## what is assumption for repeated measures analysis using proc GLM, MIXED, GENMOD?

### what is assumption for repeated measures analysis using proc GLM, MIXED, GENMOD?

For continued response data.

As we learned that there are three assumption for employ ANOVA:
independent observations
normally distributed data for each group
equal variances for each group

But for REPEATED MEASURES data the first one will be violated so we us
some appoaches to accourt for the correction. but the rest two
assumptions should still be met or not?

for example,

person    Before    Drug_A    Drug_B    Drug_C

1         94        67         90       67
2         57        52         69       55
3         81        74         69       73
4         82        59         71       72
5         67        65         74       72
6         78        72         80       72
7         87        75        106       74
8         82        68         76       59
9         90        74         82       80
....

<this is a cross-over study design, the three drug are measured at the
same subject four times. the response variable is heart reate, the
DRUG B is placebo>

my code is;

PROC GLM DATA=mydata;
MODEL Before Drug_A Drug_B Drug_C =/NOUNI ;
REPEATED hrate 4 CONTRAST(1)/ PRINTE SUMMARY;
REPEATED hrate 4 CONTRAST(4)/ PRINTE SUMMARY;
RUN;QUIT;

then i got a output about the spericity test and something else. but
any other thing i forgot? are ther other assumption should be tested?

Any suggestion or reference are appreciated! thanks in advance.

Johson Chang

Hi,

I have two problems using proc GLM and it would be really great if someone could help me...

Problem 1.
I perform a mixed anova model with proc GLM.
2 categorial variables: 1 is a fixed effect (2 modalities; ie mating system "MS": selfed versus outcrossed); 1 is a random effect (14 modalities; different families "fam")

I need the variance and covariance to compute with the delta-method the variance of the ratio of the estimates for the fixed effect within each family.

The model is size1 = MS fam MS*fam
random fam MS*fam / test

But with the command: lsmeans  MS*fam / cov
the variance-covariance matrix only includes variances, all covariances are null (valeu of 0). Is that normal ?

I performed the computation of the COV matrix with data of example 30.8, and the COV matrix is also diagonal, with null covariances.

Problem 2.To the previous data, let add a continuous variable (ie size0).
I therefore have a mixed model with:
- a fixed categorial effect "MS"
- a random categorial effect "fam"
- a fixed continuous effect "size0"

Model built is: size1= MS fam size0 MS*fam MS*size0 fam*size0
random fam MS*fam / test

When I look at the results of the analysis, the formula given for the error term for effect "MS" does not correspond to the value given by SAS.
Formula is: 0.0021*meansquares(fam*MS) + 0.9979*meansquares(error)
Given the values computed by SAS for the meansquares of fam*MS and error, this would give a value of 0.001494 whereas the value given by SAS is 0.000003017.
The effect is then highly significant whereas non-significance is expected a priori.

Is that normal ? What is the problem ? What should I do ?

Many many thanks for your help.
Pierre-Yves

_____________________________________________________

Pierre-Yves Henry
Groupe "Biomtrie et Biologie des Populations"
Groupe "Gntique et Dynamique des Populations"
Centre d'Ecologie Fonctionnelle et Evolutive
CNRS
1919 Route de Mende
34293 Montpellier Cedex 5
France

Tel: 04 67 61 32 20
Fax: 04 67 41 21 38