资源描述:
《Power failure - why small sample size undermines the reliability of neuroscience》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、ANALYSISPowerfailure:whysmallsamplesizeunderminesthereliabilityofneuroscienceKatherineS. Button1,2,JohnP. A. Ioannidis3,ClaireMokrysz1,BrianA. Nosek4,JonathanFlint5,EmmaS. J. Robinson6andMarcusR. Munafò1Abstract
2、Astudywithlowstatisticalpowerhasareducedchanc
3、eofdetectingatrueeffect,butitislesswellappreciatedthatlowpoweralsoreducesthelikelihoodthatastatisticallysignificantresultreflectsatrueeffect.Here,weshowthattheaveragestatisticalpowerofstudiesintheneurosciencesisverylow.Theconsequencesofthisincludeoverestima
4、tesofeffectsizeandlowreproducibilityofresults.Therearealsoethicaldimensionstothisproblem,asunreliableresearchisinefficientandwasteful.Improvingreproducibilityinneuroscienceisakeypriorityandrequiresattentiontowell-establishedbutoftenignoredmethodologicalprin
5、ciples.Ithasbeenclaimedanddemonstratedthatmany(andlowsamplesizeofstudies,smalleffectsorboth)nega-possiblymost)oftheconclusionsdrawnfrombiomedi-tivelyaffectsthelikelihoodthatanominallystatisticallycalresearchareprobablyfalse1.Acentralcauseforthissignificantf
6、indingactuallyreflectsatrueeffect.Wedis-importantproblemisthatresearchersmustpublishincusstheproblemsthatarisewhenlow-poweredresearchordertosucceed,andpublishingisahighlycompetitivedesignsarepervasive.Ingeneral,theseproblemscanbeenterprise,withcertainkindso
7、ffindingsmorelikelytodividedintotwocategories.Thefirstconcernsprob-bepublishedthanothers.Researchthatproducesnovellemsthataremathematicallyexpectedtoariseevenifresults,statisticallysignificantresults(thatis,typicallytheresearchconductedisotherwiseperfect:in
8、otherp<0.05)andseemingly‘clean’resultsismorelikelytobewords,whentherearenobiasesthattendtocreatesta-1SchoolofExperimentalpublished2,3.Asaconsequence,researchershavestrongtisticallysignificant(thatis,‘positive’)resultsthatarePsychology,UniversityofBristol,Br
9、istol,BS81TU,UK.incentivestoengageinresearchpracticesthatmakespurious.Thesecondcategoryconcernsproblemsthat2SchoolofSocialandtheirfindingspublishablequickly,evenifthoseprac-reflectbiasesthattendtoco-oc