GEE in SPSS

[torvon]

What model would you recommend for a repeated measurement analysis of an ordered variable (0,1,2,3) with 5 measurement points, possibly with a quadratic slope (it fits better graphically), in which the dependent variable is quite strongly correlated over time, and in which I have some baseline factors and covariates, but also 2 time-varying covariates?

After baseline measurement a major stressor happens, so there is drastic increase from time1 to time2, after that is stays pretty stable. Measurement points are only 3 months apart, therefor the high correlation. I'm currently thinking about using a GEE, but I'm not sure if that's the right way to go.

GEE only looks at variance in slope differences predicted by covariates, right? How would I check for intercept differences predicted by covariates? I want to distinguish whether (1) people are different at baseline (regarding to covariates) and (2) whether these difference carry over the stressor, or whether different groups of people react differently to the stressor.

I understand that GEEs are used because it allows for AR(1) covariance structure, which is important in my case because dependent variable is correlated over time.

What about correlated predictors? Maximum correlation between predictors is r =.3, which isn't collinearity, but it's more than nothing.

Thank you
E.

[Lemon, John S.]

I'm not in the office until 08:00 Monday 10th September 2012 - I will try and respond on my return.

The University of Aberdeen is a charity registered in Scotland, No SC013683.

[Rich Ulrich]

What model?  Well, you start with "What hypothesis?"

You think there should be a major change from time1 to time2
because there is a major stressor.  It stays stable after that.

That sounds like a t-test is the simple test for time1 to time2.
Or you build an ANCOVA if there are covariates to consider.

And you look at repeated measures - linear trend? - for the
time periods following that.  You need to figure out your
hypotheses before you worry so much about what procedure.

--
Rich Ulrich

> Date: Tue, 4 Sep 2012 14:13:15 -0700

> From: [hidden email]
> Subject: GEE in SPSS
> To: [hidden email]
>
> What model would you recommend for a repeated measurement analysis of an
> ordered variable (0,1,2,3) with 5 measurement points, possibly with a
> quadratic slope (it fits better graphically), in which the dependent
> variable is quite strongly correlated over time, and in which I have some
> baseline factors and covariates, but also 2 time-varying covariates?
>
> After baseline measurement a major stressor happens, so there is drastic
> increase from time1 to time2, after that is stays pretty stable. Measurement
> points are only 3 months apart, therefor the high correlation. I'm currently
> thinking about using a GEE, but I'm not sure if that's the right way to go.
>
> GEE only looks at variance in slope differences predicted by covariates,
> right? How would I check for intercept differences predicted by covariates?
> I want to distinguish whether (1) people are different at baseline
> (regarding to covariates) and (2) whether these difference carry over the
> stressor, or whether different groups of people react differently to the
> stressor.
>
> I understand that GEEs are used because it allows for AR(1) covariance
> structure, which is important in my case because dependent variable is
> correlated over time.
>
> What about correlated predictors? Maximum correlation between predictors is
> r =.3, which isn't collinearity, but it's more than nothing.
> ...