ABSTRAK FITIUYANI. Kajian Data Hilang Dalam R
e Linear
bimbigan ERFlANl dan AGUS M. I ~ m o rDibawah .
Dengan Meiode AFiing-
SOLEH.
Pmelitian ini dilakukan dengan tujuan untuk mmgaahui mecode tmbtik dalam pcndugaair panvneae pada regresi linear berganda dmgan dua peubah pmjelas XI clan X2. Pada X2 u d q W data hilang secara acak sedangkan peubah rrspon(Y) dan XI ~elalulaamali. Metode c ~ t p l e i ~ - c u s emcntpakm salah satu penddratan ) m g s a i n g d i g u h untuk men-i data hilans salab -1'~ dala hilang dalam model rrgresi linear Metode compkte-care m t n g h i l k a n pmduga yang b i t % tak bias, tetapi kurang efisien jika jumlah data yang hilang cukup ksar. Maode missing-idcaw rnpendekatan lain )9ng digunakan dengan mcnambahlran ptubah indikator pada peubah pcnjelas dalam model r e p s i . Maode ini ta&i dari metode misting-i&afor I dan metode missi-indimor 2. Parelitian ini menggmakan pendekatan simdasi untuk membandin-ekan metode comfletecase, metodt missing-iiwkmw I dan metode missing-indicator 2 pada muktur daia dengan b e h g i ukuran contoh (n) dan standar dcviasi sisaan (aJ. Ulnwn cantoh, standar deviasi sisaan dan parameter yang digunalcan dalam penelilian ini adalah n = 20,30, SO1 100,50; 4 = 1,2,3, 15, 120; Po = I, 8,= 4P2= 2 . Perscntase jumlah data hilang j m g dianalisis adalah 5, 10: 15,20,25, 30. U k m kebaikan model mruk pcndugun paiamater dilihat dari nilai R'+t dari masingmasing penddrstan. Hasil simdasi menunjuWcan bah\va semakin besar ukuran carrtbh ) m g diambil, maka ukuran kebaikan pcndugaan paramem metode complete-care scmakin mendtkati u l i i kcbaikan pcndugaan paranreter pada data yang lmgkap dan ulnwn bebailan pendugaan paramacr maodc mixring-idcafor 1 scmakin mendem metode missing-idcator 2. Bdasukm kcbaikan pcndugaan y y a , pads kondisi standar deviasi sisaan ka5l metode com$ete-case lebih baik dibandingkan metode missing-idcator 1 dan melode missing-indicufor 2. Unuk 51andar de\iasi sisaan yang cukup be=, tidak talihat peddaan dari kaiga maode m b n dan nilai R2a&wtcarderung mcndekati nol. Secara mum, metode compleie-caw m a u m maode M k d i b d h g k a n muode missing-indicator I dan metode mirring-indicator 2. Kata Kunci: metode complere-cuse, n k i d e missing-idxator 1, metode missing-indicator 2, data tiU IcngkaR data hilang, R2-ta#~sr
ABSTRACT FITIUYANI. The study o f Missing Dam in Mulliple Limar Regmion with Missing Indicator Methods. Advisory Cornminee by ERFIANI dan AGUS M. SOLEH. This research is doing in orda to knowing \vhicfi the best methods that can k used to &mate independent wariable XI dan X2. X2 has the parameter in multiple linear rcgcsion with missing data by random, meanww.hile response variable (Y) and XI are dwma)5measured.
Complatcasc methods is one of the approaching mahods that common uxd to handle a missins data, one example is missing data in multiple linear rrgrrssion. Complmahods pmducig an unbiased tstimauw but h e &mator is not to efficient if h e amount of missing dam is large. Missing-indicator is o m otbcr kind of the approadiing mahods can be used with adding the indicator variable in to explained variable in m i o n model. This mdhods is mnsisl of missing-indicator 1 mahods and missing-indi-r 2 methods. Sirnuhion meihods is using in this restarch in or& to compare mmplettcase methods, missing-indicator 1 methods and missing-indicator 2 for e v a y data shucaac. Amount of stMinic (n), error standard d e v i i o n (03and paramem which used in chis rrscarch are n = 20,30,50, 10(1, 50; q = 1,2,3, IS, 120; Po = I,& = IlP2= 2 . The pcmmtage of analysis m k i n g data a!r 5, 10, 15, 20, 25, 30. Smdard goodmss of model for estimating the p a r a w can be Imowm from R2fxljiu7 \due fTom each approaching mahods. According to simulation melbods indicate increasingly amount of aalisrics, tht cornpie of d l for cstimming the paramas, and missingmuhods appmarh dl dau standard g I ' indicator 1 mahods approach missing-indicator 2. According to standard goodness of model for O &mating the paramem: in condition small am standard dwiation, h e complmethods is much bcclcr than the missing-iodim I m.(hodr and misingindicator 2 mahorn. Datp with l q e error standard deviationt the R'-sdjust from the three models is not quite diffarnt and che vaiue is closely to zero. The completocase methods is the best than misingindicaua 1 mahods and missing-indicator 2 methods. -
3
W
B 3
>
Key words: complmethods, missing-indicator I melhods, missiigindicaror 2 methods, incomplete data, missing data, R'-@~LsI.