多層次模型 (Multilevel Model)

多層次模型 (Multilevel Model)

簡單介紹與使用SAS處理Multilevel Model，

本篇使用到三個比較常用的簡單模型，分別為The Variance components model、The Random intercept model、The Random slope model。

SAS Codes：

LIBNAME lib EXCEL "D:\SasData\jsp.XLSX" ;

data jsp; set lib."jsp2$"N; run;

ods graphics on;

/*The_variance_components_model*/

Proc mixed  data=jsp  method= ml noclprint covtest;

Class Gender  Social_class  School_ID;

Model  math_yr_3=   /s ddfm=bw  OUTP=p1  OUTPM=pm1   RESIDUAL;

Random intercept    /type=un   sub=School_ID;

Run;

/*畫圖*/

goptions reset=global;

axis1 label =( f= "arial/bo" h=1.2 "8-year math score")

order = (0 to 45 by 5);

axis2 label =(a=90 f= "arial/bo" h=1.2 "11-year math score")

order = (26 to 37 by 5);

title "The variance components model";

symbol i=j r=50;

proc gplot data = p1;

where school_id<51;

plot pred*math_yr_1=school_id/ haxis=axis1 vaxis=axis2;

run;

quit;

/*Random_intercept_model*/

Proc mixed  data=jsp  method= ml noclprint covtest;

Class Gender  Social_class  School_ID;

Model  math_yr_3=math_yr_1  /s ddfm=bw OUTP=p2  OUTPM=pm2   RESIDUAL;

Random intercept  /type=un

sub=School_ID;

Run;

/*畫圖*/

goptions reset=global;

axis1 label =( f= "arial/bo" h=1.2 "8-year math score")

order = (0 to 45 by 5);

axis2 label =(a=90 f= "arial/bo" h=1.2 "11-year math score")

order = (0 to 45 by 5);

title "The Random Intercept Model";

symbol i=j r=50;

proc gplot data = p2;

where school_id<51;

plot pred*math_yr_1=school_id/ haxis=axis1 vaxis=axis2;

run;

quit;

/*Random_Slope_model*/

Proc mixed  data=jsp  method= ml noclprint covtest;

      Class Gender  Social_class  School_ID;

      Model  math_yr_3=math_yr_1  /s ddfm=bw    OUTP=p3  OUTPM=pm3   RESIDUAL;

      Random intercept  math_yr_1  /type=un   sub=School_ID; 

Run;

proc print data=pm3;

run;

/*畫圖*/

goptions reset=global;

axis1 label =( f= "arial/bo" h=1.2 "8-year math score")

order = (0 to 45 by 5);

axis2 label =(a=90 f= "arial/bo" h=1.2 "11-year math score")

order = (0 to 45 by 5);

title "The Random Coefficient Model";

symbol i=j r=50;

proc gplot data=p3;

where school_id<51 ;

plot pred*math_yr_1=school_id/ haxis=axis1 vaxis=axis2 href=0;

run;

quit;

資料來源：
http://www.bristol.ac.uk/cmm/

資料下載：https://www.dropbox.com/s/hl0h1uqkin44gtf/jsp.xlsx