Posted by
Maguin, Eugene on
Feb 07, 2011; 3:24pm
URL: http://spssx-discussion.165.s1.nabble.com/Re-posting-conditional-logistic-regression-for-agreement-study-tp3373568p3374424.html
Jan,
I'm sure there are others more knowledgable but here is my thinking. I think
that what you'd like to be true is that any third variable has equal
relationships with both DVs. Basically, a common cause model. You have three
confounders: MD_Type, MDE, HxSA. MDE and HxSA might be used as predictors of
current suicidality (ignore MD_type for the moment). Ideally, each and both
would have equal relationships with both DVs. You can look at bivariate
relationships easily enough. It'd be nice to compare the two resulting
statistics much as you might compare two correlations. I don't know how to
do that. By the way, note that these are dependent associations because they
share a common variable. The best alternative, not available in spss, is to
fit a simple three variable path model and constrain the path coefficients
to equality. Mplus is the software to use although Lisrel or EQS could do
this too. However, because the DVs are categorical, I think the process may
be more complicated. I have little experience in this.
MD_Type is a different problem. I assume that either a staff or a resident
did the MD rating. Thus, MD_Type is a moderator variable. So the question is
whether staff rate the same as residents. Again, bivariate techniques will
give you insight but not a unified, i.e., single statistic, answer. Perhaps
there are ways to compare two independent chi-squares or other categorical
stats. I don't know them. The basic analysis here is just a multiple group
problem. Again, an Mplus problem but now much more complicated because of
the ordinal DV.
I'd say the moderator question comes before the common cause question
because if you can knock out MD_Type, you can legitimately combine your two
MD_Type groups. Even though you could work this a two step problem, you
could combine the everything into single path model. So, two groups defined
by MD_type. In each group both DVs are predicted by MDE and HxSA. What you
want is that the totally constrained model fits as well as the unconstrained
model.
Gene Maguin
Hi,
I'm re posting this since I didn't get any replies from my Jan 26th post. I
hope this is the correct etiquette.
Hi,
I have 2 variables (veteran rating of suicidality and psychiatrist rating of
veteran suicidality) with the same 5 level ordinal rating (n=482).
I know that the survey results and the psychiatrist's rating are not
independent since they are rating the same entity-the study participant.
I am using a combination of methods to examine agreement including:
looking at the crosstabs for patterns, calculating proportion of agreement,
kappa, ICC, and McNemar.
I understand that modeling (specifically conditional logistic regression)
can be used and thus potential confounders can be examined.
Can this be done in SPSS? If so, I'm confused as to what would be the DV and
IV. I have 3 potential confounders.
Here is a sample of my data
Data List List/ ID Vet_Rate Psych_Rate MD_Type MDE HxSA
1 4 1 1 1 0
2 3 1 2 0 1
3 0 0 1 0 0
4 3 2 1 1 1
5 4 1 1 1 0
etc.
ID
Vet_Rate= veteran rating of themselves (0-4)
Psych_Rate= psychiatrist rating of veteran(0-4)
MD_Type= type of psychiatrist (1=resident 2=staff)
MDE=presence of depression 0=no 1=yes
HxSA=history of suicide attempt 0=no 1=yes
Thanks for any help,
Jan ===================== To manage your subscription to SPSSX-L, send a
message to
[hidden email] (not to SPSSX-L), with no body text
except the command. To leave the list, send the command SIGNOFF SPSSX-L For
a list of commands to manage subscriptions, send the command INFO REFCARD
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD