Strange observations in N-way anova

[AdityaKir]

Hello,

I have run into a strange problem. I am running N-way anova and what I have found out is if i change the order of the variables that I am using in the anova, then the P-values are changing

E.g. when I run the below procedure

ods noproctitle;
ods graphics / imagemap=on;

proc glm data=WORK.INTERNETDATA;
class Exits Continent Sourcegroup Uniquepageviews Bounces Visits;
model Timeinpage=Exits Continent Sourcegroup Uniquepageviews Bounces Visits /
ss1 ss3;
lsmeans Exits Continent Sourcegroup Uniquepageviews Bounces Visits /
adjust=tukey pdiff=all alpha=0.05 cl;
quit;

The value of Sourcegroup is significant i.e. 0.0037

Now when i change the sequence of the variable Bounces to something like this

ods noproctitle;
ods graphics / imagemap=on;

proc glm data=WORK.INTERNETDATA;
class Bounces Exits Continent Sourcegroup Uniquepageviews Visits;
model Timeinpage=Bounces Exits Continent Sourcegroup Uniquepageviews Visits /
ss1 ss3;
lsmeans Bounces Exits Continent Sourcegroup Uniquepageviews Visits /
adjust=tukey pdiff=all alpha=0.05 cl;
quit;

The value of the variable Souregroup is no longer significant i.e. 0.0752

Can somebody tell me why is this occuring.

Thanks and regards,

Aditya

[Ksharp · Accepted solution]

It is difference between SS3 and SS1.

If you only use SS3 , you will get the same P-value , otherwise if you use SS1 ,you will get different one.


proc glm data=sashelp.class;
class age sex;
model weight=age sex/solution ss3;
run;

proc glm data=sashelp.class;
class age sex;
model weight=sex age/solution ss3;
run;