Calcite | Level 5

## repeated measurements mixed procedure-comparisons between treatments across data points

All statistical analyses are conducted in SAS 9.4

I have a 2x4 experimental design with Diet (4 increments of nutrient x,  0, 5, 10 and 15%) and infection (Infected vs control) as my independent factors and feed intake (FI) is the dependent variable .

The replicate is pen (6 animals per pen, 6 pens per treatment combination) and I measure feed intake over 15 days.

FI is analysed with the repeated measurements mixed procedure (PROC MIXED)

Covariance structures were chosen based on the lowest value for the Akaike and Bayesian information criteria. Given this covariance structure, fixed effects were tested, and least square means along with pooled standard errors (SEM) are calculated using the LSMEANS and PDIFF-statements of PROC MIXED, respectively. For multiple comparisons I follow the Bonferroni adjustment.

For all statistical procedures, the normality of the residuals are assessed with the Shapiro-Wilk test.  Significance was determined at P < 0.05. All values are expressed as model-predicted least square means with the SEM.

The code:

proc mixed data=Coc;

class diet infection day pen;

model fi=diet infection day diet*infection diet*infection*day;

repeated day/ subject=pen type=ANTE(1);

lsmeans diet infection day diet*infection*day/ pdiff=all Adjust=Bon; run;

1st Question Should I insert pen as a random factor in the model (e.g Pen(diet)) or not. If not can you help with syntax?

2nd I am primarily interested in the 3-way interaction and pairwise comparisons only within the same day are of interest. Pdiff gives me all possible comparisons which are of no interest to me. Should I use a different approach e.g use the SLICE option or should I use a CONTRAST statement?

3rd Could you provide a bit of help with the coding according to the approach (es) I should follow?

Calcite | Level 5

## Re: repeated measurements mixed procedure-comparisons between treatments across data points

After a lot of reading I realize that if the interaction with diet and infection is significant then I can either (correct me if I am wrong)

a) Run separate Anova's on each of of the days with diet infection and diet*infection in my model

b) The Slice approach gives  tests of simple effects

c) use pre-planned contrasts for each day which is an overkill

D) Run multiple comparisons (pdiff).

For the D) approach there are multiple ways of adjusting for multiple comparisons.

The question is whether the pdiff=all option is appropriate or whether I need to use the i.e the adjust=Tukey option.

However the adjust options will take into account all 15 days when calculating pairwise comparisons.

Since I am interested only in pairwise comparisons within the same day which would be the best approach?

Is there someone who could help?

Discussion stats