data structure question

[Janell Mensinger]

Hi everyone,
I am working with a physician on a project that has a complex data structure
that I am unsure as to how to handle.  I will try to describe briefly. He is
testing a new form for attending physicians to use to evaluate a resident
during a clinical encounter with a patient.  He recruited 40 docs from 4
sites to participate.  These 80 docs were randomized to an intervention
group (the new form), or a control group (standard form).  He has 40 docs in
each condition. He had the docs watch videotaped sessions of 6 different
residents performing a clinical interaction with a patient.  So now we have
480 completed forms (80 docs rated 6 clinical encounters). One of his aims
is to show that the quantity of comments from each clinical encounter is
greater using the new form compared to the standard form.  So my outcome is
a quantity for 480 different forms. Site and doc are clearly random effects
and my intervention is a fixed effect, so I have a mixed model (I am
assuming I can model this linearly although I have not seen the data yet). I
am not sure how to handle the variable denoting which of the 6 clinical
scenarios the form is linked to.  I am guessing that the ratings associated
with each scenario would be highly correlated thus it needs to be treated as
some form of repeated measure, no? Furthermore, I have always used Proc
Mixed in SAS to run mixed models but I don't have access to SAS in my
current institution.  Can SPSS handle a model like this?  I would greatly
appreciate any advice on this.

Thanks in advance!
Janell Mensinger

[Richard Ristow]

At 12:53 PM 9/21/2007, Janell Mensinger wrote:

>[The investigating physician] recruited 40 docs
>from 4 sites to participate.

Slip? That should be 80 docs? And I assume, 80
among the 4 sites, not 80 (or 40) from each.

>These 80 docs were randomized to an intervention
>group (the new form), or a control group
>(standard form).  He has 40 docs in each
>condition. He had the docs watch videotaped
>sessions of 6 different residents performing a
>clinical interaction with a patient.  So now we
>have 480 completed forms (80 docs rated 6 clinical encounters).

Clarifying, to make sure: It sounds from what you
say later that these are 6 encounters all told, not 480 different ones.

So, you have encounter (6 levels) crossed with
form (2 levels); 12 cells, 40 observations per
cell. Ignoring random effects, it looks like
two-way ANOVA, where you're interested in the
'form' main effect. (I wonder what the interaction would tell you?)

Then, site and physician as random effects nested
within 'form', as a mixed-effects model. With the
usual effect of (correctly) reducing apparent
statistical power, versus the naïve assumption
the ratings are independent.

>I have always used Proc Mixed in SAS to run
>mixed models but I don't have access to SAS in
>my current institution.  Can SPSS handle a model like this?

MIXED, I think in module Regression Modules, does
this. I don't know it well enough to suggest syntax.

-Good luck,
  Richard