2mifaIndex11 mifaGetcovariancematrixofincompletedatausingmultipleimputation DescriptionComputecovariancematrixofincompletedatausingmultipleimputationFormultipleimputationMultivariateImputationbyCha ID: 847331
Download Pdf The PPT/PDF document "PackagemifaJanuary222021TitleMultipleImp..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1 Package`mifa'January22,2021TitleMultiple
Package`mifa'January22,2021TitleMultipleImputationforExploratoryFactorAnalysisVersion0.2.0URLhttps://github.com/teebusch/mifaBugReportshttps://github.com/teebusch/mifa/issuesImportsstats,mice,dplyr,checkmateSuggestspsych,testthat,knitr,rmarkdown,ggplot2,tidyr,covrDescriptionImputethecovariancematrixofincompletedatasothatfactoranalysiscanbeperformed.ImputationsaremadeusingmultipleimputationbyMultivariateImputationwithChainedEquations(MICE)andcombinedwithRubinsrules.ParametricFiellercondenceintervalsandnonparametricbootstrapcondenceintervalscanbeobtainedforthevarianceexplainedbydifferentnumbersofprincipalcomponents.ThemethodisdescribedinNassirietal.(2018) oi:;.3;ݘ/;s134;(-0;-1;-;䀀LicenseMIT+leLICENSEEncodingUTF-8LazyDatatrueRoxygenNote7.1.1NeedsCompilationnoAuthorVahidNassiri[aut],AnikóLovik[aut],GeertMolenberghs[aut],GeertVerbeke[aut],TobiasBusch[aut,cre](http;s://;orci; .or;᠀g/0000-0002-8390-7892)MaintainerTobiasBuschteeb;usch;@gma;il.c;om00;RepositoryCRANDate/Publication2021-01-2208:40:08UTCRtopicsdocumented:mifa.............................................2mifa_ci_boot........................................5mifa_ci_eller........................................91 2mifaIndex11 mifaGetcovariancematrixofincompletedatausingmultipleimputation DescriptionComputecovariancematrixofincompletedatausingmultipleimputation.Formultipleimputa-tion,MultivariateImputationbyChainedEquations(MICE)fromthemicepackageisused.ThecovariancematricesoftheimputeddatasetsarecombinedusingRubin'srules.Usagemifa(data,co
2 v_vars=dplyr::everything(),n_pc,ci=FALSE
v_vars=dplyr::everything(),n_pc,ci=FALSE,conf=0.95,n_boot=1000,...)ArgumentsdataAdataframewithmissingvaluescodedasNA.cov_varsVariablesindataforwhichtocalculatethecovariancematrix.Supports(tidyselection)dplyr::select().Thisallowstoselectvariablesthatareusedfortheimputationsofmissingvalues,butnotthecalculationsofthecovariancematrix.Thisisespeciallyusefulwhentherearecategoricalpredictorsthatcanimprovetheimputationoftheresponsevariables,butforwhichcovariancecannotbecalculated.Bydefault,allvariablesindataareusedforboth,theimputationandthecovariancematrix.Note:Variablesandrowsusedfortheimputation,aswellasthemethodforimputationcanbeconguredusingthe....Seealsomice::mice().n_pcIntegerorintegervectorindicatingnumberofprincipalcomponents(eigenvec-tors)forwhichexplainedvariance(eigenvalues)shouldbeobtainedandforwhichcondenceintervalsshouldbecomputed.Defaultstoallprincipalcom-ponents,i.e.,thenumberofvariablesinthedata.ciAcharacterstringindicatingwhichtypesofcondenceintervalsshouldbecon-structedforthevarianceexplainedbytheprincipalcomponents.If"boot","fieller",or"both",thecorrespondingintervalsarecomputed.IfFALSE(thedefault)nocondenceintervalswillbecomputed.Thecomponentsforwhichcondenceintervalsshouldbecomputedcanbesetwithn_pc.Seemifa_ci_boot()andmifa_ci_fieller()fordetailsaboutthetwomethods.confCondencelevelforconstructingcondenceintervals.Thedefaultis.95thatis,95%condenceintervals. mifa3n_bootNumberofbootstrapsamplestouseforbootstrappedcondenceintervals.Thedefaultis1000....Argumentspassedontomice::micemNumberofmultipleimputations.Thedefaultism=5.met
3 hodCanbeeitherasinglestring,oravectorofs
hodCanbeeitherasinglestring,oravectorofstringswithlengthlength(blocks),specifyingtheimputationmethodtobeusedforeachcolumnindata.Ifspeciedasasinglestring,thesamemethodwillbeusedforallblocks.Thedefaultimputationmethod(whennoargumentisspecied)dependsonthemeasurementlevelofthetargetcolumn,asregulatedbythedefaultMethodargument.Columnsthatneednotbeimputedhavetheemptymethod"".Seedetails.predictorMatrixAnumericmatrixoflength(blocks)rowsandncol(data)columns,containing0/1dataspecifyingthesetofpredictorstobeusedforeachtargetcolumn.Eachrowcorrespondstoavariableblock,i.e.,asetofvariablestobeimputed.Avalueof1meansthatthecolumnvariableisusedasapredictorforthetargetblock(intherows).Bydefault,thepredictorMatrixisasquarematrixofncol(data)rowsandcolumnswithall1's,exceptforthediagonal.Note:Fortwo-levelimputationmod-els(whichhave"2l"intheirnames)othercodes(e.g,2or-2)arealsoallowed.ignoreAlogicalvectorofnrow(data)elementsindicatingwhichrowsareignoredwhencreatingtheimputationmodel.ThedefaultNULLincludesallrowsthathaveanobservedvalueofthevariabletoimputed.RowswithignoresettoTRUEdonotinuencetheparametersoftheimpu-tationmodel,butarestillimputed.Wemayusetheignoreargumenttosplitdataintoatrainingset(onwhichtheimputationmodelisbuilt)andatestset(thatdoesnotinuencetheimputationmodelestimates).Note:Multivariateimputationmethods,likemice.impute.jomoImpute()ormice.impute.panImpute(),donothonourtheignoreargument.whereAdataframeormatrixwithlogicalsofthesamedimensionsasdataindicatingwhereinthedatatheimputationsshouldbecreated.Thedefault,where=is.na(data),speciesthatthemissingdatashouldbeimp
4 uted.Thewhereargumentmaybeusedtooverimpu
uted.Thewhereargumentmaybeusedtooverimputeobserveddata,ortoskipimputationsforselectedmissingvalues.blocksListofvectorswithvariablenamesperblock.Listelementsmaybenamedtoidentifyblocks.Variableswithinablockareimputedbyamulti-variateimputationmethod(seemethodargument).Bydefaulteachvari-ableisplacedintoitsownblock,whichiseffectivelyfullyconditionalspecication(FCS)byunivariatemodels(variable-by-variableimputation).Onlyvariableswhosenamesappearinblocksareimputed.TherelevantcolumnsinthewherematrixaresettoFALSEofvariablesthatarenotblockmembers.Avariablemayappearinmultipleblocks.Inthatcase,itiseffectivelyre-imputedeachtimethatitisvisited.visitSequenceAvectorofblocknamesofarbitrarylength,specifyingthesequenceofblocksthatareimputedduringoneiterationoftheGibbssam-pler.Ablockisacollectionofvariables.Allvariablesthataremembersofthesameblockareimputedwhentheblockisvisited.Avariablethatis 4mifaamemberofmultipleblocksisre-imputedwithinthesameiteration.ThedefaultvisitSequence="roman"visitstheblocks(lefttoright)intheor-derinwhichtheyappearinblocks.Onemayalsouseoneofthefollowingkeywords:"arabic"(righttoleft),"monotone"(orderedlowtohighpro-portionofmissingdata)and"revmonotone"(reverseofmonotone).formulasAnamedlistofformula's,orexpressionsthatcanbeconvertedintoformula'sbyas.formula.Listelementscorrespondtoblocks.Theblocktowhichthelistelementappliesisidentiedbyitsname,solistnamesmustcorrespondtoblocknames.TheformulasargumentisanalternativetothepredictorMatrixargumentthatallowsformoreexibilityinspecifyingimputationmodels,e.g.,forspecifyinginteractionterms.blotsAnamedlistofalist'sthatc
5 anbeusedtopassdownargumentstolowerleveli
anbeusedtopassdownargumentstolowerlevelimputationfunction.Theentriesofelementblots[[blockname]]arepasseddowntothefunctioncalledforblockblockname.postAvectorofstringswithlengthncol(data)specifyingexpressionsasstrings.Eachstringisparsedandexecutedwithinthesampler()func-tiontopost-processimputedvaluesduringtheiterations.Thedefaultisavectorofemptystrings,indicatingnopost-processing.defaultMethodAvectoroflength4containingthedefaultimputationmethodsfor1)numericdata,2)factordatawith2levels,3)factordatawith2unorderedlevels,and4)factordatawith2orderedlevels.Bydefault,themethodusespmm,predictivemeanmatching(numericdata)logreg,logisticregressionimputation(binarydata,factorwith2levels)polyreg,polytomousregressionimputationforunorderedcategoricaldata(factor2levels)polr,proportionaloddsmodelfor(ordered,2levels).maxitAscalargivingthenumberofiterations.Thedefaultis5.printFlagIfTRUE,micewillprinthistoryonconsole.Useprint=FALSEforsilentcomputation.seedAnintegerthatisusedasargumentbytheset.seed()foroffsettingtherandomnumbergenerator.Defaultistoleavetherandomnumbergeneratoralone.data.initAdataframeofthesamesizeandtypeasdata,withoutmissingdata,usedtoinitializeimputationsbeforethestartoftheiterativeprocess.ThedefaultNULLimpliesthatstartingimputationarecreatedbyasimplerandomdrawfromthedata.Notethatspecicationofdata.initwillstartallmGibbssamplingstreamsfromthesameimputation.DetailsThefunctionalsocomputesthevarianceexplainedbydifferentnumbersofprincipalcomponentsandthecorrespondingFieller(parametric)orbootstrap(nonparametric)condenceintervals.ValueAlist:cov_
6 combinedTheestimatedcovariancematrixofth
combinedTheestimatedcovariancematrixoftheincompletedata,basedonthecombinedcovariancematricesofimputeddatasets. mifa_ci_boot5cov_imputationsAlistcontainingtheestimatedcovariancematrixesforallimputeddatasets.var_explainedAdataframecontainingtheestimatedproportionsofexplainedvarianceforeachofspeciedn_pccomponents.Dependingonci,itwillalsocontaintheestimatedFieller's(parametric)and/orbootstrap(nonparametric)condenceintervalfortheproportionofvari-anceexplainedbythedifferentnumbersofprincipalcomponentsdenedbyn_pc.midsObjectoftypemice::mids.Thisistheresultsofthemultipleimputationstepforthecovari-ancematrix.Canbeusefulfordiagnosingthemultipleimputations.ReferencesNassiri,V.,Lovik,A.,Molenberghs,G.,&Verbeke,G.(2018).Onusingmultipleimputationforexploratoryfactoranalysisofincompletedata.BehavioralResearchMethods50,501517.doi:10.3758/s1342801710134SeeAlsomifa_ci_boot(),mifa_ci_fieller(),mice::mice()Examplesif(requireNamespace("psych")){datapsych::bfimifa(data,cov_vars=-c(age,education,gender),ci="fieller",print=FALSE)} mifa_ci_bootBootstrapcondenceintervalsforexplainedvariance DescriptionComputebootstrapcondenceintervalsfortheproportionofexplainedvarianceforthecovarianceofanincompletedataimputedusingmultipleimputation.Formultipleimputation,MultivariateImputationbyChainedEquations(MICE)fromthemicepackageisused.Usagemifa_ci_boot(data,cov_vars=dplyr::everything(),n_pc,conf=0.95,n_boot=1000,progress=FALSE,...) 6mifa_ci_bootArgumentsdataAdataframewithmissingvaluescodedasNA.cov_varsVariablesindataforwhichtocalculatethecovariancematrix.Supports(tidyselection)dplyr::s
7 elect().Thisallowstoselectvariablesthata
elect().Thisallowstoselectvariablesthatareusedfortheimputationsofmissingvalues,butnotthecalculationsofthecovariancematrix.Thisisespeciallyusefulwhentherearecategoricalpredictorsthatcanimprovetheimputationoftheresponsevariables,butforwhichcovariancecannotbecalculated.Bydefault,allvariablesindataareusedforboth,theimputationandthecovariancematrix.Note:Variablesandrowsusedfortheimputation,aswellasthemethodforimputationcanbeconguredusingthe....Seealsomice::mice().n_pcIntegerorintegervectorindicatingnumberofprincipalcomponents(eigenvec-tors)forwhichexplainedvariance(eigenvalues)shouldbeobtainedandforwhichcondenceintervalsshouldbecomputed.Defaultstoallprincipalcom-ponents,i.e.,thenumberofvariablesinthedata.confCondencelevelforconstructingcondenceintervals.Thedefaultis.95thatis,95%condenceintervals.n_bootNumberofbootstrapsamplestouseforbootstrappedcondenceintervals.Thedefaultis1000.progressLogical.Whethertoshowprogressbarsforcomputationofbootstrapcondenceintervals.DefaultisFALSE....Argumentspassedontomice::micemNumberofmultipleimputations.Thedefaultism=5.methodCanbeeitherasinglestring,oravectorofstringswithlengthlength(blocks),specifyingtheimputationmethodtobeusedforeachcolumnindata.Ifspeciedasasinglestring,thesamemethodwillbeusedforallblocks.Thedefaultimputationmethod(whennoargumentisspecied)dependsonthemeasurementlevelofthetargetcolumn,asregulatedbythedefaultMethodargument.Columnsthatneednotbeimputedhavetheemptymethod"".Seedetails.predictorMatrixAnumericmatrixoflength(blocks)rowsandncol(data)columns,containing0/1dataspecifyingthesetofpredictor
8 stobeusedforeachtargetcolumn.Eachrowcorr
stobeusedforeachtargetcolumn.Eachrowcorrespondstoavariableblock,i.e.,asetofvariablestobeimputed.Avalueof1meansthatthecolumnvariableisusedasapredictorforthetargetblock(intherows).Bydefault,thepredictorMatrixisasquarematrixofncol(data)rowsandcolumnswithall1's,exceptforthediagonal.Note:Fortwo-levelimputationmod-els(whichhave"2l"intheirnames)othercodes(e.g,2or-2)arealsoallowed.ignoreAlogicalvectorofnrow(data)elementsindicatingwhichrowsareignoredwhencreatingtheimputationmodel.ThedefaultNULLincludesallrowsthathaveanobservedvalueofthevariabletoimputed.RowswithignoresettoTRUEdonotinuencetheparametersoftheimpu-tationmodel,butarestillimputed.Wemayusetheignoreargumenttosplitdataintoatrainingset(onwhichtheimputationmodelisbuilt) mifa_ci_boot7andatestset(thatdoesnotinuencetheimputationmodelestimates).Note:Multivariateimputationmethods,likemice.impute.jomoImpute()ormice.impute.panImpute(),donothonourtheignoreargument.whereAdataframeormatrixwithlogicalsofthesamedimensionsasdataindicatingwhereinthedatatheimputationsshouldbecreated.Thedefault,where=is.na(data),speciesthatthemissingdatashouldbeimputed.Thewhereargumentmaybeusedtooverimputeobserveddata,ortoskipimputationsforselectedmissingvalues.blocksListofvectorswithvariablenamesperblock.Listelementsmaybenamedtoidentifyblocks.Variableswithinablockareimputedbyamulti-variateimputationmethod(seemethodargument).Bydefaulteachvari-ableisplacedintoitsownblock,whichiseffectivelyfullyconditionalspecication(FCS)byunivariatemodels(variable-by-variableimputation).Onlyvariableswhosenamesappearinblocksareimputed.Therelevantcolumnsinthewhe
9 rematrixaresettoFALSEofvariablesthataren
rematrixaresettoFALSEofvariablesthatarenotblockmembers.Avariablemayappearinmultipleblocks.Inthatcase,itiseffectivelyre-imputedeachtimethatitisvisited.visitSequenceAvectorofblocknamesofarbitrarylength,specifyingthesequenceofblocksthatareimputedduringoneiterationoftheGibbssam-pler.Ablockisacollectionofvariables.Allvariablesthataremembersofthesameblockareimputedwhentheblockisvisited.Avariablethatisamemberofmultipleblocksisre-imputedwithinthesameiteration.ThedefaultvisitSequence="roman"visitstheblocks(lefttoright)intheor-derinwhichtheyappearinblocks.Onemayalsouseoneofthefollowingkeywords:"arabic"(righttoleft),"monotone"(orderedlowtohighpro-portionofmissingdata)and"revmonotone"(reverseofmonotone).formulasAnamedlistofformula's,orexpressionsthatcanbeconvertedintoformula'sbyas.formula.Listelementscorrespondtoblocks.Theblocktowhichthelistelementappliesisidentiedbyitsname,solistnamesmustcorrespondtoblocknames.TheformulasargumentisanalternativetothepredictorMatrixargumentthatallowsformoreexibilityinspecifyingimputationmodels,e.g.,forspecifyinginteractionterms.blotsAnamedlistofalist'sthatcanbeusedtopassdownargumentstolowerlevelimputationfunction.Theentriesofelementblots[[blockname]]arepasseddowntothefunctioncalledforblockblockname.postAvectorofstringswithlengthncol(data)specifyingexpressionsasstrings.Eachstringisparsedandexecutedwithinthesampler()func-tiontopost-processimputedvaluesduringtheiterations.Thedefaultisavectorofemptystrings,indicatingnopost-processing.defaultMethodAvectoroflength4containingthedefaultimputationmethodsfor1)numericdata,2)factordatawith2levels,3)factord
10 atawith2unorderedlevels,and4)fac
atawith2unorderedlevels,and4)factordatawith2orderedlevels.Bydefault,themethodusespmm,predictivemeanmatching(numericdata)logreg,logisticregressionimputation(binarydata,factorwith2levels)polyreg,polytomousregressionimputationforunorderedcategoricaldata(factor2levels)polr,proportionaloddsmodelfor(ordered,2levels).maxitAscalargivingthenumberofiterations.Thedefaultis5. 8mifa_ci_bootprintFlagIfTRUE,micewillprinthistoryonconsole.Useprint=FALSEforsilentcomputation.seedAnintegerthatisusedasargumentbytheset.seed()foroffsettingtherandomnumbergenerator.Defaultistoleavetherandomnumbergeneratoralone.data.initAdataframeofthesamesizeandtypeasdata,withoutmissingdata,usedtoinitializeimputationsbeforethestartoftheiterativeprocess.ThedefaultNULLimpliesthatstartingimputationarecreatedbyasimplerandomdrawfromthedata.Notethatspecicationofdata.initwillstartallmGibbssamplingstreamsfromthesameimputation.DetailsThisfunctionusestheShaoandSitter(1996)methodtocombinemultipleimputationandboot-strapping.Theimputationsaredoneusingmice::mice().Normally,thisfunctiondoesnotneedtobecalleddirectly.Instead,usemifa(...,ci="boot").ValueAdataframecontainingbootstrappedcondenceintervalsforvarianceexplainedbydifferentnum-berofprincipalcomponents.ReferencesShao,J.&Sitter,R.R.(1996).Bootstrapforimputedsurveydata.JournaloftheAmericanStatis-ticalAssociation91.435(1996):1278-1288.doi:10.1080/01621459.1996.10476997SeeAlsomifa(),mice::mice()Othermifacondenceintervals:mifa_ci_fieller()Examplesif(requireNamespace("psych")){datapsych::bfi[,1:25]mifa_ci_boot(data,n_pc=3:8,n_boot=10,pr
11 int=FALSE)} mifa_ci_eller9 mifa_ci_f
int=FALSE)} mifa_ci_eller9 mifa_ci_fiellerFiellercondenceintervalsforexplainedvariance DescriptionComputesparametriccondenceintervalsforproportionofexplainedvarianceforgivennumbersofprincipalcomponentsusingFieller'smethod.Notethatbysettingci=TRUEinmifa(),thiscondenceintervalcanbecomputedaswell.Usagemifa_ci_fieller(cov_imps,n_pc,conf=0.95,N)Argumentscov_impsListcontainingtheestimatedcovariancematrixwithineachimputeddata.Onecanusecov_imputationsreturnedbymifa().n_pcIntegerorintegervectorindicatingnumberofprincipalcomponents(eigenvec-tors)forwhichexplainedvariance(eigenvalues)shouldbeobtainedandforwhichcondenceintervalsshouldbecomputed.Defaultstoallprincipalcom-ponents,i.e.,thenumberofvariablesinthedata.confCondencelevelforconstructingcondenceintervals.Thedefaultis.95thatis,95%condenceintervals.NAscalarspecifyingsamplesize.DetailsNormally,thisfunctiondoesnotneedtobecalleddirectly.Instead,usemifa(...,ci="fieller").ValueAdataframecontainingcondenceintervalsforn_pcprincipalcomponents.ReferencesFieller,E.C.(1954).Someproblemsinintervalestimation.JournaloftheRoyalStatisticalSociety.SeriesB(Methodological):175-185.SeeAlsomifa()Othermifacondenceintervals:mifa_ci_boot() 10mifa_ci_ellerExamplesif(requireNamespace("psych")){datapsych::bfi[,1:25]mimifa(data,print=FALSE)mifa_ci_fieller(mi$cov_imputations,n_pc=3:8,N=nrow(data))} Indexmifacondenceintervalsmifa_ci_boot,5mifa_ci_fieller,9dplyr::select(),2,6mice,2,5mice::mice,3,6mice::mice(),2,5,6,8mice::mids,5mifa,2mifa(),8,9mifa_ci_boot,5,9mifa_ci_boot(),2,5mifa_ci_fieller,8,9mifa_ci_fiell