turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Why the covariate does not affect the p values of ...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

07-06-2016 10:04 AM

I am running an ANCOVA on some cross-over study data. The SAS code snippets for the model is:

proc mixed data=free_ancova1; class ptno treat period seq; model s_free = s_bl period seq treat/ddfm=kr noint residual solution; lsmeans treat/pdiff; random ptno(seq); run;

The "s_bl" is the baseline recorded during a run-in period. The SAS output is:

I also ran an ANOVA without the baseline as the covariate:

proc mixed data=free_ancova1; class ptno treat period seq; model s_free = period seq treat/ddfm=kr noint residual solution; lsmeans treat/pdiff; random ptno(seq); run;

And the output is:

Can someone please help me understand:

1. What does it mean for "s_bl" p value <0.0001?

2. Why does the baseline covariate only affect sequence (SEQ) but not other variables?

Thanks.

Accepted Solutions

Solution

07-06-2016
10:50 AM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to jj02148

07-06-2016 10:42 AM

1. The small p-value for s_bl indicates that there is a real effect for s_bl. That is, it seems to be significant in the model for predicting the response. Your other p-values indicate that the other effects are not significant. Either they do not predict the response or you do not have enough data to detect the effect.

2. When you get rid of the s_bl variable, the degrees of freedom in the model change. By looking at the "Den DF" column, you can see that the SEQ variable gained an extra degree of freedom in the denominator of the F test. But why only that effect? Well, you've specified the DDFM=KR option, which makes various adjustments to the denominator degree of freedom. I don't fully understand the details, but if you change the DDFM= option, you are likely to get different results.

But the result is the same: the PERIOD and SEQ variables are not signicant in either model that you've constructed.

All Replies

Solution

07-06-2016
10:50 AM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to jj02148

07-06-2016 10:42 AM

1. The small p-value for s_bl indicates that there is a real effect for s_bl. That is, it seems to be significant in the model for predicting the response. Your other p-values indicate that the other effects are not significant. Either they do not predict the response or you do not have enough data to detect the effect.

2. When you get rid of the s_bl variable, the degrees of freedom in the model change. By looking at the "Den DF" column, you can see that the SEQ variable gained an extra degree of freedom in the denominator of the F test. But why only that effect? Well, you've specified the DDFM=KR option, which makes various adjustments to the denominator degree of freedom. I don't fully understand the details, but if you change the DDFM= option, you are likely to get different results.

But the result is the same: the PERIOD and SEQ variables are not signicant in either model that you've constructed.