The Registered Reports project: A vaccine against research ... · Are the conclusions jus’fied...

38
The Registered Reports project: A vaccine against research bias? Chris Chambers School of Psychology, Cardiff University Email: [email protected] Twitter: @chrisdc77 1

Transcript of The Registered Reports project: A vaccine against research ... · Are the conclusions jus’fied...

  • TheRegisteredReportsproject:Avaccineagainstresearchbias?

    Chris Chambers School of Psychology, Cardiff University

    Email: [email protected] Twitter: @chrisdc77 1

  • What’sbestforscience

    Highqualityresearch,regardlessofoutcome

    What’sbestforscien:sts

    Producingalotofpublishableresults

    Sciencehasanincen:veproblem

    seeNosek,Spies&Motyl(2012).Perspec'vesonPsychologicalScience,7(6):615–631

  • Generate and specify hypotheses!

    Design study!

    Collect data!Analyse data & test hypotheses!

    Interpret data!

    Publish or conduct next experiment!

    Hypothe:co-deduc:vescien:ficmethod

    Publication bias!Lack of data sharing!

    Low statistical power!

    Significance chasing!

    Significance chasing!

    Lack of replication!

    1in1000papersMakeletal(2012)

    ~50%chancetodetectmediumeffectsCohen(1962);SedlmeierandGigerenzer(1989);BezeauandGraves(2001)

    ~50-100%prevalenceJohnetal(2012)

    ~50-90%prevalenceJohnetal(2012)Kerr(1998)

    ~92%posiWveFanelli(2010)

    ~70%failureWichertsetal(2006)

  • Whyisthishappening?

    Becauseweplacetoomuchimportanceontheresultsofexperimentsandnotenoughontheprocessesthatproducethem

    ResultsmakescienceexciWngbutjudgingthequalityofscience(andscienWsts)accordingtotheresultsis“so`”science

  • Canwefixthis?YesPhilosophy:Whatgiveshypothesis-tesWngitsscienWficvalueis•  theQUESTIONitasks•  theQUALITYofthemethodituses•  nevertheRESULTitproduces

    Ifweacceptthisphilosophytheneditorialdecisionsatjournalsshouldbeblindtoresults

  • Thisisnotanewidea

    “Whatweneedisasystemforevalua'ngresearchbasedonlyontheproceduresemployed.Iftheproceduresarejudgedappropriate,sensible,andsufficientlyrigoroustopermitconclusionsfromtheresults,theresearchcannotthenbejudgedinconclusiveonthebasisoftheresultsandrejectedbytherefereesoreditors.Whethertheprocedureswereadequatewouldbejudgedindependentlyoftheoutcome.”

    RobertRosenthal(1966).Experimentereffectsinbehavioralresearch.NewYork.

  • RegisteredReports

    FourcentralaspectsoftheRegisteredReportsmodel:

    •  Partofthepeerreviewprocesstakesplacebeforeexperimentsareconducted

    •  PassingthisstageofreviewvirtuallyguaranteespublicaWon•  Originalstudiesandhigh-valuereplicaWonsarewelcome

    •  Researchersdecidehypotheses,experimentalprocedures,andmainanalysesbeforedatacollecWon

  • AuthorssubmitSTAGE1manuscriptwithIntroducWon,ProposedMethods&

    Analyses,andPilotData(ifapplicable)

    Stage1peerreview

    IfreviewsareposiWvethenjournaloffersin-principleacceptance(IPA),

    regardlessofstudyoutcome(protocolnotpublishedyet)

    Howitworks

    Arethehypotheseswellfounded?Arethemethodsandproposedanalysesfeasibleandsufficientlydetailed?Isthestudywellpowered?(≥90%)Havetheauthorsincludedsufficientposi'vecontrolstoconfirmthatthestudywillprovideafairtest?

  • Howitworks

    Stage2peerreview Didtheauthorsfollowtheapprovedprotocol?Didposi'vecontrolssucceed?Aretheconclusionsjus'fiedbythedata?Manuscriptpublished!

    Authorsdotheresearch

    AuthorsresubmitcompletedSTAGE2manuscript:•  Introduc:onandMethods(virtuallyunchanged)•  Results(new):Registeredconfirmatoryanalyses

    +unregisteredexploratoryanalyses•  Discussion(new)•  Datadepositedinapublicarchive

  • Noneofthesethingsmaper

  • hpp://www.journals.elsevier.com/cortex/virtual-special-issues/virtual-special-issue-registered-reports

    PublishedexamplesatCortex

    Seealso:SocialPsychologyspecialissue:hpp://econtent.hogrefe.com/toc/zsp/45/3PerspecWvesonPsychologicalScience:hpp://www.psychologicalscience.org/index.php/replicaWon/ongoing-projects

    –Reproducible–•  detailed,repeatablemethods•  highstaWsWcalpower(2-3xabovenormal)

    –Transparent–•  accompaniedbyopendata&materials•  outcomesofconfirmatoryandexploratory

    analysesdisWnguished

    –Credible–•  nopublicaWonbias•  nohindsightbias•  noselecWvereporWng

  • Permanentadopters

    Specialissues

    Forfulllistseehpps://cos.io/rr/

  • RegisteredReportsatRoyalSocietyOpenScience

    NowavailableinallSTEMareas,fromphysicstopsychology

    hpp://rsos.royalsocietypublishing.org/content/registered-reports

  • 14

    RegisteredReportsatNatureHumanBehaviour

  • Whatarethebenefitsforjournals,editors,authorsandthescienWficcommunity?

  • Benefits

  • Benefits

    1.  NopublicaWonbias

  • Benefits

    1.  NopublicaWonbias

  • Benefits

    1.  NopublicaWonbias2.  Logicallyeliminatesvariousforms

    researcherbias(p-hacking,posthochypothesising)

  • Benefits

    1.  NopublicaWonbias2.  Logicallyeliminatesvariousforms

    researcherbias(p-hacking,posthochypothesising)

  • Benefits

    1.  NopublicaWonbias2.  Logicallyeliminatesvariousforms

    researcherbias(p-hacking,posthochypothesising)

    3.  HighstaWsWcalpowerrequirements

    increasereproducibility

  • Benefits

    1.  NopublicaWonbias2.  Logicallyeliminatesvariousforms

    researcherbias(p-hacking,posthochypothesising)

    3.  HighstaWsWcalpowerrequirements

    increasereproducibility

  • Benefits

    1.  NopublicaWonbias2.  Logicallyeliminatesvariousforms

    researcherbias(p-hacking,posthochypothesising)

    3.  HighstaWsWcalpowerrequirements

    increasereproducibility4.  IncenWvizesimportantreplicaWon

    studiesandothernovel,resource-intensiveprojects(wherepublicaWonwouldnormallybeconWngentonresults)

  • Benefits

    1.  NopublicaWonbias2.  Logicallyeliminatesvariousforms

    researcherbias(p-hacking,posthochypothesising)

    3.  HighstaWsWcalpowerrequirements

    increasereproducibility4.  IncenWvizesimportantreplicaWon

    studiesandothernovel,resource-intensiveprojects(wherepublicaWonwouldnormallybeconWngentonresults)

  • Benefits

    1.  NopublicaWonbias2.  Logicallyeliminatesvariousforms

    researcherbias(p-hacking,posthochypothesising)

    3.  HighstaWsWcalpowerrequirements

    increasereproducibility4.  IncenWvizesimportantreplicaWon

    studiesandothernovel,resource-intensiveprojects(wherepublicaWonwouldnormallybeconWngentonresults)

    5.  Incorporatespublicarchivingofdata

    andmaterials

  • Benefits

    1.  NopublicaWonbias2.  Logicallyeliminatesvariousforms

    researcherbias(p-hacking,posthochypothesising)

    3.  HighstaWsWcalpowerrequirements

    increasereproducibility4.  IncenWvizesimportantreplicaWon

    studiesandothernovel,resource-intensiveprojects(wherepublicaWonwouldnormallybeconWngentonresults)

    5.  Incorporatespublicarchivingofdata

    andmaterials

  • FrequentlyaskedquesWons

  • 28

    1.“AreRegisteredReportssuitableformyfield?”

    •  Applicabletoanyfieldengagedinhypothesis-drivenresearchwhereoneormoreofthefollowingproblemsapply:

    •  Publica:onbias•  Significancechasing(e.g.p-hacking)•  Posthochypothesizing(hindsightbias)•  Lowsta:s:calpower•  Lackofdirectreplica:on

    •  Notapplicablefor•  Purelyexploratoryscience•  Methodsdevelopment } NohypothesistesWng

  • 3.“What’stostopRegisteredReportsfrombecomingadumpinggroundforinconclusivenullresults?”

    •  aprioripowerrequirements(≥90%)increasereproducibilityofallfindings•  BayesianinferenWalmethodswelcomedforprovidingevidenceinfavourofH0orH1.

    2.“Couldresearcherscheatby‘pre-registering’astudythattheyhavealreadyconducted?”•  Time-stampedrawdatafilesmustbesubmipedatStage2withbasiclablogand

    cerWficaWonfromallauthorsthatdatawascollecteda`erprovisionalacceptance•  SubmiwngacompletedstudyatStage1wouldthereforebefraud•  StrategywouldbackfireanywaywhenreviewersaskforamendmentsatStage1

    4.“Pre-registraWonisfineforseniorresearcherswhohave‘madeit’butI’majuniorscienWstandneedtoplaythegame”

    •  Goingforpostdocjobs,whatyoudothinkwilllookbeperonyourCV?A)Bunchofpaperslistedas“inpreparaWon”,“submiped”B)Bunchofpaperslistedas“provisionallyacceptedat[respectedjournal]”

    •  Thegameischanging:journalpoliciesarechangingtovaluetransparencyandreproducibility,e.g.NatureHumanBehaviourhaslaunchedRRs

    RegisteredReportsaren’tdesignedtopreventfraudbuttoincen'vizegoodprac'ce

  • 5.“WillthislimitexploraWonorsWgmaWzeexploratoryresearch?”•  No.Thearenorestric:onsontherepor:ngofunregisteredexploratoryanalyses.•  Confirmatoryandexploratoryanalysesaresimplyreportedseparatelyinthefinalpaper

    •  ExploratoryReportsatCortex(indevelopment)•  nohypothesistesWng•  nopvalues•  Data-led;lightonintroducWonandtheory•  Purposeistogeneratehypothesesratherthantestthem

    WhatsWgmaWzesexploratoryresearchisposthochypothesizingtofitadeducWveframeworkExploratoryresearchissimplynotvaluedinitsnaWveform

  • 31

    6.“Whathappensifweneedtochangesomethingaboutourexperimentalproceduresa`ertheyareprovisionallyaccepted?”

    •  Minorchanges(e.g.replacingequipment)canbefootnotedinStage2manuscriptasprotocoldeviaWons

    •  Majorchanges(e.g.changingdataexclusioncriteria)arelikelytorequirewithdrawal•  EditorialteamdecideswhetherdeviaWonissufficientlyminortoconWnue

    8.“IhaveaccesstoanexisWngdatasetthatIhaven’tyetanalysed.CanIsubmitthisproposedanalysisasaRegisteredReport?”

    •  NotatCortex,butotherjournalsofferthis,suchasEuropeanJournalofNeuroscience…

    7.“Someofmyanalyseswilldependontheresults,sohowcanIpre-registereachstepindetail?”(e.g.typeofstaWsWcalmodel)

    •  Pre-registraWondoesn’trequireeachdecisiontobespecified,onlythedecisiontree•  Authorscanpre-registertheconWngencies/rulesforfuturedecisions

  • hpps://docs.google.com/spreadsheets/d/1D4_k-8C_UENTRtbPzX}jEyu3BfLxdOsn9j-otrO870/edit#gid=0

  • 9.“HowwillRegisteredReportsincenWvizereplicaWonstudies?”

    •  Conspiracyofcircumstancestellsusnottobotherdoingdirect(exact)replicaWons

    •  MethodsecWonsareo`entoovaguetoallowprecisereplicaWon•  ChroniclackofpowerinnovelresearchmeansthatreplicaWonso`en

    requireverylargesamplessizes•  ApempWngtoexactlyrepeatapreviousexperimentcanbeseen(in

    psychology)asanactofaggression(cf.physics)•  MoWvatedreasoningbyreviewerscanimpedepublicaWon•  Mostpsych/neurojournalswantnoveltyandseereplicaWonsas(usually)

    unpublishable

    •  RRs:haveproposedreplicaWonexperimentreviewedandprovisionallyacceptedbeforeyouinvestsubstanWalresourcesintodoingit;potenWallyinvolveoriginalauthorsinpeerreviewoftheprotocol;mo:vatedreasoningisprevented

  • 11.“RegisteredReportsseemslimitedtosinglestudies.Butourpapersusuallyincludesequencesofexperiments”

    •  WewelcomesequenWalregistraWonsinwhichauthorsaddstudiesiteraWvelyatStage1viaafast-trackmechanismandcompletethematStage2

    •  Witheachcompletedcycle,thepreviousacceptedversionofthepaperisguaranteedtobepublished

    •  AuthorscanalsoincludeasequenceofunregisteredexperimentsaspreliminarystudiesinaStage1RR(e.g.E1,E2,E3preliminary;manuscriptproposesE4aspre-registeredtest)

    10.“Reviewerscouldstealmyideasatthepre-registraWonstageandscoopme”

    •  OnlyahandfulofpeopleknowabouteachStage1submission•  OnceStage1protocolisaccepted,thejournalcan’trejectyourpaperbecause

    somethingsimilarwaspublished(noveltyisirrelevant)•  ManuscriptreceiveddateonpublishedRRwillbethedateofStage1submission•  HowdifferentfromgrantapplicaWons,conferencepresentaWons,seminars?

  • 35

    Twothingswe’velearnedaseditors

    1.ReviewerssomeWmesshi`thegoalpostsoncedataareinCasestudy:•  A`ermulWpleroundsofreview,areviewerapprovedprotocolatStage1•  Whenresultsfailedtoconfirmreviewer’sexpectaWonsatStage2,

    reviewerraisednewmethodologicalobjecWons&apemptedtoreject

    Casestudy:•  A`ermulWpleroundsofreview,areviewerapprovedprotocolatStage1•  WhenresultswerestaWsWcallynon-significant,reviewerdemandedthat

    authorsconductalonglistofposthocanalysesto“findsomething”

    EDITORALDECISION:Reviewerwasoverruled.BarringextremecaseswhereallparWes(authors,reviewers,editors)agreethatacriWcalflawwasoverlooked,objecWonstoStage1methodsareineligibleatStage2.LimitaWonsinsteadcoveredinDiscussion.

    EDITORALDECISION:Posthocanalysescanonlyberequiredifdeemednecessarytosupportauthor’sconclusions.Authorinvitedtoconsiderextraanalysesbutnotrequiredtodothem.Reviewerinvitedtoconductanalysesusingopendataandpublishaseparatecommentpiece.

    Upshot:RRsarerevealingreviewerbiasinwaythatisinvisibleinconvenWonalreview

  • 36

    Twothingswe’velearnedaseditors

    2.LackofposiWvecontrolsinpsychologyandcogniWveneuroscience

    •  ButfewiniWalsubmissionsproposesuchtests•  Manyfieldshavenosuchtests

    STAGE1CRITERION6Whethertheauthorshaveconsideredsufficientoutcome-neutralcondiWonsforensuringthattheresultsobtainedareabletotestthestatedhypotheses

    STAGE2CRITERION1Whetherthedataareabletotesttheauthors’proposedhypothesesbypassingtheapprovedoutcome-neutralcriteria

    PrevailingassumpWon:astudyissaidtohave“worked”ifthemainhypothesiswassupported(p

  • 37

    Goingevenfurther…

    CanweintegrateclinicaltrialregistraWon(whereitapplies),ethicalreview,grantfundingandRegisteredReports?

    •  PossiblesoluWon:RegisteredReportsfundingmodel•  Authorssubmittheirresearchproposalbeforetheyhavefunding.

    •  Followingsimultaneousreviewbytheboththefunderandthejournal,thestrongestproposalswouldbeofferedfinancialsupportbythefunderANDin-principleacceptanceforpublicaWonbythejournal.

  • Informa:onHubattheCenterforOpenScience

    hpps://cos.io/rr/

    Formoreinfo,emailme([email protected])orDavidMellorattheCOS([email protected])

    •  DetailedFAQs•  Tablecomparingjournalfeatures