资源描述:
《[20] An Interpretation of Partial Least Squares.pdf》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、AnInterpretationofPartialLeastSquaresAuthor(s):PaulH.GarthwaiteReviewedwork(s):Source:JournaloftheAmericanStatisticalAssociation,Vol.89,No.425(Mar.,1994),pp.122-127Publishedby:AmericanStatisticalAssociationStableURL:http://www.jstor.org/stable/2291207.Accessed:31/12/201203:02YouruseoftheJ
2、STORarchiveindicatesyouracceptanceoftheTerms&ConditionsofUse,availableat.http://www.jstor.org/page/info/about/policies/terms.jsp.JSTORisanot-for-profitservicethathelpsscholars,researchers,andstudentsdiscover,use,andbuilduponawiderangeofcontentinatrusteddigitalarchive.Weuseinformationtechn
3、ologyandtoolstoincreaseproductivityandfacilitatenewformsofscholarship.FormoreinformationaboutJSTOR,pleasecontactsupport@jstor.org..AmericanStatisticalAssociationiscollaboratingwithJSTORtodigitize,preserveandextendaccesstoJournaloftheAmericanStatisticalAssociation.http://www.jstor.orgThisc
4、ontentdownloadedonMon,31Dec201203:02:38AMAllusesubjecttoJSTORTermsandConditionsAnInterpretationofPartialLeastSquaresPaulH.GARTHWAITE*Univariatepartialleastsquares(PLS)isamethodofmodelingrelationshipsbetweenaYvariableandotherexplanatoryvariables.Itmaybeusedwithanynumberofexplanatoryvariabl
5、es,evenfarmorethanthenumberofobservations.Asimpleinterpretationisgiventhatshowsthemethodtobeastraightforwardandreasonablewayofformingpredictionequations.ItsrelationshiptomultivariatePLS,inwhichtherearetwoormoreYvariables,isexamined,andanexampleisgiveninwhichitiscomparedbysimulationwithoth
6、ermethodsofformingpredictionequations.WithunivariatePLS,linearcombinationsoftheexplanatoryvariablesareformedsequentiallyandrelatedtoYbyordinaryleastsquaresregression.Itisshownthattheselinearcombinations,herecalledcomponents,maybeviewedasweightedaveragesofpredictors,whereeachpredictorholds
7、theresidualinformationinanexplanatoryvariablethatisnotcontainedinearliercomponents,andthequantitytobepredictedisthevectorofresidualsfromregressingYagainstearliercomponents.AsimilarstrategyisshowntounderliemultivariatePLS,exceptthatthequantitytobepredictedisaweighted