15

Click here to load reader

Exploratory Research is More Reliable Than Confirmatory Research

Embed Size (px)

Citation preview

Page 1: Exploratory Research is More Reliable Than Confirmatory Research

ExploratoryResearchisMoreReliablethanConfirmatoryResearch

ClarkGlymourCarnegieMellonUniversity

1

Page 2: Exploratory Research is More Reliable Than Confirmatory Research

Confirmatory“Logic”

•  HypothesisH:A,Barecausallyconnected•  Nullhypothesis:A,Bareindependent•  HavedataD,chooseteststaIsIcS,chosealpha

•  RejectnullhypothesisifS(D)<alpha•  Ifnullisrejected,Hisconfirmed.

2

Page 3: Exploratory Research is More Reliable Than Confirmatory Research

TheArgumentAgainst“ConfirmatoryResearch”

•  Many,manypossibleposiIvehypotheses(typicallyofcausaleffects)inadomain(psychology,epidemiology,biomedicalscience).

•  Mostarefalse(wayfewerthan10%aretrue).•  SelecIonofwhichhypothesestotestisindependentoftheirtruth.

•  Mosttestsareatalpha=.05andpower<.8•  PosiIvepublishedresultsarefromrejectednullhypotheses.

•  Conclusion:“ifyouusep=0.05asacriterionforclaimingthatyouhavediscoveredaneffectyouwillmakeafoolofyourselfatleast30%oftheIme”,Colquhuon,2012,JRoySoc.Open

3

Page 4: Exploratory Research is More Reliable Than Confirmatory Research

TheBaseRateCalculaIonIllustrated•  SupposeN>>>1hypotheses,10%ofwhicharetrueposiIves(cause-

effect),foreachofwhichthenullhypothesisofindependenceistestedindependentlywithalpha=Pr(rejecIngnull|nullistrue)=0.05;powerw=(probabilityofrejecIngthenullwhenthealternaIveistrue=.8

•  ProbabilityoffindingafalseposiIveassociaIons:Pr(rejectnull|nulltrue)xPr(nulltrue)=.05x.9=.045•  ProbabilityoffindingaposiIveassociaIon:[.045+Pr(rejectnull|alttrue)xPr(alttrue)=.045+(wx.1)•  RaIooftrueposiIvesfoundtoallposiIvesfound:•  .045/(.045+.8x.1)•  E.g.,ifPower=.8,alpha=.05expectedpropor9onoffalseposi9ves

is.36 4

Page 5: Exploratory Research is More Reliable Than Confirmatory Research

WhyDoScienIstsPublish“ConfirmatoryStudies”at.05?

•  BecausetheythinktheyknowmostofthecausalrelaIons?

•  Becauseiftheyusedaloweralphatheirresultswouldnotbe“significant”?

•  Becausearealsearchwouldfindthattheycan’tinfermuchfromthedata?

•  Publishorperish!

5

Page 6: Exploratory Research is More Reliable Than Confirmatory Research

TheArgumentthat“Exploratory,”“HighThroughput”ResearchIstheWorst

•  Becauseittestsmorehypotheses,andsoproducesmorefalseposi:veeffects:

•  ‘”Thegreaterthenumberandthelesser[sic]theselecIonoftestedrelaIonshipsinascienIficfield,thelesslikelytheresearchfindingsaretobetrue.Thusresearchfindingsaremorelikelytrueinconfirmatorydesigns…thaninhypothesisgeneraIngdesigns”[becauseinexploratorystudiesalotoffalsehypothesesaretested,andthemorethataretested,themoreerrorswillbemade.]”FieldsconsideredhighlyinformaIveandcreaIvegiventhewealthoftheassembledandtestedinformaIon,suchasmicroarraysandotherhigh-throughputdiscoveryorientedresearch…shouldhaveextremelylowPPV”[PosiIvePredicIveValue,theprobabilitythatareportedresultistrue].(p.0698).(Ioannidis,2005,PLOSMedicine).

6

Page 7: Exploratory Research is More Reliable Than Confirmatory Research

Balderdash!Ignorance!Dogma!SupersIIon!

•  SearchforcausalrelaIonsisjustparameteres:ma:on.

•  StopthinkingintermsoftesIngandconfirmaIon,thinkaccuraciesofes:mators—hypothesistestsarejustcogsinanesImaIonprocedure.

•  ConsistentesImatorsforrarerelaIonsexist,employingeither(quasi)BayesiancalculaIonsorclassicalhypothesistests,orboth.

•  TheproceduresneverpostulateaconnecIonwithoutmulIpleassessments.

•  Appropriatelyused,theproceduresareamazinglyaccurate.

7

Page 8: Exploratory Research is More Reliable Than Confirmatory Research

Example:SGSAlgorithm

Variables:X1,X2,…,XNXp1–X2isinferredifandonlyifthenullhypothesesX1||X2|ZarerejectedforeachandeverysetZ⊆{X2,…,XN}.InPCsubsetstotestonareselecteddynamically,butthetestsareequivalenttoSGSassumingFaithfulness.

8

Page 9: Exploratory Research is More Reliable Than Confirmatory Research

Example:FGSAlgorithm

•  IteraIvealgorithmstarIngwithtotallydisconnectedgraphofvariables.ConnecIonisaddedonlyifitimprovesthelikelihoodmorethananyotheraddiIonornoneandmorethankln(S)wherekisposiIveandSisthesamplesize.

•  E.g.,inthefirststepasingleconnecIonisaddedonlyifitimprovesthelikelihoodsufficientlyandmorethantheaddiIonofanyotheredge.Foramillionvariablecaseaedgeisaddedonlyifitslikelihoodisbe{erthan~1012alternaIves.

9

Page 10: Exploratory Research is More Reliable Than Confirmatory Research

SimulaIons

AccuraciesforcausalgraphrecoverywithFastGreedySearch(FGS)for,LinearGaussiandata,samplesize1,000:

SimilarresultswithPC-Max.10

Page 11: Exploratory Research is More Reliable Than Confirmatory Research

The“AnI-ExploraIonArgument”HasEverythingBackwards

•  WithverysparsecausalrelaIons,automatedsearchnumberofvariables>>samplesize

hasinthesimulaIons<2%falseposiIvecausalconnecIons<2%falsedirecIonsThesparserthegraph,themoreaccuratetheposiIveresultsoftheprocedure. 11

Page 12: Exploratory Research is More Reliable Than Confirmatory Research

EmpiricalResultsfrom

•  Economics•  Ecology•  Planetaryscience•  Climatescience•  GeneregulaIon•  EducaIonalResearch•  Neuropsychology•  Etc.

12

Page 13: Exploratory Research is More Reliable Than Confirmatory Research

Why?•  TheproceduresareasymptoIcallycorrect.•  TheyusedatainwhichLOTSofvariableshavebeen

measured.•  EachposiIvecausalclaimistestedorassessedmulIple

Imes,againstmulIplecompeInghypothesesinmulIplesubsamplesofthedata.

•  TheproceduresarebiasedagainstposiIveresults.•  Theprocedureshaveanadjustablebiasagainstweak

effectsandinfavorofstrongeffects,andcanbeusedtofindthevariableswiththestrongesttotaleffectsizeforanoutcomeofinterest.

•  ThereverseofIonnaidis’concernaboutrareposiIverelaIonsholds:theproceduresaremostreliablyaccurate,mostinformaIve,andmostfeasiblewhenthetrueposiIvecausalrelaIonsarerare. 13

Page 14: Exploratory Research is More Reliable Than Confirmatory Research

Morals

•  Researchonfast,reliablealgorithmsforcausalesImaIoninavarietyofse}ngsiswheretheacIonisandshouldbe:Latentstructure,feedbackrelaIons,mixedpopulaIons,sampleselecIonbias,Imeseries—allin“highdimensions.”

•  Almosteverythingsaidandwri{eninstaIsIcsaboutthesuperiorityof“confirmatory”researchandtheevilsofdatadrivenhypothesissearchiswrong,verywrong.

•  “Iandmyfriendscan’tthinkofawaytodoX,thereforeXisimpossible”isacrummyinference.

14

Page 15: Exploratory Research is More Reliable Than Confirmatory Research

Read,ThenWork•  P.Spirtes,etal.,Causa:on,Predic:onandSearch•  B.Shipley,Causa:onandCorrela:oninBiology,2ndediIon

•  J.D.Ramsey•  Buhlmann•  Webpagesof:

P.SpirtesT.RichardsonMarloesMaatuisDavidBesslerKevinHoover,forastart

15