ofanimallethergo

SjobergBehavBrainFunct(2017)13:3DOI10.
1186/s12993-017-0121-8RESEARCHLogicalfallaciesinanimalmodelresearchEspenA.
Sjoberg*AbstractBackground:Animalmodelsofhumanbehaviouraldeficitsinvolveconductingexperimentsonanimalswiththehopeofgainingnewknowledgethatcanbeappliedtohumans.
Thispaperaimstoaddressrisks,biases,andfalla-ciesassociatedwithdrawingconclusionswhenconductingexperimentsonanimals,withfocusonanimalmodelsofmentalillness.
Conclusions:Researchersusinganimalmodelsaresusceptibletoafallacyknownasfalseanalogy,whereinferencesbasedonassumptionsofsimilaritiesbetweenanimalsandhumanscanpotentiallyleadtoanincorrectconclusion.
Thereisalsoariskoffalsepositiveresultswhenevaluatingthevalidityofaputativeanimalmodel,particularlyiftheexperimentisnotconducteddouble-blind.
Itisfurtherarguedthatanimalmodelexperimentsarereconstructionsofhumanexperiments,andnotreplicationsperse,becausetheanimalscannotfollowinstructions.
Thisleadstoanexperimentalsetupthatisalteredtoaccommodatetheanimals,andtypicallyinvolvesasmallersamplesizethanahumanexperiment.
Researchersonanimalmodelsofhumanbehaviourshouldincreasefocusonmechanisticvalidityinordertoensurethattheunderlyingcausalmechanismsdrivingthebehaviourarethesame,asrelyingonfacevalid-itymakesthemodelsusceptibletologicalfallaciesandahigherriskofType1errors.
Wediscussmeasurestoreducebiasandriskofmakinglogicalfallaciesinanimalresearch,andprovideaguidelinethatresearcherscanfollowtoincreasetherigouroftheirexperiments.
Keywords:Argumentfromanalogy,Confirmationbias,Type1error,Animalmodels,Double-downeffect,ValidityTheAuthor(s)2017.
ThisarticleisdistributedunderthetermsoftheCreativeCommonsAttribution4.
0InternationalLicense(http://creativecommons.
org/licenses/by/4.
0/),whichpermitsunrestricteduse,distribution,andreproductioninanymedium,providedyougiveappropriatecredittotheoriginalauthor(s)andthesource,providealinktotheCreativeCommonslicense,andindicateifchangesweremade.
TheCreativeCommonsPublicDomainDedicationwaiver(http://creativecommons.
org/publicdomain/zero/1.
0/)appliestothedatamadeavailableinthisarticle,unlessotherwisestated.
LogicalfallacyAlogicalfallacyisajudgmentorargumentbasedonpoorlogicalthinking.
Itisanerrorinreasoning,whichusuallymeansthateitherthelineofreasoningisflawed,ortheobjectsinthepremiseoftheargumentaredissimi-lartotheobjectsintheconclusion[1].
Scientistsarenotimmunetologicalfallaciesandaresusceptibletomakingargumentsbasedonunsoundreasoning.
Forinstance,acommonfallacyisaffirmingtheconsequent.
Thisinvolvesthefollowinglineofreasoning:ifAistrue,thenXisobserved.
WeobserveX,thereforeAmustbetrue.
ThisargumentisfallaciousbecauseobservingXonlytellsusthatthereisapossibilitythatAistrue:theruledoesnotspecifythatAfollowsX,evenifXalwaysfollowA.
1Stud-iesthathaveexplicitlyinvestigatedthisinascientistsam-plefoundthat25–33%ofscientistsmakethefallacyofaffirmingtheconsequentandconcludethatX→Aisavalidargument[2,3].
Makinglogicalfallaciesisahumancondition,andthereisalargerangeoffallaciescommonlycommitted[1,4,5].
Inthepresentpaper,wewillfocusonaselectfewthatareofparticularrelevancetoanimalmodelresearch,espe-ciallyinthecontextofvalidityandreliabilityofconclu-sionsdrawnfromanexperiment.
1Ifyoustruggletofollowthislineofreasoning,aconcreteexamplemakesiteasier:Ifitiswine,thenthedrinkhaswaterinit.
Waterisinthedrink.
Therefore,itmustbewine.
Nowheredoestherulespecifythatonlywinecontainswaterasaningredient,sosimplymakingthisobservationdoesnotallowustoconcludethatitiswine.
OpenAccessBehavioralandBrainFunctions*Correspondence:espen.
sjoberg@hioa.
noDepartmentofBehavioralSciences,OsloandAkershusUniversityCollegeofAppliedSciences,St.
OlavsPlass,P.
O.
Box4,0130Oslo,NorwayPage2of13SjobergBehavBrainFunct(2017)13:3ConfirmationandfalsificationThefallacyofaffirmingtheconsequentisconnectedwithatendencytoseekevidencethatconfirmsahypothesis.
Manyscientistsconducttheirexperimentsundertheassumptionthattheirexperimentalparadigmisalegiti-mateextensionoftheirhypothesis,andthustheirresultsareusedtoconfirmtheirbeliefs.
Asanexample,imagineahypothesisthatstatesthatpatientswithbipolardisor-derhavereducedcognitiveprocessingspeed,andwedoareactiontimetesttomeasurethis.
Thus,afallaciouslineofreasoningwouldbe:ifbipolarpatientshavereducedcognitiveprocessingspeed,thenwewillobserveslowerreac-tiontimeonatest.
Weobserveaslowerreactiontime,andthereforebipolarpatientshavereducedcognitiveprocessingspeed.
Thiswouldbeaffirmingtheconsequent,becausetheobservedoutcomeisassumedtobetheresultofthemechanismoutlinedinthehypothesis,butwecannotwithcertaintysaythatthisistrue.
Theresultscertainlysuggeststhispossibility,anditmayinfactbetrue,butthepatientsmayhaveexhibitedslowerreactiontimesforavarietyofreasons.
Ifasignificantstatisticaldifferencebetweenbipo-larpatientsandcontrolsisfound,itmaybecommontoconcludethattheresultssupportthecognitiveprocessingspeedhypothesis,butinrealitythisanalysisonlyrevealsthatthenullhypothesiscanberejected,notnecessarilywhyitcanberejected[6,7].
Themanipulationoftheinde-pendentvariablegivesusaclueastothecauseoftherejec-tionofthenullhypothesis,butthisdoesnotmeanthatthealternativehypothesisisconfirmedbeyonddoubt.
Popper[8]claimedthathypothesescouldneverbeconfirmed;onlyfalsified.
Heclaimedthatwecouldnotconcludewithabsolutecertaintythatastatementistrue,butitispossibletoconcludethatitisnottrue.
Theclas-sicexampleisthewhiteswanhypothesis:evenifwehaveonlyobservedwhiteswans,wecannotconfirmwithcertaintythestatement"allswansarewhite",butifweobserveasingleblackswanthenwecanrejectthestate-ment.
Lookingforconfirmation(searchingforwhiteswans)includestheriskofdrawingthewrongconclusion,whichinthiscaseisreachedthroughinduction.
How-ever,ifweseekevidencethatcouldfalsifyahypothesis(searchingforblackswans),thenourobservationshavethepotentialtorejectourhypothesis.
Notethatrejectingthenullhypothesisinstatisticalanalysesisnotnecessar-ilysynonymouswithfalsifyinganexperimentalhypoth-esis.
Null-hypothesistestingisatool,andwhenweusestatisticalanalysesweareusuallyanalysinganumericalanalogyofourexperimentalhypothesis.
Whenahypothesiswithstandsmultipletestsoffalsifi-cation,Poppercalleditcorroborated[9].
Wecouldarguethatifahypothesisiscorroborated,thenitslikelihoodofbeingtrueincreases,becauseithassurvivedagaunt-letofcriticismbyscience[10].
However,itisimportanttonotethatPoppernevermadeanysuchsuggestion,asthiswouldbeinductivereasoning:exactlytheproblemhewastryingtoavoid!
Evenifahypothesishassupportingevidenceandhaswithstoodmultipleroundsoffalsifica-tion,Poppermeantthatitisnotmorelikelytobetruethananalternativehypothesis,andcannotbeconfirmedwithcertainty[11].
Instead,hefeltthatacorroboratedtheorycouldnotberejectedwithoutgoodreason,suchasastrongeralternativetheory[12].
Poppermaybecor-rectthatwecannotconfirmahypothesiswithabsolutecertainty,butinpracticeitisacceptabletoassumethatahypothesisislikelytrueifithaswithstoodmultipleroundsoffalsification,throughmultipleindependentstudiesusingdifferentmanipulations(see"Animalmodelexperimentsarereconstructions"section).
However,inthequestfortruthwemustalwaysbeawareofthepos-sibility,howeverslight,thatthehypothesisiswrong,evenifthecurrentevidencemakesthisseemunlikely.
ConfirmationbiasConfirmationbiasisthetendencytoseekinformationthatconfirmsyourhypothesis,ratherthanseekinginfor-mationthatcouldfalsifyit[13].
Thiscaninfluencetheresultswhentheexperimenterisinformedofthehypoth-esisbeingtested,andisparticularlyproblematiciftheexperimentreliesonhumanobservationsthathasroomforerror.
Theexperimentersimpactonthestudyisoftenimplicit,andmayinvolvesubtlyinfluencingparticipantsorunderminingmethodologicalflaws,somethingalsoknownasexperimenterbias[14].
Thetendencytoexpressconfirmationbiasinscienceappearstobemoderatedbywhatfieldofstudywebelongto.
Physicists,biologists,psychologists,andmathemati-ciansappeartobesomewhatbetteratavoidingconfir-mationbiasthanhistorians,sociologists,orengineers,althoughperformancevariesgreatlyfromstudytostudy[3,15–18].
Insomecases,thetendencytoseekconfirm-ingevidencecanbearesultofthephilosophyofsciencebehindadiscipline.
Forinstance,Sidman's[19]bookTac-ticsofScientificResearch,consideredalandmarktext-bookonresearchmethodsinbehavioranalysis[20–22],activelyencouragesresearcherstolookforsimilaritiesbetweentheirresearchandothers,whichislikelytoincreaseconfirmationbias.
Confirmationbiashasbeenshowninanimalresearchaswell,butthisfallacyisreducedwhenanexperimentisconducteddouble-blind[23].
VanWilgenburgandElgarfoundthat73%ofnon-blindstudieswouldreportasig-nificantresultsupportingtheirhypothesis,whilethiswasonlythecasein21%ofdouble-blindstudies.
Aninterest-ingnewapproachtoreduceconfirmationbiasinanimalresearchistofullyautomatizetheexperiment[24,25].
ThisinvolvessettinguptheequipmentandprotocolsPage3of13SjobergBehavBrainFunct(2017)13:3inadvance,sothatlargeportionsofanexperimentcanberunautomatically,withminimalinterferencebytheexperimenter.
Alongwithdouble-blindedstudies,thisisapromisingwaytoreduceconfirmationbiasinanimalexperiments.
Itisimportanttonotethattheconfirmationbiasphe-nomenonoccursasanautomatic,unintentionalprocess,andisnotnecessarilyaresultofdeceptivestrategies[26].
Ashumans,weaddlabelstophenomenaandestablishcertainbeliefsabouttheworld,andconfirmationbiasisawaytocementthesebeliefsandreinforceoursenseofidentity.
2Scientistsmaythereforebepronetoconfirma-tionbiasduetoalackofeducationonthetopic,andnotnecessarilybecausetheyareactivelyseekingtofindcor-roboratingevidence.
ArgumentfromanalogyandanimalmodelresearchTheissuesreportedinthispaperapplytoallofscience,andwediscussprinciplesandphenomenathatanyscien-tistwouldhopefullyfinduseful.
However,theissueswillprimarilybediscussedinthecontextofresearchonani-malmodels,assomeoftheprincipleshavespecialappli-cationsinthisfield.
Inthissection,weoutlinehowananimalmodelisdefined,andproblemsassociatedwitharguingfromanalogyinanimalresearch.
DefiningananimalmodelTheterm"animalmodel"isnotuniversallydefinedintheliterature.
Here,wedefineananimalmodelasananimalsufficientlysimilartoahumantargetgroupinitsphysi-ologyorbehaviour,basedonanatural,bred,orexperi-mentallyinducedcharacteristicintheanimal,andwhichpurposeistogenerateknowledgethatmaybeextrapo-latedtothehumantargetgroup.
Inthisarticle,wefocusontranslationalanimalmodelsinthecontextofbehav-iouraltesting,whichusuallyinvolveaspecificspeciesorstrain,orananimalthathaveundergoneamanipulationpriortotesting.
Ananimalmodelcanofcoursemodelanothernon-humananimal,butforthemostparttheaimofitistostudyhumanconditionsindirectlythroughanimalresearch.
Thatresearchisconductedonanimalsdoesnotnecessarilymeanthattheanimalactsasamodelforhumans.
Itisonlyconsideredananimalmodelwhenitsfunctionistorepresentatargetgrouporconditioninhumans,e.
g.
peoplewithdepression,autism,orbraininjury.
Thecurrentpaperfocusesonanimalmodelsofmentalillness,butanimalmodelsasawholerepresentalargevarietyofconditions,andareparticularlycommon2ThankstoRachaelWilnerforpointingoutthisargument.
touseindrugtrials.
SeeTable1foranoverviewofcom-monanimalmodelsofmentalillnesses.
Itshouldalsobenotedthattheterm"animalmodel"referstoananimalmodelthathasatleastbeenvali-datedtosomeextent,whileamodelnotyetvalidatedisreferredtoasa"putativeanimalmodel".
Thatamodelis"validated"doesnotmeanthatthestrengthofthisvali-dationcannotbequestioned;itmerelymeansthatprevi-ousresearchhasgiventhemodelcredibilityinonewayoranother.
ArguingfromanalogyInresearchonanimalmodels,scientistssometimesuseanapproachcalledtheargumentfromanalogy.
Thisinvolvesmakinginferencesaboutapropertyofonegroup,basedonobservationsfromasecondgroup,becausebothgroupshavesomeotherpropertyincom-mon[1].
Analogiescanbeveryusefulinourdailylivesaswellasinscience:amathematicalmeasurement,suchas"onemeter",isessentiallyananalogywherenumbersandquantitiesactasrepresentationsofpropertiesinnature.
Whenapplyingforajob,apersonmightarguethatshewouldbeagoodsupervisorbecauseshewasalsoagoodbasketballcoach,asthejobshavethepropertyoflead-ershipincommon.
Concerninganimalmodels,arguingfromanalogyusuallyinvolvesmakinginferencesabouthumans,basedonanearlierobservationwhereitwasfoundthattheanimalsandhumanshavesomeprop-ertyincommon.
Arguingfromanalogyisessentiallyapotentiallyerroneousjudgmentbasedonsimilaritiesbetweenentities.
However,thisdoesnotmaketheargu-mentinvalidbydefault,becausethestrengthoftheargu-mentrelieson:(1)howrelevantthepropertyweinferistothepropertythatformsthebasisoftheanalogy;(2)towhatdegreethetwogroupsaresimilar;(3)andifthereisanyvarietyintheobservationsthatformthebasisoftheargument[1].
Animalmodelsthemselvesareanalogies,astheirexist-enceisbasedontheassumptionthattheyaresimilartoatargetgroupinsomerespect.
Ifthetwothingswearedrawinganalogiesonaresimilarenoughsothatwewillreasonablyexpectthemtocorrelate,anargumentfromanalogycanbestrong!
However,whenwedrawthecon-clusionthattwothingsshareacharacteristic,becausewehaveestablishedthattheyalreadyshareanother,differentcharacteristic,thenweareatriskofmakingthefallacyoffalseanalogy[27].
ThefalseanalogyAfalseanalogyisessentiallyaninstancewhenanargu-mentbasedonananalogyisincorrect.
Thiscanoccurwhenthebasisofsimilaritybetweenobjectsdonotjus-tifytheconclusionthattheobjectsaresimilarinsomePage4of13SjobergBehavBrainFunct(2017)13:3otherrespect.
Forinstance,ifJackandJillaresiblings,andJackhasthepropertyofbeingclumsy,wemightinferthatJillisalsoclumsy.
However,wehavenoinformationtoassertthatJillisclumsy,andthepremiseforourargu-mentisbasedsolelyontheobservationthatJackandJillhavegeneticpropertiesincommon.
Weareassumingthatclumsinessishereditary,andthereforethisisprob-ablyafalseanalogy.
Notethatknowledgegainedlatermayindicatethat—infact—clumsinessishereditary,butuntilwehaveobtainedthatknowledgeweareoperat-ingunderassumptionsthatcanleadtofalseanalogies.
Evenifclumsinesswashereditary,wecouldstillnotsaywithabsolutecertaintythatJillisclumsy(unlessgeneticsaccountedfor100%ofthevariance).
Thisnewknowledgewouldmeanthatouranalogyisnolongerfalse,asJill'sclumsinesscanprobablyatleastinpartbeexplainedbygenetics,butwearestillarguingfromanalogy:wecannotknowforcertainifJillisclumsy,basedsolelyonobserva-tionswithJack.
ThefalseanalogyinanimalmodelsWithanimalmodels,thefalseanalogycanoccurwhenonegroup(e.
g.
ananimal)sharesomecharacteristicswithanothergroup(e.
g.
humans),andweassumethatthetwogroupsalsoshareothercharacteristics.
Forinstance,becausechimpanzeescanfollowthegazeofahuman,itcouldbeassumedthatthenon-humanpri-matesunderstandwhatothersperceive,essentiallydis-playingtheoryofmind[28–30].
However,Povinellietal.
[31]arguethatthisisafalseanalogy,becausewearedrawingconclusionsabouttheinnerpsychologicalstateoftheanimal,basedonbehaviouralobservations.
Itmayappearthattheanimalisperformingabehaviourthatrequirescomplexthinking,whileinrealityitonlyremindsusofcomplexthinking[32],mostlikelybecauseweareanthropomorphizingtheanimal'sbehaviour[33]—particularlytheassumptionthatthemindofanapeissimilartothemindofahuman[30].
Adifferentexam-plewouldbebirdsthatareabletomimichumanspeech:thebirdsaresimplyrepeatingsounds,andweareanthro-pomorphisingifwebelievethebirdsactuallygraspourconceptoflanguage.
Robbins[34]pointedoutthathomologyisnotguar-anteedbetweenhumansandprimates,evenifboththebehaviouralparadigmandtheexperimentalresultareidenticalforbothspecies:differentprocessesmayhavebeenusedbythetwospeciestoachievethesameout-come.
Sinceananimalmodelisbasedoncommonprop-ertiesbetweentheanimalandhumans,wemayassumethatnewknowledgegainedfromtheanimalmodelisalsoapplicabletohumans.
Inreality,theresultsareonlyindicativeofevidenceinhumans.
Arguingfromanalogy,therefore,involvestheriskofapplyingknowledgegainedfromtheanimalovertohumans,withoutknowingwithcertaintyifthisapplica-tionistrue.
Imaginethefollowinglineofreasoning:wefindresultAinahumanexperiment,andinananimalmodelwealsofindresultA,establishingfacevalidityfortheanimalmodel.
Consequently,wethenconductadif-ferentexperimentontheanimalmodel,findingresultB.
IfweassumethatBalsoexistinhumans,withouttry-ingtorecreatetheseresultsinhumanexperiments,thenwearearguingfromanalogy,potentiallydrawingafalseanalogy.
Illustration:argumentfromanalogyintheSHRmodelofADHDAnillustrationofargumentfromanalogycomesfromtheSHR(spontaneouslyhypertensiverat)modelofADHD(Attention-Deficit/HyperactivityDisorder)[35,Table1Asummaryofsomeavailableanimalmodelsofmentalillnesses,wheretheanimalsthemselvesactasthemodelforthetargetgroupTheanimalsaregeneticallymodified,bredforaspecifictrait,ormanipulatedinsomephysiologicalfashion(e.
g.
alesionordruginjection)MentalillnessModelReferencesAnxietySerotoninreceptor1Aknockoutmice[114]Corticosteronetreatedmice[115]Attention-Deficit/HyperactivityDisorderSpontaneouslyHypertensiverat[35]Thyroidreceptorβ1transgenicmice[116]AutismValproicAcidrat[81]DepressionCorticosteronetreatedratsandmice[117]ChronicMildStressratsandmice[118]ObsessiveCompulsiveDisorderQuinpiroletreatedrats[119]Post-TraumaticStressSyndromeCongenitallearnedhelplessrat[120]SchizophreniaVentralhippocampuslesionedrats[121]Methylazoxymethanolacetatetreatedrats[122]DevelopmentalvitaminDdeficientrats[123]Page5of13SjobergBehavBrainFunct(2017)13:336].
Comparedtocontrols,usuallytheWistarKyotorat(WKY),theSHRsexhibitmanyofthesamebehaviouraldeficitsobservedinADHDpatients,suchasimpulsivebehaviour[37–42],inattention[35,37],hyperactivity[37,43],andincreasedbehaviouralvariability[44–47].
Onemeasureofimpulsivebehaviourisatestinvolvingdelaydiscounting.
Inthisparadigm,participantsarefacedwiththechoiceofeitherasmall,immediatereinforceroralarger,delayedreinforcer.
BothADHDpatients[48]andSHRs[41]tendtoshowapreferenceforthesmallerreinforcerasthedelaybetweenresponseandreinforcerincreasesforthelargereinforcer.
Researchondelaydis-countingwithADHDpatientssuggeststhattheyaredelayaverse,meaningthatimpulsivityisdefinedasmakingchoicesthatactivelyseektoreducetriallength(oroveralldelay)ratherthanimmediacy[48–56],butthisisusuallyachievedbychoosingareinforcerwithashortdelay.
ThereisnodirectevidencetosuggestthatSHRsoper-atebythesameunderlyingprinciplesasADHDpatients.
StudiesondelaydiscountingusingSHRstendtomanipu-latethedelayperiodbetweenresponseandreinforcerdelivery,butdonotcomparetheresultswithalternativeexplanations.
Thisisbecausetheratscannotbetoldthedetailsoftheprocedure(e.
g.
iftheexperimentendsafteraspecifictimeoraspecificnumberofresponses).
There-fore,mostauthorswhohaveinvestigateddelaydiscount-ingusuallyavoidthetermdelayaversion[57].
However,someauthorsmaketheargumentfromanalogywheretheyassumethattheratsshowasimilareffecttoADHDchildren:Bizotetal.
[58]concludedthat"…SHRarelesspronetowaitforarewardthantheothertwostrains,i.
e.
exhibitahigherimpulsivitylevel…(p.
220)",andPardey,Homewood,TaylorandCornish[59]concludedthat"…SHRsaremoreimpulsivethantheWKYastheyarelesswillingtowaitforanexpectedreinforcer(p.
170).
"EventhoughtheevidenceshowsthatSHRspreferenceforthelargereinforcerdropswithincreaseddelay,wecan-notconcludewithcertaintythatthisoccursbecausetheSHRsdonotwanttowait.
Theexperimentalsetupdoesnottellusanythingconclusiveabouttheanimal'smotiva-tion,noritsunderstandingoftheenvironmentalcondi-tions.
Hayden[60]hasarguedthatthedelaydiscountingtaskisproblematicinmeasuringimpulsivityinanimalsbecauseitisunlikelythattheanimalsunderstandtheconceptoftheinter-trialinterval.
Furthermore,iftheSHRswerelesswillingtowaitforareinforcer,thenwemayarguethatthisshowsimmediacy,andnotnecessar-ilydelayaversion.
Inthiscase,itmayinsteadsupportthedualpathwaymodelofADHD,whichtakesintoaccountbothdelayaversionandanimpulsivedriveforimmediatereward[56,61,62].
Assumingthattheratsaredelayaverseorimpulsiveisarguingfromanalogy.
Theevidencemayonlysuggeststhattheratsareimpulsive,notnecessarilywhytheyareimpulsive.
TheresultsmayalsonotspeaktowhetherthereasonforthisbehaviouristhesameinADHDandSHRs(mechanisticvalidity—see"Mechanisticvalidity"sec-tion).
Ifweweretomanipulatethemagnitudeofthelargereinforcerthenwewillalsofindachangeinperformance[57,63].
HowdoweknowthattheSHRsaresensitivetotemporaldelays,andnottootherchangesintheexperi-mentalsetup,suchastheinter-trialinterval[60],rein-forcermagnitude[63],ortherelativelong-termvalueofthereward[64]ThevaliditycriteriaofanimalmodelsBeforeanyfurtherdiscussiononlogicalfallaciesinani-malmodels,thevaliditycriteriaofthesemodelsmustbeaddressed.
Wemustalsopointoutthattherearetwoapproachestoanimalmodelresearch:(1)validatingaputativeanimalmodel,and(2)conductingresearchonanalreadyvalidatedmodel.
Whenassertingthecriteriaforvalidatinganputativeanimalmodel,thepaperbyWillner[65]isoftencited,claimingthatthecriteriaforavalidanimalmodelrestsonitsface,construct,andpredictivevalidity.
Thismeansthatthemodelmustappeartoshowthesamesymp-tomsasthehumantargetgroup(facevalidity),thattheexperimentmeasureswhatitclaimstomeasureandcanbeunambiguouslyinterpreted(constructvalidity),andthatitcanmakepredictionsaboutthehumanpopula-tion(predictivevalidity).
However,thereisnouniver-sallyacceptedstandardforwhichcriteriamustbemetinorderforananimalmodeltobeconsideredvalid,andthecriteriaemployedmayvaryfromstudytostudy[66–70].
Basedonthis,BelzungandLemoine[71]attemptedtobroadenWillner'scriteriaintoalargerframework,pro-posingninevaliditycriteriathatassessthevalidityofanimalmodelsforpsychiatricdisorders.
TricklebankandGarner[72]havearguedthat,inadditiontothethreecriteriabyWillner[65],agoodanimalmodelmustalsobeevaluatedbasedonhowitcontrolsforthirdvariableinfluences(internalvalidity),towhatdegreeresultscanbegeneralized(externalvalidity),whethermeasuresexpectedtorelateactuallydorelate(convergentvalid-ity),andwhethermeasuresexpectedtonotrelateactu-allydonotrelate(discriminantvalidity).
Theseauthorsarguethatnoknownanimalmodelcurrentlyfulfilsallofthesecriteria,butwemightnotexpectthemto;whatisofutmostimportanceisthatwerecognizethelimitationofananimalmodel,includingitsapplication.
Indeed,itcouldbearguedthatareliableanimalmodelmaynotneedtotickallthevalidityboxesaslongithaspredic-tivevalidity,becauseintheenditsforemostpurposeistomakeempiricalpredictionsaboutitshumantargetgroup.
However,beawarethatarguingfromanalogyPage6of13SjobergBehavBrainFunct(2017)13:3reducesthemodel'spredictivevalidity,becauseitspre-dictivecapabilitiesmaybelimitedtotheanimalstudied.
MechanisticvalidityBehaviouralsimilaritiesbetweenaputativemodelanditshumantargetgroupisnotsufficientgroundstovali-dateamodel.
Inotherwords,facevalidityisnotenough:arguably,mechanisticvalidityismoreimportant.
Thisisatermthatnormallyreferstotheunderlyingcogni-tiveandbiologicalmechanismsofthebehaviouraldefi-citsbeingidenticalinbothanimalsandhumans[71],thoughwecanextendthedefinitiontoincludeexternalvariablesaffectingthebehaviour,ratherthanattributingcausalitytoonlyinternal,cognitiveevents.
Whethertheobservedbehaviourisexplainedintermsofneurologi-calinteractions,cognitiveprocesses,orenvironmentalreinforcementdependsonthecaseinquestion,butthecoreofmatteristhatmechanisticvalidityreferstothecauseoftheobservedbehaviouraldeficitorsymptom.
Ifwecanidentifythecauseoftheobservedbehaviourinananimalmodel,andinadditionestablishthatthisisalsothecauseofthesamebehaviourinhumans,thenwehaveestablishedmechanisticvalidity.
Thisvaliditycrite-riondoesnotspeaktowhathastriggeredtheonsetofacondition(triggervalidity),orwhatmadetheorganismvulnerabletotheconditioninthefirstplace(ontopatho-genicvalidity),butratherwhatfactorsareproducingthespecificsymptomsorbehaviour[71].
Forinstance,fallingdownthestairsmighthavecausedbraininjury(triggervalidity),andthisinjuryinturnreduceddopaminetrans-missioninthebrain,whichleadtoimpulsivebehaviour.
Whenananimalmodelisalsoimpulsiveduetoreduceddopaminetransmissions,wehaveestablishedmechanis-ticvalidity(evenifthetriggerwasdifferent).
ThevalidityofmodelsofconditionswithlimitedetiologyFacevalidityhasbeenarguedtobeofrelativelylowimportanceinananimalmodel,becauseitdoesnotspeakaboutwhythebehaviouroccurs[33,69],i.
e.
theevidenceisonlysuperficial.
However,itcouldbearguedthatfacevalidityisofhigherimportanceinanimalmod-elsofADHD,becausethecompleteetiologyunderlyingtheconditionisnotyetfullyknown,andthereforeanADHDdiagnosisisbasedentirelyonbehaviouralsymp-toms[73].
ThereislimitedknowledgeofthepathophysiologyonmanyofthementalillnessesintheDiagnosticandSta-tisticalManualofMentalDisorders[74];depressionandbipolardisorderareexamplesofheterogeneousconditionswhereanimalmodelshavebeendifficulttoestablish[75,76].
Whendealingwithaheterogeneousmentaldisorder,itisinherentlyharderforanimalmodelstomimicthebehaviouraldeficits,particularlyarangeofdifferentdeficits[75,77–80].
Wecouldargue,therefore,thatmechanisticvalidityinanimalmodelsisdifficult,ifnotimpossible,toestablishfromtheoutsetwhenourknowledgeofcausalityinhumansmightbelimited.
ModelscanbeholisticorreductionistAnimalmodelscanbeapproachedwithdifferentapplica-tionsinmind:itcanaimtoactholisticorreductionist.
Aholisticapproachassumesthatthemodelisagoodrep-resentationofthetargetgroupasawhole,includingallormostsymptomsandbehaviouralorneurologicalchar-acteristics.
Alternatively,areductionistapproachusesananimalmodeltomimicspecificaspectsofatargetgroup,suchasonlyonesymptom.
Thisseparationmaynotbeapparent,becauseanimalmodelsareusuallyaddressedasiftheyareholistic;forinstance,thevalproicacid(VPA)ratmodelofautismistypicallyjustlabelledasan"animalmodelofautism"inthetitleortext[81],butexperimentstypicallyinvestigatespecificaspectsofautism[82–84].
Thisdoesnotmeanthatthemodelisnotholistic,butratherthatitspredictivevalidityislimitedtotheaspectsofautisminvestigatedsofar.
Similarly,theSHRistypi-callylabelledasan"animalmodelofADHD"[35],butithasbeensuggestedthatthemodelisbestsuitedforthecombinedsubtypeofADHD[36,73],whileWistarKyotoratsfromCharlesRiverLaboratoriesaremoresuitedfortheinattentivesubtype[85].
Thepointofthisdistinctionbetweenholisticandreductionistapproachesistounder-linethatanimalmodelshavemanyuses,andfalsifyingamodelinthecontextofonesymptomdoesnotmeanthemodelhasbecomeredundant.
Aslongasthemodelhaspredictivevalidityinoneareaoranother,thenitcanstillgeneratehypothesesandexpandourunderstand-ingofthetargetgroup,evenifthemodelisnotagoodrepresentationofthetargetgroupasawhole.
Indeed,ananimalmodelmayactuallybetreatedasholisticuntilitcanbeempiricallysuggestedthatitshouldinfactbereductionist.
However,researchersshouldtakecarenottoassumethatamodelisholisticbasedonjustafewobservations:thiswouldbearguingfromanalogyandbearstheriskofmakingapplicationsabouthumansthatarecurrentlynotestablishedempirically.
Theexactappli-cationsandlimitationsofananimalmodelshouldalwaysbeclearlydefined[33,86].
AnimalmodelexperimentsarereconstructionsTheterms"replicate"and"reproduce"areoftenusedinterchangeablyintheliterature[87],butwithregardstoanimalmodelstheirdistinctionisparticularlyimportant.
Replicationinvolvesrepeatinganexperimentusingthesamemethodsastheoriginalexperiment,whilearepro-ductioninvolvesinvestigatingthesamephenomenonusingdifferentmethods[88].
ReplicationsassurethatthePage7of13SjobergBehavBrainFunct(2017)13:3effectsarestable,butareproductionisneededtoensurethattheeffectwasnotduetomethodologicalissues.
Wesuggestathirdterm,reconstruction,whichhasspecialapplicationsinanimalmodels.
Areconstructioninvolvesredesigninganexperiment,whilemaintainingtheoriginalhypothesis,inordertoaccommodatediffer-entspecies.
Whenananimalexperimentaimstoinvesti-gateaphenomenonpreviouslyobservedonhumans,wehavetomakecertainchangesforseveralreasons.
First,theanimalsareadifferentspeciesthanhumans,andhaveadifferentphysiologyandlifeexperience.
Second,theanimalsdonotfollowverbalinstructionsandmustoften(butnotalways)betrainedtorespond.
Third,theexperi-mentalsetupmustoftenbeamendedsothatabehaviourequivalenttoahumanbehaviourismeasured.
Afourthobservationisthatanimalstudiestendtousesmallersamplesizesthanhumanexperiments,whichmakesthemmorelikelytoproducelargeeffectsizeswhenasig-nificantresultisfound[89].
Ananimalmodelexperimentactivelyattemptstoreconstructtheconditionsofwhichweobservedaneffectwithhumans,butmakesalterationssothatwecanberelativelycertainthatanequivalenteffectisobservedintheanimals(orviceversa,whereahumanexperimentmeasuresanequivalenteffecttowhatwasobservedinananimalstudy).
Thisquestionstheconstructvalidityofthestudy:howcertainarewethatthetaskaccuratelyreflectsthehumanbehaviourweareinvestigatingAnotherproblemconcernedwithreconstruc-tionisthestandardizationfallacy[90].
Thisreferstothefactthatanimalexperimentsarebestrepli-catedifeveryaspectoftheexperimentisstandard-ized.
However,byincreasingexperimentalcontrolweloseexternalvalidity,meaningthattheresultsarelesslikelytoapplytoothersituations[91].
Thedif-ficultyisthereforetofindabalancebetweenthetwo,andfindingthisbalancemaydependontheresearchquestionweseektoanswer[33,92].
Oneapproachistoinitiallybeginwithreplications,andifthesearesuccessfulmoveontoperformreproductions,andeventuallyreconstructions.
ThisisessentiallywhatvanderStaay,ArndtandNordquist[92]haveprevi-ouslysuggested:successfuldirectreplicationisfol-lowedbyextendedreplicationwheremodificationsaremadewithintheprocedure,theanimal'senviron-ment(e.
g.
housingorrearing),ortheirgender.
Shouldtheeffectpersevere,thenwehavesystematicallyestablishedahigherdegreeofgeneralizationwithoutlosinginternalvalidity.
Atthefinalstage,quasi-repli-cationsareconductedusingdifferentspecies,whichissimilartoourconceptofreconstructions,anditisatthisstagethatthetranslationalvalueofthefind-ingsareevaluated.
ThedoubledowneffectWhenwerunanimalmodelexperiments,wehavetouseacontrolgroupforcomparison.
Whenweareevaluatingaputativemodel,wearethereforeindirectlyevaluatingbothanimalgroupsfortheirappropriatenessasanani-malmodelforthephenomenoninquestion,evenifwehypothesizedbeforehandthatjustonegroupwouldbesuitable,andthisisthedouble-downeffect.
Ifweweretodiscoverthatthecontrolgroup,ratherthantheexperi-mentgroup,showsthepredictedcharacteristic,thenitmaybetemptingtousehindsightbiastorationalizethattheresultwaspredictedbeforehand,somethingthatshouldalwaysbeavoided!
Inactuality,thisisanoccasionthatcanbeusedtomaptheobservablecharacteristicsoftheanimals,whichiscalledphenotyping.
Thismayshowthatthecontrolgrouphasapropertythatmakesthemasuitablecandidateasanewputativemodel.
Follow-upstudiescanthenformallyevaluatewhetherthisputa-tiveanimalmodelhasvalidity.
Thisapproachisperfectlyacceptable,providedthattheinitialdiscoveryofthecon-trolgroup'ssuitabilityisseenassuggestiveandnotcon-clusive,untilfurtherstudyprovidemoreevidence.
Whenananimalmodelhasalreadybeenvalidated,thedouble-downeffectstillapplies:wearestillindirectlyevaluatingtwoanimalgroupsatonce,butitislesslikelythatthatthecontrolgroupwilldisplaytheanimal'schar-acteristicduetopreviousvalidation.
Failuretoreplicatepreviousfindingscanbeinterpretedinmanyways;itcouldbeanerrorinmeasurement,differencesinexperi-mentalmanipulations,orthattheanimalmodelissimplynotsuitableasamodelinthisspecificparadigm(butstillviableinothers).
Shouldweobservethatcontrolsexpressaphenomenonthatwasexpectedoftheexperimentalgroup,thenweshouldreplicatethestudytoruleoutthatthefindingoccurredbychanceorthroughsomemeth-odologicalerror.
Thismayleadustosuggestthecontrolgroupasaputativemodel,pendingfurthervalidation.
ThedoubledowneffectandthefiledrawerproblemSincethepurposeofanimalmodelsistoconductresearchonnon-humananimals,withtheaimtoadvanceknowledgeabouthumans,theninevitablytheanimalmodelandthehumanconditionitmimicsmustbesimi-larinsomerespect.
Iftheywerenot,thenthepursuitofthemodelwouldberedundant.
Therefore,fromtheout-set,thereislikelytobepublicationbiasinfavourofdatathatshowssupportforaputativeanimalmodel,becauseotherwiseithasnoapplications.
Thedouble-downeffectofevaluatingtwoanimalgroupsatoncemakesanimalmodelsparticularlysuscep-tibletothefiledrawerproblem.
Thisisaproblemwheretheliteratureprimarilyreflectspublicationsthatfoundsignificantresults,whilenullresultsarepublishedlessPage8of13SjobergBehavBrainFunct(2017)13:3frequently[93,94].
ThisaversiontothenullcreateswhatFergusonandHeenecalled"undeadtheories",whicharetheoriesthatsurviverejectionindefinitely,becausenullresultsthatrefutethemarenotpublished[95].
Theori-ginofthistrendisnotentirelyclear,butitprobablycameintoexistencebytreatingthepresenceofaphenomenonasmoreinterestingthanitsabsence.
Onceaneffecthasbeendocumented,replicationsmaynowbepublishedthatsupporttheunderlyinghypothesis.
Thefiledrawereffectisprobablyrelatedtothesunk-costeffect:thisisatendencytocontinueonaprojectduetopriorinvestment,ratherthanswitchingtoamorevia-blealternative[96].
Thus,ifwepublishnullresults,itmayseemthatpreviouspublicationswithsignificantfindingswerewasteful,andwemayfeelthatwearecontributingtowardsdissentratherthantowardsfindingsolutions.
Itmaybeintheresearcher'sinteresttofindevidencesup-portingthetheoryinordertojustifytheirinvestedtime,thusbecomingvictimofconfirmationbias.
Furthermore,ifnullresultsarefound,theymightbetreatedwithmoreskepticismthanasignificantresult.
Thisis,ofcourse,afallacyinitselfasbothresultsshouldbetreatedthesame:whywouldanullresultbesubjectedtomorescrutinythanasignificantresultWhentheCERNfacilityrecordedparticlestravellingfasterthanthespeedoflight,theobservationappearedtofalsifythethe-oryofrelativity[97].
Thisresultwasmetwithskepticism[98],anditwasassumedthatitwasduetoameasure-menterror(whichintheenditturnedouttobe).
Nev-ertheless,iftheresulthadsupportedrelativity,wouldthedegreeofskepticismhavebeenthesameInthecontextofanimalstudies,thedouble-downeffectmakesitmorelikelythatasignificantresultisfoundwhencomparingtwoanimalgroups.
Eithergroupmaybeasuitablecandidateforaputativeanimalmodel,evenifonlyonegroupwaspredictedtobesuit-ablebeforehand.
Ifanyresultotherthananullresultwillshowsupportforananimalmodel(oraputativemodel),thenmultipleviablemodelswillbepresentinthelitera-ture,allofwhichwillbehardtofalsify(asfalsifyingonemodelmaysupportanother).
Indeed,thisiscurrentlythecaseforanimalmodels,wheretherearemultipleavail-ablemodelsforthesamehumanconditions[80,99–103].
Thefiledrawerproblemisaseriousissueinscience[104],andthetrendmayoftenbeinvisibletothenakedeye,butmethodssuchasmeta-analyseshavemultipletoolstohelpdetectpublicationbiasintheliterature[105].
MeasurestoimproveanimalmodelresearchThemainpurposeofthispaperwastoaddresssev-eralrisksandfallaciesthatmayoccurinanimalmodelresearch,inordertoencouragearigorousscientificpursuitinthisfield.
Wedonotintendtodiscourageresearchersfromusinganimalmodels,butratherhopetoincreaseawarenessofpotentialrisksandfalla-ciesinvolved.
Inordertomaketheissuesaddressedinthepapermoreoverviewable,wehavecreatedalistforresearcherstoconferwhendesigninganimalexperimentandinterpretingtheirdata.
1.
Beawareofyourownlimitations.
Someofthefalla-ciesandrisksaddressedinthispapermaybeuna-voidableforavarietyofreasons.
Nevertheless,thefirststeptowardsimprovingone'sresearchistobeawareoftheexistenceoftheserisks.
Whenwritingthediscussionsectionofareport,itmaybeneces-sarytopointoutpossiblelimitations.
Eveniftheyarenotexplicitlystated,itisstillhealthyforanyscientisttobeawareofthem.
32.
Establishpredictiveandmechanisticvalidity.
Ifyouareattemptingtovalidateaputativeanimalmodel,ensurethattheexperimentisassimilaraspos-sibletoexperimentsdoneonhumans.
Ifthisisnotpossible,explainwhyinthewrite-up.
Iftheexperi-mentisnovel,andtheanimalmodelisalreadyvali-datedthroughpreviousresearch,thenthisprincipledoesnotnecessarilyapply,becausethepurposeistouncovernewknowledgethatmaybetranslatedtohumans.
Insuchinstances,anewhypothesisgainsvalidityinafollow-upexperimentonhumans.
Rememberthatthereareseveralcriteriaavailableforvalidatingananimalmodel,butthereisnouni-versalagreementonwhichsetofcriteriashouldbefollowed.
However,thetwomostimportantcrite-riaarearguablypredictivevalidityandmechanisticvalidity,becausefacevalidityispronetologicalfal-lacies.
Establishingmechanisticvalidityensuresthatthemechanismscausingtheobservedbehaviourarethesameinthemodelandhumans,whileestablish-ingpredictivevaliditymeansthatknowledgegainedfromthemodelismorelikelytoapplytohumans.
3.
Defineanapriorihypothesisandplanthestatisticalanalysisbeforehand.
Itiscrucialtohaveanapriorihypothesispriortoconductingtheexperiment,otherwiseonemightbeaccusedofdatadredgingandreasoningafter-the-factthattheresultswereexpected[107,108].
Whenvalidatingaputativeani-malmodel,thisdrasticallyreducesthedouble-downeffect.
Ifthedatadonotshowthepredictedpat-ternthenitisperfectlyacceptabletosuggestanew3Theauthorofthismanuscriptonceheldaconferencetalkwherehesug-gestedthepossibilitythatoneofhisownresearchresultsmayhavebeeninfluencedbyconfirmationbias[106].
Neverassumethatonlyothersarepronetobias—evenauthorsoflogicalfallacypapersmaycommitfallacies!
.
Page9of13SjobergBehavBrainFunct(2017)13:3hypothesisand/oraputativeanimalmodelforfur-therresearch.
Whendesigningtheexperiment,keepinmindwhichstatisticalanalysiswouldbeappropriateforanalysingthedata.
Ifthestatisticalmethodischosenposthoc,thenitmaynotcorrespondtothechosendesign,andonemightbeaccusedofdatadredging,whichinvolveschoosingastatisticalprocedurethatismorelikelytoproducesignificantresults[107].
Also,keepinmindwhichposthoctestsareplanned,andthatthecorrectoneischosentoreducefamilywiseerrorwhentherearemultiplecomparisonstobemade.
Itishighlyrecommendedthateffectsizesarereportedforeverystatisticaltest:thiswillgiveinsightintothestrengthoftheobservedphenomenon,andalsoallowamoredetailedcomparisonbetweenstudies[109].
4.
Doapoweranalysis.
Forlogistical,practical,oreco-nomicreasons,animalmodelresearchmaybeforcedtousesamplesizessmallerthanwhatisideal.
Nev-ertheless,oneshouldconductapoweranalysistoascertainhowmanyanimalsshouldbetestedbeforetheexperimentstarts.
Whendoingmultiplecom-parisons,itmaybedifficulttoestablishthesamplesizebecausethepoweranalysismayonlygrantthesamplesizeofanomnibusanalysis(theanalysisofthewhole,notitsindividualparts),andnotwhatisrequiredtoreachsignificancewithposthoctests[110].
Ifalltheposthocanalysesareofequalinterest,choosethesamplesizerequiredtoachievepowerof0.
8inallcomparisons.
Alternatively,useacompar-ison-of-most-interestapproachwherethesamplesizeisdeterminedbythepoweranalysisoftheposthoccomparisonthatisofhighestinterest[110].
Ifapoweranalysisisnotconducted,ornotadheredto,itmaybeprudenttouseasamplesizesimilartopreviouslyconductedexperimentsintheliterature,andthendoaposthocpoweranalysistodeterminethepowerofyourstudy.
Oncetheexperimentiscompletedandthedataanalysed,onemustneverincreasethesamplesize,becausethiswillincreaseyourchancesoffindingasignificantresult(confirma-tionbias)[109,111,112].
5.
Double-blindtheexperiment.
Bydoingtheexperi-mentdouble-blind,weseverelyreducetheriskofconfirmationbias.
Thismeansthattheexperimenterisblindtotheapriorihypothesisofthestudy,aswellaswhatgroupeachanimalbelongsto.
How-ever,insomecasesitmaybedifficultorimpossibletodothis.
Forinstance,iftheexperimentalgrouphasaphenotypethatdistinguishesthemfromcon-trols(e.
g.
whitevs.
brownrats),thenitisdifficulttoblindtheexperimenter.
Forlogisticalandmonetaryreasonsitmayalsobeimpracticaltohaveaqualifiedexperimenterwhoisblindtotherelevantliteratureofthestudy.
Also,avoidanalysingdatapriortotheexperiment'scompletion,becauseifthedataarenotinlinewithyourpredictionsthenonemightimplic-itlyinfluencetheexperimenttogetthedataneeded(experimenterbias[14]).
Beawarethatitisneverthe-lessperfectlyacceptabletoinspectthedataonocca-sionwithoutstatisticallyanalysingit,justtoensurethattheequipmentisworkingasitissupposedto(orstateinadvanceatwhatpointitisacceptabletocheckthedata,incasetherearecircumstanceswhereyoumaywanttoterminatetheexperimentearly).
6.
Avoidanthropomorphizing.
Whileitisinevitabletodescribeourresultsinthecontextofhumanunder-standingandlanguage,wemustbecarefulnottoattributetheanimalswithhuman-likequalities.
Avoidmakinginferencesabouttheanimal'sthoughts,feelings,innermotivation,orunderstandingofthesituation.
Wecanreportwhattheanimalsdid,andwhatthismeansinthecontextofourhypothesis,buttakecarenottomakeassumptionsoftheinnerwork-ingsoftheanimal.
7.
Avoidarguingfromanalogy.
Nomatterhowvali-datedananimalmodelis,wecannotbecertainthatanewlyobservedeffectalsoappliestohumans.
Ifresearchonananimalmodelyieldsnewinformationthatcouldgiveinsightintothehumantargetgroup,ensuretomentionthatthedataissuggestive,notconclusive,pendingfurthervalidation.
Rememberthatthestrengthofananimalmodelistogeneratenewknowledgeandhypothesesrelevanttothetargetgroup,includingtheassessmentofpotentiallyusefultreatments,butthatthesenewpossibilitiesareonlyhypotheticaloncetheyarediscovered.
8.
Attempttopublish,despiteanullresult.
Ifyoupre-dictedaspecificresultbasedontrendsinthelitera-ture,butfailedtofindthisresult,donotbediscour-agedfrompublishingthedata(especiallyifyoufailedtoreplicatearesultinaseriesofexperiments).
Thisisparticularlyimportantiftheexperimenthadalowsamplesize,asnullresultsfromsuchstudiesareprobablytheleastlikelytobepublished,thusfuellingthefiledrawerproblem.
Bymakingthedataavail-ableviaeitheranarticle(forinstancethroughJour-nalofArticlesinSupportoftheNullHypothesis)oradatasetonline,thenyouareactivelycontributingtoreducethefiledrawerproblem.
9.
Replicate,reproduce,andreconstruct.
Replicatinganexperimentinordertoestablishintervalvalidityandreliabilityofananimalmodelisessential.
Whenrep-licatingexperimentsmultipletimes,wereducetheriskthattheoriginalfindingwasachanceresult.
Ifpreviousreplicationshavesucceeded,thenattemptPage10of13SjobergBehavBrainFunct(2017)13:3toincludeanewhypothesis,experimentalmanipu-lation,orfollow-upexperimentduringthestudytoexpandourknowledgeoftheresearchquestion.
Thisprocessestablishesbothinternalandexternalvalid-ity.
Finally,reconstructtheexperimentonhumans,sothatthefindingsmaybeattributedacrossspecies.
AnoteonneurologicalsimilaritiesTheprinciplesdiscussedinthispaperhavebeenaddressedinabehaviouralcontext,butitshouldbenotedthattheyalsoapplytoneurologicalevidenceforanimalmodels,thoughincreasingthevalidityinthiscasecanoperatesomewhatdifferently.
Whenwefindneurologicalelementsthatarethesameinboththeanimalmodelandthehumantargetgroup(thatdonotexistincontrols),weshouldbecarefultodrawanyconclusionsbasedonthis.
Justlikebehaviouralevidence,thelinksaresuggestiveandnotnecessarilycon-clusive.
Itisriskytoassumethatthephysiologicalprop-ertiessharedbetweenhumansandanimalsoperatethesameway.
Indrugresearch,over90%ofdrugsthatshoweffectivenessonanimalmodelsfailtoworkonhumans,aproblemcalledattrition[113].
Inthecontextofanimalmodelsofmentalillness,BelzungandLemoine[71]pro-posedtheconceptbiomarkervalidity,whichmeansthatthefunctionofaneurologicalmechanismisthesameintheanimalmodelandhumans,evenifthebiomarkerresponsibleforthisfunctionmaybedifferentacrossthespecies.
Inotherwords,thetwospeciesmayhavediffer-entbiologicalmarkers,butaslongastheyoperatethesameway,andinturnproducethesamesymptoms,thenthisaddsvaliditytothemodel.
Ofcourse,inrealitythingsarenotthissimple.
Neuro-logicalevidenceisusuallynotbasedonthepresenceofasinglecomponent,butrathermultipleelementssuchasrateofneurotransmitterrelease,reuptake,polymor-phism,neuralpathways,drugeffectiveness,oracombi-nationoffactors.
Thecoremessageisthatwemustbeawarethatfindingsimilarneurologicalelementsinbothanimalsandhumansdoesnotmeanthattheyoperatethesameway.
Ifwemakethisassumption,wearearguingfromanalogy.
Itshouldbenotedthatconfirmationbiascouldalsobeaproblematicissueinneuroscientificresearch.
Garner[113]illustratesthiswithacarexample:ifwebelievethatthegaspedalofacaristhecauseofcaraccidents,thenremovingthegaspedalfromacarwilldrasticallyreducetheaccidentrateofthatcar,confirmingthatindeedthegaspedalwasthecauseofcaraccidents.
Inneuroscience,wemayknockoutageneorselectivelybreedstrainstoaddorremoveageneticcomponent.
Whenthehypoth-esizedbehaviourisshown(ornotshown),wemightconcludethatwehaveconfirmedourhypothesis.
Theconclusioncouldbewrongbecauseitisbasedoncorrela-tion,andthusfuturereplicationsofthisresultislikelytomakethesamelogicalerror[113].
ClosingremarksInthispaper,ithasbeendiscussedhowanimalmodelscanbesusceptibletologicalfallacies,bias,andariskofgettingresultsthatcouldgiveafalsesenseofsupportforaputativeanimalmodel.
Researchersshouldrememberthatbehaviouralresultsfoundinananimalmodelofahumanconditiondoesnotguaranteethatthisknowl-edgeisapplicabletohumans.
Replicating,reproducingandreconstructingresultsovernumerousstudieswilldrasticallyreducetheprobabilitythattheresultsaresimilarbychancealone,althoughthisdoesnotnecessar-ilyshedlightonwhythebehaviouroccurs.
Researchersshouldthereforebeencouragedtoinvestigatemecha-nisticvalidity,meaningwhatunderlyingprocessesarecausingthebehaviour.
Bysimplylookingatfacevalid-ity,wehaveanincreasedriskofmakingerrorsthroughcomparisons.
Animalmodelscanbeveryusefulforinvestigatingthemechanismsbehindahumancondition.
Thisnewknowl-edgecanhelpimproveourunderstandingandtreatmentofthiscondition,buttheresearchermustnotassumethattheobservedanimalbehaviouralsoappliestohumans.
Ultimately,animalmodelsonlyprovidesolidevidencefortheanimalused,andindicativeevidenceofhumanbehav-iour.
However,thisisalsothestrengthofanimalmodels:indicativeevidencemayopenthedoortonewideasabouthumanbehaviourthatwerenotpreviouslyconsidered.
Throughreconstructions,itcanbeestablishedwhetherornotthephenomenonexistsinhumans,andifthemodelhasmechanisticvalidityandpredictivevaliditythenthiscertainlyincreasestheapplicationofthemodel,aswellasitsvaluefortheprogressofhumanhealth.
AbbreviationsADHD:Attention-Deficit/HyperactivityDisorder;CERN:EuropeanOrganiza-tionforNuclearResearch;DSM:DiagnosticandStatisticalManualofMentalDisorders;SHR:spontaneouslyhypertensiverat;VPA:valproicacidrat;WKY:WistarKyotorat.
AcknowledgementsRachaelWilnergavevaluableinsightandfeedbackthroughoutmultiplever-sionsofthemanuscript,especiallyintoimprovingthelanguageandstructureofthepaper,aswellasclarifyingseveralarguments.
AconversationwithysteinVogtwaslargelyinspirationalintermsofwritingthisarticle.
MagnusH.
Blystadgavefeedbackthatsubstantiatedseveralclaims,particularlytheneurologysection.
EspenBorgJohansenofferedcriticalinputonseveraloccasions,whichleadtosomeargumentsbeingempiricallystrengthened.
CarstaSimon'sfeedbackimprovedsomeofthedefinitionsemployedinthearticle.
OthermembersoftheresearchgroupExperimentalBehaviorAnalysis:TranslationalandConceptualResearch,OsloandAkershusUniversityCollege,istobethankedfortheircontributionandfeedback,particularlyPerHolth,RasmiKrippendorf,andMonicaVandbakk.
Page11of13SjobergBehavBrainFunct(2017)13:3CompetinginterestsTheauthordeclarethathehasnocompetinginterests.
Received:21January2016Accepted:1February2017References1.
SalmonM.
Introductiontologicandcriticalthinking.
Boston:Wads-worthCengageLearning;2013.
2.
BarnesB.
Aboutscience.
NewYork:BasilBlackwellInc.
;1985.
3.
KernLH,MirelsHL,HinshawVG.
Scientists'understandingofpropositionallogic:anexperimentalinvestigation.
SocStudSci.
1983;13:131–46.
4.
TverskyA,KahnemanD.
Extensionalversusintuitivereasoning:thecon-junctionfallacyinprobabilityjudgment.
PsycholRev.
1983;90:293–315.
5.
KahnemanD.
Thinking,fastandslow.
London:Macmillan;2011.
6.
HallerH,KraussS.
Misinterpretationsofsignificance:aproblemstu-dentssharewiththeirteachers.
MethodsPsycholRes.
2002;7:1–20.
7.
Badenes-RiberaL,Frías-NavarroD,Monterde-i-BortH,Pascual-SolerM.
InterpretationofthePvalue:anationalsurveystudyinacademicpsychologistsfromSpain.
Psicothema.
2015;27:290–5.
8.
PopperKR.
TheLOGICOFSCIENTIfiCDISCOVery.
London:Hutchinson;1959.
9.
LewensT.
Themeaningofscience.
London:Pelican;2015.
10.
LeaheyTH.
Themythicalrevolutionsofamericanpsychology.
AmPsychol.
1992;47:308–18.
11.
LawS.
Thegreatphilosophers.
London:Quercus;2007.
12.
KeuthH.
ThePhilosophyofKarlPopper.
Cambridge:CambridgeUniver-sityPress;2005.
13.
NickersonRS.
Confirmationbias:aubiquitousphenomenoninmanyguises.
RevGenPsychol.
1998;2:175.
14.
RosenthalR,FodeKL.
Theeffectofexperimenterbiasontheperfor-manceofthealbinorat.
BehavSci.
1963;8:183–9.
15.
InglisM,SimpsonA.
Mathematiciansandtheselectiontask.
In:Pro-ceedingsofthe28thinternationalconferenceonthepsychologyofmathematicseducation;2004.
p.
89–96.
16.
JacksonSL,GriggsRA.
Educationandtheselectiontask.
BullPsychonSoc.
1988;26:327–30.
17.
HergovichA,SchottR,BurgerC.
Biasedevaluationofabstractsdepend-ingontopicandconclusion:furtherevidenceofaconfirmationbiaswithinscientificpsychology.
CurrPsychol.
2010;29:188–209.
18.
MahoneyMJ.
Scientistassubject:thepsychologicalimperative.
Phila-delphia:Ballinger;1976.
19.
SidmanM.
Tacticsofscientificresearch.
NewYork:BasicBooks;1960.
20.
MooreJ.
Aspecialsectioncommemoratingthe30thanniversaryoftac-ticsofscientificresearch:evaluatingexperimentaldatainpsychologybyMurraySidman.
BehavAnal.
1990;13:159.
21.
HolthP.
Aresearchpioneer'swisdom:aninterviewwithDr.
MurraySid-man.
EurJBehavAnal.
2010;12:181–98.
22.
MichaelJ.
Flightfrombehavioranalysis.
BehavAnal.
1980;3:1.
23.
vanWilgenburgE,ElgarMA.
Confirmationbiasinstudiesofnestmaterecognition:acautionarynoteforresearchintothebehaviourofani-mals.
PLoSONE.
2013;8:e53548.
24.
PoddarR,KawaiR,lveczkyBP.
Afullyautomatedhigh-throughputtrainingsystemforrodents.
PLoSONE.
2013;8:e83171.
25.
JiangH,HannaE,GattoCL,PageTL,BhuvaB,BroadieK.
Afullyauto-mateddrosophilaolfactoryclassicalconditioningandtestingsystemforbehaviorallearningandmemoryassessment.
JNeurosciMethods.
2016;261:62–74.
26.
OswaldME,GrosjeanS.
Confirmationbias.
In:PohlR,editor.
Cognitiveillusions:ahandbookonfallaciesandbiasesinthinking,judgementandmemory.
Hove:PsychologyPress;2004.
p.
79.
27.
MillJS.
Asystemoflogic.
London:JohnW.
Parker;1843.
28.
PremackD,WoodruffG.
DoesthechimpanzeehaveatheoryofmindBehavBrainSci.
1978;1:515–26.
29.
CallJ,TomaselloM.
Doesthechimpanzeehaveatheoryofmind30yearslater.
TrendsCognSci.
2008;12:187–92.
30.
GomezJ-C.
Non-humanprimatetheoriesof(non-humanprimate)minds:someissuesconcerningtheoriginsofmind-reading.
In:Car-ruthersP,SmithPK,editors.
Theoriesoftheoriesofmind.
Cambridge:CambridgeUniversityPress;1996.
p.
330.
31.
PovinelliDJ,BeringJM,GiambroneS.
Towardascienceofotherminds:escapingtheargumentbyanalogy.
CognSci.
2000;24:509–41.
32.
DuttonD,WilliamsC.
Aviewfromthebridge:subjectivity,embodimentandanimalminds.
Anthrozos.
2004;17:210–24.
33.
vanderStaayFJ,ArndtSS,NordquistRE.
Evaluationofanimalmodelsofneurobehavioraldisorders.
BehavBrainFunct.
2009;5:11.
34.
RobbinsT.
Homologyinbehaviouralpharmacology:anapproachtoanimalmodelsofhumancognition.
BehavPharmacol.
1998;9:509–19.
35.
SagvoldenT.
Behavioralvalidationofthespontaneouslyhypertensiverat(Shr)asananimalmodelofattention-deficit/hyperactivitydisorder(Ad/Hd).
NeurosciBiobehavRev.
2000;24:31–9.
36.
SagvoldenT,JohansenEB,WienG,WalaasSI,Storm-MathisenJ,BergersenLH,etal.
ThespontaneouslyhypertensiveratmodelofADHD—theimportanceofselectingtheappropriatereferencestrain.
Neuropharmacology.
2009;57:619–26.
37.
SagvoldenT,AaseH,ZeinerP,BergerD.
Alteredreinforcementmecha-nismsinattention-deficit/hyperactivitydisorder.
BehavBrainRes.
1998;94:61–71.
38.
WultzB,SagvoldenT.
Thehyperactivespontaneouslyhypertensiveratlearnstositstill,butnottostopburstsofresponseswithshortinter-responsetimes.
BehavGenet.
1992;22:415–33.
39.
Malloy-DinizL,FuentesD,LeiteWB,CorreaH,BecharaA.
Impulsivebehaviorinadultswithattentiondeficit/hyperactivitydisorder:char-acterizationofattentional,motorandcognitiveimpulsiveness.
JIntNeuropsycholSoc.
2007;13:693–8.
40.
EvendenJL.
ThepharmacologyofimpulsivebehaviourinratsIv:theeffectsofselectiveserotonergicagentsonapacedfixedconsecutivenumberschedule.
Psychopharmacology.
1998;140:319–30.
41.
FoxAT,HandDJ,ReillyMP.
Impulsivechoiceinarodentmodelofatten-tion-deficit/hyperactivitydisorder.
BehavBrainRes.
2008;187:146–52.
42.
Sonuga-BarkeEJ.
PsychologicalheterogeneityinAd/Hd—adualpathwaymodelofbehaviourandcognition.
BehavBrainRes.
2002;130:29–36.
43.
BergerDF,SagvoldenT.
Sexdifferencesinoperantdiscriminationbehaviourinananimalmodelofattention-deficithyperactivitydisor-der.
BehavBrainRes.
1998;94:73–82.
44.
UebelH,AlbrechtB,AshersonP,BrgerNA,ButlerL,ChenW,etal.
Performancevariability,impulsivityerrorsandtheimpactofincentivesasgender-independentendophenotypesforADHD.
JChildPsycholPsychiatry.
2010;51:210–8.
45.
JohansenEB,KilleenPR,SagvoldenT.
Behavioralvariability,eliminationofresponses,anddelay-of-reinforcementgradientsinShrandWkyrats.
BehavBrainFunct.
2007;3:1.
46.
AdrianiW,CaprioliA,GranstremO,CarliM,LaviolaG.
Thespontane-ouslyhypertensive-ratasananimalmodelofADHD:evidenceforimpulsiveandnon-impulsivesubpopulations.
NeurosciBiobehavRev.
2003;27:639–51.
47.
ScheresA,OosterlaanJ,SergeantJA.
Responseexecutionandinhibi-tioninchildrenwithAD/HDandotherdisruptivedisorders:theroleofbehaviouralactivation.
JChildPsycholPsychiatry.
2001;42:347–57.
48.
Sonuga-BarkeE,TaylorE,SembiS,SmithJ.
Hyperactivityanddelayaversion—I.
Theeffectofdelayonchoice.
JChildPsycholPsychiatry.
1992;33:387–98.
49.
Sonuga-BarkeEJ,WilliamsE,HallM,SaxtonT.
HyperactivityanddelayaversionIII:theeffectoncognitivestyleofimposingdelayaftererrors.
JChildPsycholPsychiatry.
1996;37:189–94.
50.
KuntsiJ,OosterlaanJ,StevensonJ.
Psychologicalmechanismsinhyperactivity:Iresponseinhibitiondeficit,workingmemoryimpair-ment,delayaversion,orsomethingelseJChildPsycholPsychiatry.
2001;42:199–210.
51.
SolantoMV,AbikoffH,Sonuga-BarkeE,SchacharR,LoganGD,WigalT,etal.
Theecologicalvalidityofdelayaversionandresponseinhibi-tionasmeasuresofimpulsivityinAD/HD:asupplementtotheNIMHmultimodaltreatmentstudyofAD/HD.
JAbnormChildPsychol.
2001;29:215–28.
Page12of13SjobergBehavBrainFunct(2017)13:352.
DalenL,Sonuga-BarkeEJ,HallM,RemingtonB.
Inhibitorydeficits,delayaversionandpreschoolAD/HD:implicationsforthedualpathwaymodel.
NeuralPlast.
2004;11:1–11.
53.
BitsakouP,PsychogiouL,ThompsonM,Sonuga-BarkeEJ.
Delayaversioninattentiondeficit/hyperactivitydisorder:anempiricalinvestigationofthebroaderphenotype.
Neuropsychologia.
2009;47:446–56.
54.
TrippG,AlsopB.
Sensitivitytorewarddelayinchildrenwithatten-tiondeficithyperactivitydisorder(ADHD).
JChildPsycholPsychiatry.
2001;42:691–8.
55.
MarxI,HübnerT,HerpertzSC,BergerC,ReuterE,KircherT,etal.
Cross-sectionalevaluationofcognitivefunctioninginchildren,adolescentsandyoungadultswithADHD.
JNeuralTransm.
2010;117:403–19.
56.
MarcoR,MirandaA,SchlotzW,MeliaA,MulliganA,MüllerU,etal.
DelayandrewardchoiceinADHD:anexperimentaltestoftheroleofdelayaversion.
Neuropsychology.
2009;23:367–80.
57.
GarciaA,KirkpatrickK.
Impulsivechoicebehaviorinfourstrainsofrats:evaluationofpossiblemodelsofattentiondeficit/hyperactivitydisor-der.
BehavBrainRes.
2013;238:10–22.
58.
BizotJ-C,ChenaultN,HouzéB,HerpinA,DavidS,PothionS,etal.
MethylphenidatereducesimpulsivebehaviourinJuvenileWistarrats,butnotinadultWistar,ShrandWkyrats.
Psychopharmacology.
2007;193:215–23.
59.
PardeyMC,HomewoodJ,TaylorA,CornishJL.
Re-evaluationofananimalmodelforADHDusingafree-operantchoicetask.
JNeurosciMethods.
2009;176:166–71.
60.
HaydenBY.
Timediscountingandtimepreferenceinanimals:acriticalreview.
PsychonBullRev.
2015;23:1–15.
61.
ScheresA,DijkstraM,AinslieE,BalkanJ,ReynoldsB,Sonuga-BarkeE,etal.
Temporalandprobabilisticdiscountingofrewardsinchildrenandadolescents:effectsofageandADHDsymptoms.
Neuropsychologia.
2006;44:2092–103.
62.
Sonuga-BarkeEJ,SergeantJA,NiggJ,WillcuttE.
Executivedysfunctionanddelayaversioninattentiondeficithyperactivitydisorder:nosologicanddiagnosticimplications.
ChildAdolescPsychiatrClinNAm.
2008;17:367–84.
63.
BotanasCJ,LeeH,delaPeaJB,delaPeaIJ,WooT,KimHJ,etal.
Rearinginanenrichedenvironmentattenuatedhyperactivityandinattentioninthespontaneouslyhypertensiverats,ananimalmodelofattention-deficithyperactivitydisorder.
PhysiolBehav.
2016;155:30–7.
64.
SjobergEA,HolthP,JohansenEB.
theeffectofdelay,utility,andmagni-tudeondelaydiscountinginananimalmodelofattention-deficit/hyper-activitydisorder(ADHD):asystematicreview.
In:Associationofbehavioranalysisinternational42ndannualconvention.
Chicago,IL;2016.
65.
WillnerP.
Validationcriteriaforanimalmodelsofhumanmentaldisor-ders:learnedhelplessnessasaparadigmcase.
ProgNeuropsychophar-macolBiolPsychiatry.
1986;10:677–90.
66.
GeyerMA,MarkouA.
Animalmodelsofpsychiatricdisorders.
In:BloomFE,KupferDJ,editors.
Psychopharmacology:thefourthgenerationofprogress.
NewYork:RavenPress;1995.
p.
787–98.
67.
McKinneyW.
Animalmodelsofdepression:anoverview.
PsychiatrDev.
1983;2:77–96.
68.
KoobGF,HeinrichsSC,BrittonK.
Animalmodelsofanxietydisorders.
In:SchatzbergAF,NemeroffCB,editors.
TheAmericanPsychiatricPresstextbookofpsychopharmacology.
2nded.
Washington:AmericanPsychiatricPress;1998.
p.
133–44.
69.
SarterM,BrunoJP.
Animalmodelsinbiologicalpsychiatry.
In:D'HaenenH,denBoerJA,WillnerP,editors.
Biologicalpsychiatry.
Chichester:Wiley;2002.
p.
37–44.
70.
WeissJM,KiltsCD.
Animalmodelsofdepressionandschizophrenia.
In:SchatzbergAF,NemeroffCB,editors.
TheAmericanPsychiatricPresstextbookofpsychopharmacology.
2nded.
Washington:AmericanPsychiatricPress;1998.
p.
89–131.
71.
BelzungC,LemoineM.
Criteriaofvalidityforanimalmodelsofpsychi-atricdisorders:focusonanxietydisordersanddepression.
BiolMoodAnxietyDisord.
2011;1(1):9.
doi:10.
1186/2045-5380-1-9.
72.
TricklebankM,GarnerJ.
Thepossibilitiesandlimitationsofanimalmodelsforpsychiatricdisorders.
Cambridge:RSCDrugDiscoveryRoyalSocietyofChemistry;2012.
p.
534–57.
73.
SagvoldenT,JohansenEB.
RatmodelsofADHD.
In:StanfordC,TannockR,editors.
Behavioralneuroscienceofattention-deficit/hyperactivitydisorderanditstreatments.
Berlin:Springer;2012.
p.
301–15.
74.
AssociationAP.
Diagnosticandstatisticalmanualofmentaldisorders(Dsm-5).
ArlingtonCounty:AmericanPsychiatricPub;2013.
75.
NestlerEJ,HymanSE.
Animalmodelsofneuropsychiatricdisorders.
NatNeurosci.
2010;13:1161–9.
76.
GouldTD,EinatH.
Animalmodelsofbipolardisorderandmoodstabi-lizerefficacy:acriticalneedforimprovement.
NeurosciBiobehavRev.
2007;31:825–31.
77.
KaratekinC.
AcomprehensiveanddevelopmentaltheoryofADHDistantalizing,butpremature.
BehavBrainSci.
2005;28:430–1.
78.
WillcuttEG,DoyleAE,NiggJT,FaraoneSV,PenningtonBF.
Validityoftheexecutivefunctiontheoryofattention-deficit/hyperactivitydisorder:ameta-analyticreview.
BiolPsychiatry.
2005;57:1336–46.
79.
EinatH,ManjiHK.
Cellularplasticitycascades:genes-to-behaviorpathwaysinanimalmodelsofbipolardisorder.
BiolPsychiatry.
2006;59:1160–71.
80.
SontagTA,TuchaO,WalitzaS,LangeKW.
Animalmodelsofattentiondeficit/hyperactivitydisorder(ADHD):acriticalreview.
ADHDAttenDeficitHyperactDisord.
2010;2:1–20.
81.
SchneiderT,PrzewockiR.
Behavioralalterationsinratsprenatallyexposedtovalproicacid:animalmodelofautism.
Neuropsychophar-macology.
2005;30:80–9.
82.
MehtaMV,GandalMJ,SiegelSJ.
Mglur5-antagonistmediatedreversalofelevatedstereotyped,repetitivebehaviorsintheVPAmodelofautism.
PLoSONE.
2011;6:e26077.
83.
MarkramK,RinaldiT,LaMendolaD,SandiC,MarkramH.
Abnormalfearconditioningandamygdalaprocessinginananimalmodelofautism.
Neuropsychopharmacology.
2008;33:901–12.
84.
SnowWM,HartleK,IvancoTL.
AlteredmorphologyofmotorcortexneuronsintheVPAratmodelofautism.
DevPsychobiol.
2008;50:633–9.
85.
SagvoldenT,DasbanerjeeT,Zhang-JamesY,MiddletonF,FaraoneS.
Behavioralandgeneticevidenceforanovelanimalmodelofattention-deficit/hyperactivitydisorderpredominantlyinattentivesubtype.
BehavBrainFunct.
2008;4:b54.
86.
vanderStaayFJ.
Animalmodelsofbehavioraldysfunctions:basicconceptsandclassifications,andanevaluationstrategy.
BrainResRev.
2006;52:131–59.
87.
GómezO,JuristoN,VegasS.
Replication,reproductionandre-analysis:threewaysforverifyingexperimentalfindings.
In:Proceedingsofthe1stinternationalworkshoponreplicationinempiricalsoftwareengi-neeringresearch(RESER2010).
CapeTown,SouthAfrica;2010.
88.
CartwrightN.
Replicability,reproducibility,androbustness:commentsonHarryCollins.
HistPolitEcon.
1991;23:143–55.
89.
SlavinR,SmithD.
Therelationshipbetweensamplesizesandeffectsizesinsystematicreviewsineducation.
EducEvalPolicyAnal.
2009;31:500–6.
90.
WürbelH.
Behaviourandthestandardizationfallacy.
NatGenet.
2000;26:263.
91.
RichterSH,GarnerJP,WürbelH.
Environmentalstandardization:cureorcauseofpoorreproducibilityinanimalexperimentsNatMethods.
2009;6:257–61.
92.
JosefvanderStaayF,ArndtS,NordquistR.
Thestandardization-general-izationdilemma:awayout.
GenesBrainBehav.
2010;9:849–55.
93.
RosenthalR.
Thefiledrawerproblemandtolerancefornullresults.
PsycholBull.
1979;86:638.
94.
SterlingTD.
Publicationdecisionsandtheirpossibleeffectsoninfer-encesdrawnfromtestsofsignificance—orviceversa.
JAmStatAssoc.
1959;54:30–4.
95.
FergusonCJ,HeeneM.
Avastgraveyardofundeadtheoriespublicationbiasandpsychologicalscience'saversiontothenull.
PerspectPsycholSci.
2012;7:555–61.
96.
ArkesHR,BlumerC.
Thepsychologyofsunkcost.
OrganBehavHumDecisProcess.
1985;35:124–40.
97.
BrumfielG.
Particlesbreaklight-speedlimit.
Nature.
2011.
doi:10.
1038/news.
2011.
554.
98.
MatsonJ.
Faster-than-lightneutrinosPhysicsluminariesvoicedoubts.
SciAm.
2011.
https://www.
scientificamerican.
com/article/ftl-neutri-nos/.
Accessed13Feb2017.
99.
DavidsE,ZhangK,TaraziFI,BaldessariniRJ.
Animalmodelsofattention-deficithyperactivitydisorder.
BrainResRev.
2003;42:1–21.
100.
KlauckSM,PoustkaA.
Animalmodelsofautism.
DrugDiscovTodayDisModels.
2006;3:313–8.
Page13of13SjobergBehavBrainFunct(2017)13:3101.
ArguelloPA,GogosJA.
Schizophrenia:modelingacomplexpsychiatricdisorder.
DrugDiscovTodayDisModels.
2006;3:319–25.
102.
SchmidtMV,MüllerMB.
Animalmodelsofanxiety.
DrugDiscovTodayDisModels.
2006;3:369–74.
103.
DeussingJM.
Animalmodelsofdepression.
DrugDiscovTodayDisModels.
2006;3:375–83.
104.
PautassoM.
Worseningfile-drawerproblemintheabstractsofnatural,medicalandsocialsciencedatabases.
Scientometrics.
2010;85:193–202.
105.
RothsteinHR,SuttonAJ,BorensteinM.
Publicationbiasinmeta-analysis:prevention,assessmentandadjustments.
Chichester:Wiley;2006.
106.
SjobergEA,D'SouzaA,ColeGG.
Anevolutionaryhypothesisconcern-ingfemaleinhibitionabilities:aliteraturereview.
In:Norwegianbehav-ioranalysissocietyconference.
Storefjell,Norway;2016.
107.
SmithGD,EbrahimS.
Datadredging,bias,orconfounding:theycanallgetyouintotheBMJandthefridaypapers.
BrMedJ.
2002;325:1437–8.
108.
SimmonsJP,NelsonLD,SimonsohnU.
False-positivepsychologyundisclosedflexibilityindatacollectionandanalysisallowspresentinganythingassignificant.
PsycholSci2011:0956797611417632.
109.
SullivanGM,FeinnR.
Usingeffectsize—orwhythePvalueisnotenough.
JGradMedEduc.
2012;4:279–82.
110.
BrooksGP,JohansonGA.
Samplesizeconsiderationsformulti-plecomparisonproceduresinAnova.
JModApplStatMethods.
2011;10:97–109.
111.
RoyallRM.
Theeffectofsamplesizeonthemeaningofsignificancetests.
AmStat.
1986;40:313–5.
112.
NakagawaS,CuthillIC.
Effectsize,confidenceintervalandstatisticalsignificance:apracticalguideforbiologists.
BiolRev.
2007;82:591–605.
113.
GarnerJP.
Thesignificanceofmeaning:whydoover90%ofbehavioralneuroscienceresultsfailtotranslatetohumans,andwhatcanwedotofixitILARJ.
2014;55:438–56.
114.
RambozS,OostingR,AmaraDA,KungHF,BlierP,MendelsohnM,etal.
Serotoninreceptor1aknockout:ananimalmodelofanxiety-relateddisorder.
ProcNatlAcadSci.
1998;95:14476–81.
115.
DavidDJ,SamuelsBA,RainerQ,WangJ-W,MarstellerD,MendezI,etal.
Neurogenesis-dependentand-independenteffectsoffluoxetineinananimalmodelofanxiety/depression.
Neuron.
2009;62:479–93.
116.
SiesserW,ZhaoJ,MillerL,ChengSY,McDonaldM.
TransgenicmiceexpressingahumanmutantΒ1thyroidreceptorarehyperactive,impulsive,andinattentive.
GenesBrainBehav.
2006;5:282–97.
117.
GourleySL,TaylorJR.
Recapitulationandreversalofapersistentdepression-likesyndromeinrodents.
CurrProtocNeurosci.
2009;Chap-ter9:Unit-9.
32.
doi:10.
1002/0471142301.
ns0932s49.
118.
WillnerP.
Chronicmildstress(CMS)revisited:consistencyandbehav-ioural-neurobiologicalconcordanceintheeffectsofCMS.
Neuropsy-chobiology.
2005;52:90–110.
119.
SzechtmanH,SulisW,EilamD.
Quinpiroleinducescompulsivecheck-ingbehaviorinrats:apotentialanimalmodelofobsessive-compulsivedisorder(OCD).
BehavNeurosci.
1998;112:1475.
120.
KingJA,AbendS,EdwardsE.
Geneticpredispositionandthedevelop-mentofposttraumaticstressdisorderinananimalmodel.
BiolPsychia-try.
2001;50:231–7.
121.
LipskaBK,JaskiwGE,WeinbergerDR.
Postpubertalemergenceofhyperresponsivenesstostressandtoamphetamineafterneonatalexcitotoxichippocampaldamage:apotentialanimalmodelofschizo-phrenia.
Neuropsychopharmacology.
1993;9:67–75.
122.
LodgeDJ,BehrensMM,GraceAA.
Alossofparvalbumin-containinginterneuronsisassociatedwithdiminishedoscillatoryactivityinananimalmodelofschizophrenia.
JNeurosci.
2009;29:2344–54.
123.
KesbyJP,BurneTH,McGrathJJ,EylesDW.
DevelopmentalvitaminDdeficiencyaltersMk801-inducedhyperlocomotionintheadultrat:ananimalmodelofschizophrenia.
BiolPsychiatry.
2006;60:591–6.