductedwithparticipantsdrawn eightwordpairswasrandomizedsuchthateachofthefirst
fromthesamesubjectpopulationasourcuedrecallstudy,wesuggestthat fourstudysetsusedadifferentsemanticcategory.
oursubjectivenormsarethemoreprecisemeasureofhowsemantically
Thedistractortaskconsistedoffoursimplearithmetic
relatedourcuedrecallparticipantswouldperceivethepairedwordstobe.
problems, in the form of A+B+C= , where A, B, and C
Ourcategorized wordswerealso morecohesive asacategorythanour —
neutralwords[t(126)=3.59,p<.001,d=.63]. were randomly selected digits between two and eight.
700 C.R.Madanetal./JournalofMemoryandLanguage66(2012)695–716
Each problem remained in the center of the screen for ing probability of final free recall (Fig. 1). Participants in
5000ms.Theparticipantwasaskedtotypethecorrectan- the Negative group recalled more negative words than
swerduringthisfixedinterval,afterwhichthescreenwas neutral words [t(119)=2.72, p<.01, d=.32; Negative:
clearedfor200ms. mean=.22;Neutral:mean=.19;forCIs,seeFig.1].Partic-
Duringcued recall, a probe word was presented along ipants in the Category group recalled more categorized
with a blank line either to the left or right of the word, words than neutral words [t(112)=6.17, p<.001, d=.58;
dependingonwhetherthetargetwordhadbeenpresented Categorized: mean=.23; Neutral: mean=.19]. However,
asthefirstorseconditeminthepair,respectively.Partic- there was no significant difference between recall rates
ipantswereaskedtorecallthewordthatwaspairedwith for negative and categorized words [t(231)=1.26, p>.1,
the probe word during the study phase, type their re- d=.16]. Similar to previous studies, our results suggest
sponses into the computer, and press the ‘‘Enter’’ key. that enhancement of item-memory due to arousal, com-
Withineachstudyset,halfthecuedrecallprobesforeach paring negative and neutral words, can be explained as
pair type were in the forward direction and half were in an effect of semantic cohesiveness alone (cf. Buchanan
the backward direction. Note, as the participant was un- etal.,2006;Talmi&Moscovitch,2004).Betweenthetwo
able to predict the testing direction for a given pair, participant groups, no difference was observed in recall
thisfurtherpreventedanypotentialencodingpreferences oftheneutralwords[t(231)=1.36,p>.1,d=.18].
for one part of the association (e.g., ‘central/peripheral’
trade-offsinattention). 4.2.Cuedrecallaccuracy—negativegroup
Participants had a maximum of 15,000ms to respond,
after which the screen was cleared for 250ms. If partic- We conducted a TARGET TYPE[2](cid:2)ASSOCIATION
ipants could not recall a target word for the probe word, TYPE[2](cid:2)TEST DIRECTION[2] repeated-measures ANO-
they were instructed to type ‘‘PASS’’. Misspellings or VA,withcuedrecalltargetaccuracyasthedependentmea-
variants of the correct word were scored as incorrect re- sure. The levels of TARGET TYPE were ‘negative’ and
sponses. Note that responses were also analyzed using a ‘neutral’.ASSOCIATIONTYPEwaseither‘pure’or‘mixed’.
spell-checking search algorithm using the common UNIX TESTDIRECTIONwaseither‘forward’or‘backward’.‘Pure’
spell-checking program aspell (Philips, 1990, 2000). All pair accuracy included cued recall accuracy from the NN
incorrect responses were processed by the algorithm and nn pairs—pairs for which both the probe and target
and were marked as correct if the correct response was wereofthesameitem-type(negativeorneutral).‘Mixed’
found in the list of possible corrections, as done in our pair accuracy was the cued recall accuracy from the Nn
previous study (Madan et al., 2010). Since analyses with ornNpairs—forwhichtheprobeandtargetareofdifferent
responses both before and after spell-checking were not item-types. See Table 2 for a demonstration of how our
substantially different, we report only analyses using experimentalconditionsmappedontoourANOVAfactors.
the strict spelling criterion for accuracy. Response time This ANOVA design was also conducted on the Category
was logged both when the participant pressed the first groupinananalogousmanner.
key of their response, as well as after they pressed ‘‘En- CuedrecallaccuracyfortheNegativegroupisplottedin
ter’’ to submit their response. However, response time Fig.2a.WefoundasignificantmaineffectofASSOCIATION
measures yielded no additional information (e.g., no TYPE,wherepurepairswererecalledwithgreateraccuracy
h i
speed-accuracy trade-off) and as such will not be dis- than mixed pairs Fð1;119Þ¼15:65; p<:001; g2¼:12 .
p
cussed further. Thismaineffectwasqualifiedbytheinteractionbetween
Followingtheeightstudysetstherewasafinalfreere- TARGET TYPE and ASSOCIATION TYPE ½Fð1;119Þ¼
call task. Participants had five minutes to recall as many 43:95; p<:001; g2¼:27(cid:3). Simple effects analyses found
p
wordsastheycouldrememberfromtheexperiment.Par- that pure-pair negative words were recalled worse than
ticipantswereinstructedtotypeinawordandpressthe pure-pair neutral words [t(119)=4.66, p<.001, d=.35].
‘‘Enter’’ key. Once a participant pressed the ‘‘Enter’’ key, However,mixed-pairnegativewordswererecalledbetter
thescreenclearedandtheparticipantwasallowedtotype than mixed-pair neutral words [t(119)=4.74,p<.001,
in another word. Repeated responses were only counted d=.35]. TARGET TYPE had no significant main effect
h i
once.ThetaskwasimplementedwiththePythonexperi- p>:1; g2<:001 .
p
mentallibrary(pyEPL;Geller,Schleifer,Sederberg,Jacobs, TEST DIRECTION had no significant main effect
&Kahana,2007). h i
p>:1; g2¼:001 andwasnotinvolvedinanysignificant
p
interactions [all g2’s <.02]. This means that test direction
4.Resultsanddiscussion p
per se did not significantly affect performance. However,
theremainingfactorsmaytelluswhetheritmatterswhich
AllanalysesarereportedwithGreenhouse-Geissercor-
item was the probe or target, regardless of test direction
rectionfornon-sphericitywhereappropriate.Effectswere
(i.e., in the form of an interactionbetween ASSOCIATION
consideredsignificantbasedonanalphalevelof0.05.
TYPE and TARGET TYPE). The symmetry this implies in
the pure pairs is a replication of numerous prior findings
4.1.Finalfreerecall of equivalent forward and backward recall in homoge-
neouspairs(Kahana,2002).Furthermore,thelackofsignif-
We first asked whether our methods replicated the icant effects of TEST DIRECTION means we found no
enhancementofitem-memoryduetoarousal,byexamin- evidence for asymmetric influences of arousal on recall,
C.R.Madanetal./JournalofMemoryandLanguage66(2012)695–716 701
0.4
Experiment 1 Experiment 2
0.35
0.3
0.25
y
c
a
ur 0.2
c
c
A
0.15
0.1
0.05
0
Negative Neutral Category Neutral Taboo Neutral
Word Type
Fig.1. Proportionofstudieditemsrecalledinfinalfreerecall,forbothexperiments.Errorbarsrepresent95%confidenceintervals,correctedforinter-
individualdifferences(Loftus&Masson,1994).
implied by previous findings of temporally asymmetric beingcompletedsuccessfully.Bythislogic,theprobability
effectsofarousalonmemory(e.g.,Miuetal.,2005;Strange that all three processes will be successful is the result of
etal.,2003). multiplyingtheprobabilitiesfromthethreeprocessesto-
gether.Inthemodel,Acc(PairType,TestDirection),denotes
4.3.Cuedrecallaccuracy—categorygroup accuracy for all eight possible pair type(cid:2)test direction
conditions,andistheproductofprobabilitiesrepresenting
PerformanceinthisgroupisshowninFig.2b.Weob- (a)howeffectivelytheprobeitemcanbeusedtoaccessthe
served main effects of TARGET TYPE and ASSOCIATION associationinmemory,(b)thestrengthofarepresentation
TYPE.ThemaineffectofTARGETTYPEfoundthatcatego- of the association in memory, and (c) whether the target
rized words were recalled better as targets than neutral itemwillbeproducedcorrectly.
h i
words Fð1;112Þ¼9:74; p<:01; g2p¼:08 . The main ef- AccðPairType;TestDirectionÞ
fectofASSOCIATIONTYPEshowedthataccuracyinpure ¼PðProbeÞ(cid:2)PðRelatÞ(cid:2)PðTarget Þ
i j k
pairs was better than accuracy in mixed pairs
h i Using the Negative group as an example, P(Probe) de-
Fð1;112Þ¼45:38; p<:001; g2¼:28 . The interaction i
p notes the probability of effectively using the probe item
h i
was not significant p>:1; g2<:001 . Similar to the to access the association in memory, where, i=N,n. P(Re-
p
lat)denotestheprobabilitycorrespondingtothestrength
Negativegroup,TESTDIRECTIONneitherhadasignificant j
h i of the association in memory, where j=NN, Nn, nN, nn.
maineffect p>:1; g2¼:01 norwasnotinvolvedinany
p P(Targetk) denotes the probability of retrieving the target
significantinteractions[allg2’s<.01]. item, where k=N, n. This results in the following system
p
ofequations:
4.4.Model-basedestimationofthearousal/categoryeffectson
AccðNN;ForwardÞ¼PðProbe Þ(cid:2)PðRelat Þ(cid:2)PðTarget Þ
item-andassociation-memory N NN N
AccðNN;BackwardÞ¼PðProbe Þ(cid:2)PðRelat Þ(cid:2)PðTarget Þ
N NN N
Toquantifytherelativeeffectsofarousalandcategory AccðNn;ForwardÞ¼PðProbeNÞ(cid:2)PðRelatNnÞ(cid:2)PðTargetnÞ
properties on item-memory versus association-memory AccðNn;BackwardÞ¼PðProbe Þ(cid:2)PðRelat Þ(cid:2)PðTarget Þ
n Nn N
we fit mean accuracy with a probabilistic ‘‘item-relation- AccðnN;ForwardÞ¼PðProbe Þ(cid:2)PðRelat Þ(cid:2)PðTarget Þ
n nN N
ship’’ model (Madan et al., 2010). Our model is based on
AccðnN;BackwardÞ¼PðProbe Þ(cid:2)PðRelat Þ(cid:2)PðTarget Þ
multiplicationasitassumesthattherearethreeseparable N nN n
Accðnn;ForwardÞ¼PðProbe Þ(cid:2)PðRelat Þ(cid:2)PðTarget Þ
processesinvolvedinsuccessfullyrecallingatargetitemin n nn n
cued recall. Each of these processes has a probability of Accðnn;BackwardÞ¼PðProbenÞ(cid:2)PðRelatnnÞ(cid:2)PðTargetnÞ
702 C.R.Madanetal./JournalofMemoryandLanguage66(2012)695–716
Table2
Factorialdesignofthepairtypesandtestdirectionsusedinourstudy,usingtheNegativegroupasanexample.Typesofprobe,relationship,andtargetare
listedforallpossiblepairtype(cid:2)testingdirectioncombinations.
Pairtype Testingdirection Probe Relationship Target
NN Forward N NN(Pure) N
Backward N NN(Pure) N
Nn Forward N Nn(Mixed) n
Backward n Nn(Mixed) N
nN Forward n nN(Mixed) N
Backward N nN(Mixed) n
nn Forward n nn(Pure) n
Backward n nn(Pure) n
Theleft-handsidesoftheseequationscorrespondtothe versuseffectsofarousalthatinfluencecuedrecallaccuracy
conditionsinourfactorialdesign(Table2).Becauseweare but would be better understood as effects due to item-
interestedintherelativeeffectsofstimuluspropertieson memory. Next we explain the model and give concrete
eachofthesethreeprocesses,wedefinethefollowingratio examples of cognitive processes that each of the model
parameters which will be used as free parameters in the parameterscouldplausiblyreflect.
modelfits: Thetparameterrepresentsaninfluenceoncuedrecall
performance that depends only on characteristics of the
PðProbe Þ
p¼ N target word. Note that during study, participants cannot
PðProbe Þ
n know which of the two words of a given pair will later
r ¼ PðRelatNNÞ becomethetargetincuedrecall.Forthisreason,treflects
1 PðRelatNn;nNÞ effects of word properties that influence item retrieval
PðRelat Þ processes. As an example, in many models of recall (e.g.,
r ¼ Nn;nN
2 PðRelat Þ SAM; Raaijmakers & Shiffrin, 1981), one assumes that
nn
PðTarget Þ manycandidatewordsaresampled(analogoustodrawing
t¼ N apieceofpaperfromahat).Someitempropertiesmight
PðTarget Þ
n make them more likely to be sampled (e.g., a word with
If any ratio parameter were equal to the value 1, this ahighwordfrequencywouldbewrittenonseveralpieces
wouldrepresentanulleffectofarousalonthemechanism ofpaperinthehat).
in the model. If the parameter were significantly greater Similarly, the p parameter represents influences on
than1,thiswouldindicateasignificantadvantagedueto cuedrecallthatdependonlyonthepropertiesoftheprobe
arousalintherespectivememoryprocess.Forexample,if word.Again,becauseeitherofthewordspresentedduring
t>1, negative words were more retrievable target items studymaylaterbecometheprobeduringcuedrecall,pre-
than neutral words. Similarly, if p>1, negative words flects word properties that influence cued recall perfor-
wouldbemoreeffectivelyaccessedasprobesthanneutral mance at test. For example, one type of word might be
words.Iftheratioparameterweresignificantlylessthan1 more easily identifiable due to unique or distinctive fea-
in the model fit, this would suggest that the respective tures.Crissetal.(2011)demonstratedthatcontextualvar-
memoryprocesswouldbeimpairedbyarousal. iability, a measure of the number of different contexts a
Thus far, our item-relationship model is underdeter- wordcanbeusedin,canprimarilyinfluenceprobeeffec-
mined(therearemultiplewaystoexplainthedatausing tiveness. Specifically, low, as opposed to high, contextual
various combinations of parameters). For this reason, we variabilitywordsservedasbetterprobeitemsincuedre-
only used further-constrained model variants wherein a call. This significant probe effect is likely due, at least in
subset of the parameters p, r1, r2, and t was fixed to 1 part, to low contextual variability words having less pre-
andtheremainingparameterswerefreetovary.Aftercon- existingassociationscompetingwithnewepisodicassoci-
straining the model, the system of equations could be ations learned in the experiment, than compared to high
solvedalgebraicallyforeachparticipantandthenparame- contextual variability words. As a result, low contextual
tervaluesandmodelfitsweresummarizedacrosspartici- variabilitywordsaremoredistinctive(byhavinglesscon-
pants.WecalculatedBIC(BayesianInformationCriterion) textualvariability)andaremoreeffectiveasprobeitems,
asameasureofmodel-fitnessthattakesintoaccountthe withoutenhancingmemoryforassociationsperse.
numberoffreeparameters.Byconvention,ifthedifference Theremainingtwoparameters,r andr ,refertomem-
1 2
betweentwomodelfits,DBIC<2,neitherofthemodels’fit oryfortheassociationandrepresentinfluencesoncuedre-
tothedataissignificantlybetter—thuswereportallscores call due to the properties of both paired items together,
as DBIC relative to the best-fitting model (Burnham & regardlessofwhichonewillbecometheprobeandwhich
Anderson,2002,2004). willbecomethetarget.r representstheratioofrecalling
1
NN relative to Nn and Nn pairs. r represents the ratio of
2
4.5.Interpretationofmodelparameters recalling Nn and nN relative to nn pairs. Because these
parameters refer only to the properties of the word
Thegoalofthemodel-basedanalysisistobeabletodis- pair, they are agnostic as to which item is the probe or
tinguish effects of arousal on association-memory per se, target;thus,r andr canidentifyeffectsincuedrecallthat
1 2
C.R.Madanetal./JournalofMemoryandLanguage66(2012)695–716 703
Experiment 1: Negative group propertiesoftheprobewordorthetargetword.Theresult-
(a) 0.8 ingassociation-memorychangeswouldonlydependonthe
Forward
presenceofthearousingwordwithintherespectivepairbut
0.7 Backward notonpropertiesofthearousingwordbyitself.Ifcuedre-
callperformanceisalinearfunctionoftheitemproperty
0.6
beingmanipulated,with,hypothetically,recallbeingbest
0.5 for NN pairs, intermediate for mixed pairs (Nn and nN),
y
urac 0.4 aitnidsqwuoitrestpflaoursninblpeatihras,twreecawllovualrdieesxnpoenc-tlirn1e=arrl2y.Hasoawfeuvnecr-,
c
c
A tionofitemproperty.Itisalsopossiblethatrecallisanon-
0.3
monotonicfunctionoftheitemproperty.Forinstance,pure
0.2 pairsmaybebetterrememberedthanmixedpairs,dueto
associabilityorcompatibilitybetweenthewordsthatmake
0.1
upeachpairs.Wemaytheninsteadfindthatr =1/r .To
1 2
illustratethiswithinourexperiment,itislikelythatpure
0
NN Nn nN nn pairs of words belonging to the same semantic category
(Categorygroup)wouldbebetterrecalledthanmixedpairs.
Experiment 1: Category group As these relationships are usually unknown, we allow r
1
(b) 0.8 andr tobefitseparatelyandindependently.
Forward 2
Note,becauseofassociativesymmetry(Kahana,2002),
0.7 Backward
thesekindsofeffectsshouldinfluenceforwardandback-
0.6 wardprobesofagivenpairtypeequivalently,i.e.,anNpair
testedintheforwarddirectionwouldbethesameasaNn
y 0.5 pair tested in the backward direction. In both cases the
c
a probe word is negative (N), the association is mixed (Nn
cur 0.4 ornN),andthetargetwordisneutral(n).OurANOVAre-
c
A 0.3 sults confirmed that test direction was not involved in
anymaineffectsorinteractionsoncuedrecall,andthere-
0.2 fore,testdirectionwasnotincludedinourmodels.
0.1
4.6.Approachtomodelselection
0
CC Cn nC nn Wefitsixconstrainedmodels,withoneormoreparam-
eters fixed to 1, to our behavioral data. In the first con-
Experiment 2: Taboo strained model, we tested how much of the cued recall
(c) 0.8 performancewasaccountedforbyarousaleffectsonitem
Forward
retrieval. This model is called the ‘Target-only’ model,
0.7 Backward
whereonlythetparameterwasallowedtofreelyvary.In
0.6 thesecondmodelwetestedeffectsofpureitem-memory
ina‘Probe-only’model(onlythepparameterwasallowed
y 0.5 tovaryfreely).Thismodeltestedhowmuchofthecuedre-
c
a call performance was accounted for by arousal effects on
ur 0.4
c the ability to effectively use the probe item. In the third
c
A 0.3 constrainedmodel,wetesteda‘Relationship-only’model,
where p(cid:4)t(cid:4)1, leaving two free parameters, r and r .
1 2
0.2 Although the pure item-effect models (the ‘Target-only’
modeland‘Probe-only’model)hadfewerdegreesoffree-
0.1
dom,ourfitnessmeasure(DBIC)includesapenaltyforde-
0 grees of freedom. Fig. 3 illustrates how these highly
TT Tn nT nn constrained models might explain our data. An enhance-
ment of item retrieval due to arousal, with no effect on
Fig. 2. Cuedrecallaccuracyacrosspairtypes.Errorbarsrepresent95%
association-memory,wouldproduceapatternofbehavior
confidenceintervals,correctedforinter-individualdifferences(Loftus&
Masson,1994).(a)AccuracyfortheNegativegroupofExperiment1.Pair similartoFig.3a.Anenhancementoftheefficacyofprobe
types:negative–negative(NN),negative–neutral(Nn),neutral–negative itemsduetoarousalwouldappearsimilartothepatternin
(nN), and neutral–neutral (nn). (b) Accuracyfor theCategory group of Fig. 3b. If arousal enhanced association-memory without
Experiment 1. Pair types: categorized–categorized (CC), categorized–
any effect on item-memory,results would appear similar
neutral(Cn),neutral–categorized(nC),andneutral–neutral(nn).(c)Accu-
racyforthetaboomanipulationofExperiment2.Pairtypes:taboo–taboo to Fig. 3c. Finally, an impairment of association-memory
(TT),taboo–neutral(Tn),neutral–taboo(nT),andneutral–neutral(nn). wouldappearsimilartoFig.3d.
Thefourthconstrainedmodelcontainedthetwoitem-
relate to enhanced association-memory, independent of memoryparameters:pandt(twofreeparameters;‘Probe
any memory enhancements or impairments due to the and Target’ model). The final two constrained models
704 C.R.Madanetal./JournalofMemoryandLanguage66(2012)695–716
involvedtherelationshipparametersandeitheroneofthe influence cued recall performance, which is reflected by
itemparameters,resultinginfreeparametersr andr in thefactthatallbest-fittingmodelsincludedtherelation-
1 2
addition to either p or t (three free parameters in total; shipparameters.Basedonconvergingevidencefromthefi-
‘Relationship and Probe’ and ‘Relationship and Target’ nal free recall task, where more categorized words were
models,respectively). recalled than neutral words, cued recall performance
To compute 95% confidence intervals of the model shouldalsobeinfluencedbyenhanceditem-retrievability
parametersobtainedforeachmodelvariant,weapplieda duetosemanticcohesiveness(t). Again,the‘Relationship
bootstrap, randomly sampling participants with replace- andProbe’modelcannotaccountforanydifferencesintar-
ment(animprovementonthemethodweusedinMadan get retrievability due to semantic cohesiveness, and also
et al., 2010). Each of 10,000 such random samples con- suggestsnosignificanteffectofsemanticcohesivenesson
tained the same amount of data. We then fit each model probeeffectiveness(p).Thus,strongestsupportwaspres-
variant to each resampled data set and report the corre- entforthe‘RelationshipandTarget’model.
sponding2.5and97.5percentilevaluesasthelimitsofa
two-tailed95%confidenceinterval. 4.6.3.Between-groupcomparisons
Afterdeterminingthefitsforeachofthesixconstrained We included the Category group as a control to test
models, multiple models can sometimes fit the data whether arousal effects on memory could be explained
equally well. In the final stage of model selection, we bya closersemanticcohesivenessamongnegativewords
examinedtheconfidenceintervalsforeachoftheparame- thanneutralwords.Ther parameterbeingbelowzeroin
1
tersinthebest-fittingmodelsanddrewonconvergingevi- the Negative group implies difficulties in association-
dencefromthefinalfreerecalldataandpreviousresearch. memorywhentwonegativewordsarelinkedtoeachother
ThesameapproachwasusedtofitthedataintheCategory asopposedtoonenegativeandoneneutralword.r being
1
groupandinExperiment2. above zero in the Category group in turn suggests that
association-memory of two words from a semantically
4.6.1.Negativegroup cohesive category helps rather than hinders association-
Thetwomodelsinvolvingtherelationshipandoneitem memory. That is, semantic relatedness alone cannot ac-
parameter(‘RelationshipandProbe’and‘Relationshipand countforthereductioninassociation-memoryfornegative
Target’models)providedthebestfitbyfar,suggestingthat words.
theeffectofarousaloncuedrecallperformanceinfluences IndependentmodelselectionfortheNegativeandCat-
bothitem-andassociation-memory(Table3;Fig.4f).Note egory groups suggested the same best-fitting model—the
thatduetomodelmimicry,both‘RelationshipandProbe’ ‘Relationship and Target’ model. Thus, here we compare
and‘RelationshipandTarget’modelsproduceidenticalfits thebest-fittingmodelparametersforeachgroupdirectly.
totheempiricaldata,buttheydosowithdifferentparam- Toaskwhetherbest-fittingmodelparameterswerediffer-
etervalues.Corroboratingevidencefromourowndataand entbetweenthetwogroups,wecomparedeachparameter
previous research was then used to drive our ultimate between groups with t-tests (see Table 3 for parameter
modelselection.Infinalfreerecall,wefoundbettermem- fits). Parameter values were log-transformed to satisfy
ory for negative words than for neutral words. This sug- the normality assumption of the t-test. r and t were not
2
gested that cued recall accuracy should incorporate significantly different between groups [p>.1]. This lack
enhancement of item-memory (t>1). Additionally many of significant difference of t between Negative and Cate-
studies have found an enhancement of item-memory for gorygroupssupportsthefindingsofTalmiandMoscovitch
arousingwords(Anderson,Wais,&Gabrieli,2006;Phelps, (2004),whofoundthattheenhancementofitem-memory
2004;Rubin&Friendly,1986).However,the‘Relationship due toarousalfor singlewordscouldbe explainedsolely
andProbe’modelcannotaccountforanydifferencesintar- bysemanticcohesiveness.Importantly,r wassignificantly
1
getretrievabilityduetoarousal(becausetwasfixedto1), differentbetweenconditions[t(228)=4.28,p<.001].Thus,
andinsteadsuggeststhatarousalimpairsprobeeffective- between-group differences in r suggest that arousal-in-
1
ness, a result that is not supported by existing research duced impairments of association-memory observed in
intoemotionandmemory.Thus,weselectedthe‘Relation- our data cannot be explained by semantic cohesiveness
ship and Target’ model as the more plausible model. The alone.Incontrast,thefindingthatr wasbelow1inboth
2
best-fittingparametersinthismodelsuggestthatarousing the Negative and Category group provides evidence that
items damage association learning (r and r <1), that sharedaspectsofbothmanipulations(i.e.,semanticcohe-
1 2
arousalalsoenhancedtargetretrieval(t>1),butthatarou- siveness) was negatively associated with association-
salhadnoinfluenceonprobeeffectiveness(p=1). memory.
OnecaveattousingdatafromourCategorygroupasa
4.6.2.Categorygroup controlforourNegativegroup,isthatparticipantsinour
We found three equivalently best-fitting models: the Category group were given words from four distinct
‘Relationship-only’ model, the ‘Relationship and Probe’ semanticcategories(‘driving’,‘school’,‘house’,‘business’),
model, and the ‘Relationship and Target’ model (Table 3; whereas participants in our Negative group were only
Fig. 5). The only difference between categorized words givenwordsfromasinglesetthatcouldbefunctioninglike
and neutral words was the relationship between the a semantic category (i.e., ‘negative’). Thus, it is possible
words, based on stronger pre-experimental associations that Category group has reduced category effects due to
between some of the items (the categorized words) than the use of multiple categories. To account for this we
others (the neutral words). Thus, this relationship should re-analyzed the data using only participants’ first study