Abstrak
M etode rank nonparametrik pada model regresi linear
Oleh :
Kusuma - M0102004 - Fak. MIPA
AB STRAK Ku su ma, 2007. ME T ODE RAN K NONP ARAME T RI K P AD A MODE L
RE GRE SI LI NE AR. Faku lt as Mat e mat ika d an I lmu Pe nget ahu a n Ala m. Univers it a s Sebe la s Maret .
Y X X X b b b b e = + + + + + L merupaka n mo de l r egres i linear de nga n
Per sa ma a n
i
0 1 1 2 2 k k
b ada la h para met er r egresi ya ng d ie st ima s i berdasarka n dat a penga mat a n. Meto de kuadrat t erkecil merupak an met o de est imas i para met er
r egres i ya ng dapat me mber ika n has il ya ng o pt ima l jik a ses at annya d ia su ms ika n berd ist r ibus i no r ma l, ( )
2
, 0 ~ s e N . J ik a ke no r ma la n t id ak d ip e nu hi mak a est ima s i para met er r egres i ya ng d ipero le h t id ak t epat. Sesat an ya ng t idak
berd ist r ibus i no r ma l dapat diind ika s ik a n deng a n adanya pe nc ila n ( outlier) . Met o de r ank no npara met r ik merupaka n met o de est imas i para met er r egres i ya ng dapat digu naka n unt uk me nga na lis is d at a jik a sesat annya t idak berd ist r ibus i no r ma l ya ng d iind ika s ika n de nga n ada nya pe nc ila n. T u jua n da la m pe nu lisa n skr ips i ad a la h me ne nt ukan e st ima s i para met er r egres i da n u ji s ig nifik a ns i para met er r egres i u nt uk me nget ahu i hu bu nga n ant ara var ia be l be bas de nga n var ia be l t ak be ba s me nggu naka n met o de r ank no npara met r ik. Met o de ya ng d igu naka n da la m p e nu lis a n skr ip s i ad a la h st udi lit e r at ur . Berdasarka n has il pe mba has a n dapat dis impu lka n ba hwa est ima s i para met er r egres i d ipero le h de nga n me min imu mka n ju mla h r ank s is aa n ber bo bo t. Hipo t es is
ya ng d igu naka n pada r egres i linear sed erha na ada la h 0 :
= b H dan 0 :
„ b H denga n st at ist ik u ji
( ) U S D U
t = . H ipo t esis no l
0
0
1
H dit o lak jika a < p denga n p = Prob [ ] t T ‡ dan nila i p d ipero le h me nggu naka n t abe l d ist r ibus i t de ng an dera jat be ba s 2 - n . Pada
r egres i linear ga nda, hipo t esis ya ng d igu naka n ad a la h
: 0
H b
l k
1 1, ,
+
K
„
denga n st at ist ik u ji
JR S B JR S B F
- =
ter edu ksi p enu h rank
( )
k l c t
-
: 0
H b b
. H ipo t esis no l
0
l k
0 1
+
= = = L da n
H d it o lak jika a < p de nga n p = P rob [ ]
F F ‡ dan nila i p d ipero le h me nggu naka n t abe l d ist r ibus i F de ngan dera jat be ba s k l - dan
1 n k - - . Kata kun ci: model regresi linear, metode rank nonparametrik
ran k
AB STRACT Ku su ma, 2007. NONP AR AME T RI C R AN K ME T HOD ON LI NE AR
RE GRE SS I ON MODEL. Facu lt y o f Mat he mat ic s and Nat ur a l Sc ie nce s. Sebe la s Maret Univers it y.
Y X X X b b b b e = + + + + + L is a mo de l o f a line ar r egress io n w it h
T he equat io n
i
0 1 1 2 2 k k
b are r egress io n para met ers whic h are est imat ed based o n t he o bservat io ns o f dat a. T he least square met ho d is a met ho d to est imat e t he
r egress io n para met ers t hat gives a n o pt ima l r e su lt if t he err o r t erms as su med ha ve no r ma lly d ist r ibut ed, ( )
2
, 0 ~ s e N . I f t he no r ma lit y as su mpt io n is no t sat is fie d t hen e st imat io n o f r egress io n para met ers is not exact . T he vio lat io n o f no r ma lit y
assu mpt io n is ind icat ed by t he o ccur ence o f o ut lier s. T he no np ara met r ic r a nk met ho d can be used t o ana lyze t he dat a if t he err or s ha ve no t nor ma lly d ist r ibut io n w hic h ind icat ed by t he o ccur ence o f o ut liers. T he a ims o f t he fina l pr o ject are to est imat e t he r egress io n para met ers and t o t est t he s ig nif ica nc e o f r egress io n para met ers to kno w the r e lat io ns hip o f indepe nde nt var ia ble w it h depe nd e nt var ia ble, us ing t he met ho d o f no npara met r ic r ank. T he met ho d used in t his fina l pr o ject is a lit erar y st ud y. Based o n t he d is cus s io n, it ca n be co nc luded t hat est imat io n o f r egress io n para met ers is o bt a ined by min imiz ing t he su m o f r ank – we ig ht ed r es idua ls. T he
hypo t hes is used o n s imp le linear r egress io n is 0 :
= b H ver sus 0 :
„ b H w it h t he t est st at ist ic s
( ) U S D U
t = . T he zer o hypo t hes is
0
0
1
H is r e ject ed whe n a < p where p = Prob [ ] t T ‡ and p va lue is o bt ained by u s ing t d ist r ibut io n t able w it h n – 2 degrees o f fr eedo m. O n
t he mu lit ip le linear r egres s io n, t he hypo t hes is used is
versu s
: 0
H b
l k
1 1, ,
+
K
„
w it h t he t est st at ist ic s
JR S B JR S B F
- =
ter edu ksi p enu h rank
. T he zer o hypo t hes is
0
( )
k l c t
-
: 0
H b b
l k
0 1
+
= = = L
H is r e ject ed whe n a < p w here p = Prob [ ]
F F ‡ and p va lue is o bt a ined by u s ing F d ist r ibut io n t able w it h k – l a nd n – k – 1 degree s
o f fr eedo m. Key words: linear reg res sion model , nonparametric rank method
ran k