Repeated Measures Analysis

[Jep]

Hello,

I need some help! My first question is that,I am trying to estimate the slope for women(coded 0) and men(coded 1) and the difference in slopes between men and women. There are 4 time points( baseline, 3 months, 6 months and 12 months). As you can see in the proc. mixed statement below, there is a quadratic term in the model and it is significant. My problem is writing an estimate or contrast statement for a model with quadratic terms.

My second question is that I modelled time as a continuous outcome but the rate of change is not constant overtime, (it is curvilinear) is it better to model time as discrete in such a situation? If that is the case, how can I obtain the overall mean slope for the 12 month period  for both genders and the overall gender difference?

I am using the Enterprise guide environment.

proc mixed data=merged;

class  patient gender(ref ='0') t ;

model wt = time gender time*time gender*time  gender*time*time  / s chisq;

repeated t /subject=patient type=un;

run;

[sld]

Although it is possible to fit a quadratic model to 4 time points, so few points provide very little support for the fit. Unless I had a strong theoretical foundation for using the quadratic model, I probably would incorporate time as a classification (discrete) factor.

If the relationship between wt and time is quadratic and if by "slope" you mean the parameter associated with the linear term in the model, then what does slope measure that is of use to you? Looking at only slope in a quadratic model does not make a lot of sense to me, but there may be something about your context that justifies looking at slopes.

If time is incorporated as a classification factor and you want to estimate polynomial trends (including linear), then see SAS Usage Note 22912. The examples use GLM, but the ideas are the same for MIXED. As the Usage Note illustrates, regression is interchangeable with ANOVA with polynomial contrasts.

If you are interested solely in the linear trend (even if the data appear to follow a quadratic trend), then you could just fit a linear model and skip the need for contrast statements.

I hope this helps.