comparing pre- post- percentages (proportions)

[Scott Roesch]

Hi all:

Can anyone provide syntax to statistically compare two percentages
(proportions) that come from a pre-post (repeated measures design). I
know that you can use a McNemar test via the Crosstabs procedure, but
I am looking for a more comparable analog to the paired-sample t-test.
Thanks!

Jordan

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

[Maguin, Eugene]

Scott,
I think the only option is Genlinmixed. I can offer no advice about a syntax setup because I have never used it and there are no repeated measures examples that I saw.

Gene Maguin


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Scott Roesch
Sent: Thursday, June 07, 2012 2:48 PM
To: [hidden email]
Subject: comparing pre- post- percentages (proportions)

Hi all:

Can anyone provide syntax to statistically compare two percentages
(proportions) that come from a pre-post (repeated measures design). I know that you can use a McNemar test via the Crosstabs procedure, but I am looking for a more comparable analog to the paired-sample t-test.
Thanks!

Jordan

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

[Ryan]

Jordan,

 

One option would be to fit a generalized estimating equations (GEE) log-binomial regression model via the GENLIN procedure. No time to provide details at this moment.

 

Ryan

On Thu, Jun 7, 2012 at 2:48 PM, Scott Roesch <[hidden email]> wrote:

Hi all:

Can anyone provide syntax to statistically compare two percentages
(proportions) that come from a pre-post (repeated measures design). I
know that you can use a McNemar test via the Crosstabs procedure, but
I am looking for a more comparable analog to the paired-sample t-test.
Thanks!

Jordan

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

[Rich Ulrich]

McNemar's is exactly an analog to the paired-sample t-test,
so far as I can see.  I can't imagine what else you have in mind.

Both tests focus on the changes.  McNemar's can be seen
to be a sign-test on the changes, plus versus minus.

The paired-sample t-test can also be computed as a one-
sample t-test on continuous "changes" as compared to zero. 
(In SAS, that used to be the prescribed method to compute it,
if I'm not misinformed.)

--
Rich Ulrich

> Date: Thu, 7 Jun 2012 11:48:25 -0700

> From: [hidden email]
> Subject: comparing pre- post- percentages (proportions)
> To: [hidden email]
>
> Hi all:
>
> Can anyone provide syntax to statistically compare two percentages
> (proportions) that come from a pre-post (repeated measures design). I
> know that you can use a McNemar test via the Crosstabs procedure, but
> I am looking for a more comparable analog to the paired-sample t-test.
> Thanks!

...

[Ryan]

Aside from McNemar's test, an option, Rich, would be to fit a repeated measures (a.k.a. GEE)  log-binomial or a GEE logistic regression model via the GENLIN procedure. The GENLIN procedure provides the option of testing the pre-post difference on the original (probability in this situation) scale using the EMMEANS subcommand.

 

The reason I recommended the GEE log-binomial model initially was because my mind immediate went to relative risks. But if the OP is only interested in testing the difference in proportions, as opposed to ratio of proportions, then in the context of GEE-generalized linear models, I'd recommend a GEE logistic regression model via GENLIN with the EMMEANS subcommand incorporated into the code.  

 

Ryan

On Thu, Jun 7, 2012 at 5:09 PM, Rich Ulrich <[hidden email]> wrote:

McNemar's is exactly an analog to the paired-sample t-test,
so far as I can see.  I can't imagine what else you have in mind.

Both tests focus on the changes.  McNemar's can be seen
to be a sign-test on the changes, plus versus minus.

The paired-sample t-test can also be computed as a one-
sample t-test on continuous "changes" as compared to zero. 
(In SAS, that used to be the prescribed method to compute it,
if I'm not misinformed.)

--
Rich Ulrich

> Date: Thu, 7 Jun 2012 11:48:25 -0700
> From: [hidden email]

> Subject: comparing pre- post- percentages (proportions)

> To: [hidden email]

>
> Hi all:
>
> Can anyone provide syntax to statistically compare two percentages
> (proportions) that come from a pre-post (repeated measures design). I
> know that you can use a McNemar test via the Crosstabs procedure, but
> I am looking for a more comparable analog to the paired-sample t-test.
> Thanks!

...