Hi,
I have survey data consisting of dichotomized answers to 19
questions (suicide warning signs) and the dichotomous results from 3
standardized tools for major depression (MDE) , PTSD, and alcohol
misuse which I have made into one variable called clinical
characteristics.
Because of small expected cell sizes I have collapsed the original
combinations of clinical characteristics into 4 categories. This
solved the cell size problem.
The overall chi-square is .000 for each of the suicide warning
signs. I want to do a post hoc comparison similar to that done with
ANOVA.
I have a crosstabs with suicide warning signs questions as the row
variables and the 4 categories of clinical characteristics as the
column variable (none, MDE only, PTSD only, and MDE and PTSD)
It seems that partitioning the overall chi-square as previously
suggested by Bruce Weaver (orthogonal contrasts and maximum
likelihood ratio test) is the best method.
QUESTION: In looking at the table I can make several observations
based on the frequencies but I read this caution:
Examination of percentages in the contingency table and expected
frequency table can be misleading. The residual, or the
difference, between the observed frequency and the expected
frequency is a more reliable indicator
so I think I should be looking at the standardized residuals.
However, I'm confused as to how to understand them in a way as to
decide which subtables I should test using Bruce's explanation.
For example, for active suicidal ideation, the frequencies are none
10%, MDE only 53%, PTSD only 25%, and MDE ples PTSD 52%. The n's are
different for the groups (145, 199, 65, and 73 respectively).
The standardized residuals are none -5.3, MDE only 3.5, PTSD only
-1.5, and MDE plus PTSD 3.1.
I think the residuals for all but PTSD only tell me that they make a
significant contribution. What I think I want to know is whether
there is a sig difference between the two MDE categories and whether
there is a difference between those combined vs. PTSD only.
Considering that I would need to present the results in some
meaningful way, would I decide on one partitioning strategy for all
the suicide warning signs variables even though a few show different
residual patterns?
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
Locked
