Prev: Binary data
Next: Is Regression Using Proc IML Faster?
From: Ryan on 23 Jan 2010 12:37
Hi,

I recently read an article in the Journal of Consulting and Clinical
Pyschology whose authors rerpoted analyzing data from a clinical trial
study testing the efficacy of a family cognitive behavioral
intervention for children with depressed parents. The study was
designed such that those in the intevervention condition were treated
within a group setting (with all memers of one family present) part of
the time, while those in the control condition were not treated in a
group setting at any time (mailed informational packets). In the data
analytic section, they report running a "mixed-effects" model in PROC
MIXED to account for partial nesting. There was also discussion
comparing models of homogeneous variances versus heterogeneous
variances across time and condition.

Does anyone know how to write a partially nested model in proc mixed?
How would I write the proc mixed code assuming, say, two conditions
(experimental and control), four time points (baseline, post, 1st f/u,
and 2nd f/u), 20 family units/groups in the intervention condition,
and a continuous outcome.

If for *both* conditions participants were treated within group/family
settings, I imagine a potentially valid parameterization of the model
wold be something like this code (assuming group values do not repeat
for each condition):

proc mixed data=mydata;
class person group condition time;
outcome = condition|time / s ;
repeated time / subject=person type={Unstructured, Autoregressive...};
random intercept / subject=group;
run;

I'm just not sure how to modify this code to account for partial
nesting. All thoughts/suggestions are welcome.

Thanks!

Ryan
From: Ryan on 23 Jan 2010 12:45
On Jan 23, 12:37 pm, Ryan <ryan.andrew.bl...(a)gmail.com> wrote:
> Hi,
>
> I recently read an article in the Journal of Consulting and Clinical
> Pyschology whose authors rerpoted analyzing data from a clinical trial
> study testing the efficacy of a family cognitive behavioral
> intervention for children with depressed parents. The study was
> designed such that those in the intevervention condition were treated
> within a group setting (with all memers of one family present) part of
> the time, while those in the control condition were not treated in a
> group setting at any time (mailed informational packets). In the data
> analytic section, they report running a "mixed-effects" model in PROC
> MIXED to account for partial nesting. There was also discussion
> comparing models of homogeneous variances versus heterogeneous
> variances across time and condition.
>
> Does anyone know how to write a partially nested model in proc mixed?
> How would I write the proc mixed code assuming, say, two conditions
> (experimental and control), four time points (baseline, post, 1st f/u,
> and 2nd f/u), 20 family units/groups in the intervention condition,
> and a continuous outcome.
>
> If for *both* conditions participants were treated within group/family
> settings, I imagine a potentially valid parameterization of the model
> wold be something like this code (assuming group values do not repeat
> for each condition):
>
> proc mixed data=mydata;
> class person group condition time;
> outcome = condition|time / s ;
> repeated time / subject=person type={Unstructured, Autoregressive...};
> random intercept / subject=group;
> run;
>
> I'm just not sure how to modify this code to account for partial
> nesting. All thoughts/suggestions are welcome.
>
> Thanks!
>
> Ryan

As a point of clarification, the study to which I refered provided the
intervention [in the experimental condition] in multiple group
settings (within families). In other words, if I understood correctly,
participants were treated in their family units. I didn't mean to
suggest that everyone in the experimental condition was treated in one
group. Each family unit was treated together, while in the control
condition, informational packets were sent out. I'm interested in
learning how to run such a model in proc mixed.

Thanks again,


Ryan
 | 
Pages: 1
Prev: Binary data
Next: Is Regression Using Proc IML Faster?