Re: Question about CrossTab Column proportions statistics

[Maguin, Eugene]

Crosstabs is good for making a table but why not use logistic regression?

By the way, do cases represent the response of unique persons to only one of the three medications?

Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of devoidx
Sent: Wednesday, December 09, 2015 10:44 AM
To: [hidden email]
Subject: Question about CrossTab Column proportions statistics

Hi all, sorry this is long but hopefully an easy read. So in a simplified way, I have two categorical variables "Medication" and "sideeffect".

"Medication" has 3 values MedA MedB and MedC. "SideeffectX" has 2 values Yes and No. My goal was to see which medication is associated with higher rate of sideeffectX (so which medication had a higher proportion of "YES") and whether its significantly different  (P<0.05) from the other 2 medications.

First i ran a simple Chi square and saw that the two variables have an association. Then i wanted to see which medications specifically were different so i ran the Z-test (compare column proportions) in the crosstab.
All three medications came up with a different subscript meaning they're all statistically different from each other with regards to having "YES"
sideeffectX (P<0.05 among all three groups of meds).

Then i did it another way, I excluded MedA cases (so now "Medication" is a binary variable) and ran a Chi Square just to compare MedB and MedC together with regards to having the "YES" sideffectX and the Pvalue is now 0.08 which is non-significant.

Which one is it? in column proportions, the sideeffect difference between MedB and MedC was P<0.05, in Chi square, its P>0.05...I can't figure out why.




--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Question-about-CrossTab-Column-proportions-statistics-tp5731101.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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

[devoidx]

Thanks, i think you replied too quickly lol because i deleted this post shortly after i posted it. I realized the error was on my part.

but as a general comment the problem with logistic regression i have in these scenarios is that it would assign one of the medications as reference and compare the other two with it. so i have to run it a number of times each time assigning a different medication as reference to be able to get all 3 interaction. It's a hassle especially when my categorical variable has 5-6 values.

[Maguin, Eugene]

Look the GenLin procedure. It will do logistic regression and compare outcomes using emmeans.
Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of devoidx
Sent: Wednesday, December 09, 2015 11:43 AM
To: [hidden email]
Subject: Re: Question about CrossTab Column proportions statistics

Thanks, i think you replied too quickly lol because i deleted this post shortly after i posted it. I realized the error was on my part.

but as a general comment the problem with logistic regression i have in these scenarios is that it would assign one of the medications as reference and compare the other two with it. so i have to run it a number of times each time assigning a different medication as reference to be able to get all 3 interaction. It's a hassle especially when my categorical variable has
5-6 values.



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Re-Question-about-CrossTab-Column-proportions-statistics-tp5731102p5731105.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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

[devoidx]

Interesting, I'll look into that, thanks for the tip