1. 用图表检验Analyze -> regression -> Linear-> Plots
5 {% l$ F# Q3 UScatter plot of the standardised residuals on the standardised predicted values (ZRESID as the Y variable, and ZPRED as the X variable% @$ _ L7 P+ I
7 z6 i7 w* r0 t1 v9 _ j3 u
如果图表显示有可能存在异方差,需要用统计检验来进一步检测异方差是否确实存在。7 a9 s2 K7 o- e8 ]7 Y2 ~$ ~5 f3 a
2. 用统计检验
3 L. w8 ]% X/ L/ g. ?( }Heteroscedasticity——Testing and Correcting in SPSS.pdf
Gwilym Pryce March 2002.doc
(172.5 KB, 下载次数: 4)
- G( N7 k8 t0 [. F
( W2 R5 O# c3 u9 P0 W! }
: L; D6 T* v; q/ \Levene’s Test
/ _9 N+ r) X0 ~8 s/ GGoldfeld-Quandt Test
7 ^* I7 |4 i: S2 ~1 A: }' XBreusch-Pagan Test/ b/ Y) R9 S; p: d/ T) u
White‘s Test (比较常用来检验异方差)
9 z0 i: }4 P/ ~) k( E/ H1 cAssume you want to run a regression of wage on age, work experience,education, gender, and a dummy for sectorofemployment (whether employed in the public sector). wage = function(age, workexperience, education, gender, sector) or, as your textbook will have it, wage = b1 + b2*age + b3*work experience+ b4*education + b5*gender + b6*sector The White’s test is usually used as a test for heteroskedasticity. In this test, a regression of the squares ofthe residuals is run on the variables suspected of causing theheteroskedasticity, their squares, and cross products. (residuals)2 = b0 + b1 educ + b2 work_ex+ b3 (educ)2 + b4 (work_ex)2 + b5(educ*work_ex)
# n2 R( e* C( O, |' g9 ?8 N F$ h1 u! s
. i, M0 ]+ ?4 o) }. S* Y# r
4 n( C4 K# {* L) o! |
6 I& C4 v8 b7 z9 U/ W/ Z
White’s Test · Calculate n*R2 à R2 = 0.037, n=2016 à Thus, n*R2 = .037*2016 = 74.6.
. ~. q {1 Q4 R. D# |6 i+ M· Compare this value with c2 (n), i.e.with c2 (2016)
3 k4 {' N) V1 q4 b0 h% y v; E4 A(c2 is the symbol for theChi-Square distribution)
, a- ` c; {% u- g! [+ C3 }/ O6 Fc2 (2016) = 124obtained from c2 table. (For 955 confidence) As n*R2 < c2 ,heteroskedasticity can not be confirmed.
' a" z* r) _; n& [4 z0 t4 I! q, x, F
请参考:regression_explained_SPSS
regression_explained_SPSS.doc
(368 KB, 下载次数: 0)
+ o* |. A% k4 j& c" o
|