group by time test for variability?

[Harris, Betty A.]

Hi,

I'm looking at student early literacy scores over five time period for two groups using a repeated measures anova. That's situated.

I'm also running independent t-tests at each time period comparing average scores by group.  The t-test also gives me whether the variability in the scores is significantly different between the groups at each time period. This will allow me to talk about the spread of scores between the groups. e.g. I would expect the intervention group to have a sd that gets progressive smaller over time on several of the subtests.

The question is.... Is there are way to do a group by time test on the standard deviations to see if the intervention group's variability decreases more over time?

Thanks,
Betty Harris
University of Oklahoma E-TEAM
http://eteam.ou.edu/

[Ryan]

Betty,

Specify a linear mixed model (via the MIXED procedure) that allows for
group-BY-time heterogeneous residual variances. Then review the output
for the R matrix to see whether or not your hypothesis holds up. You
could also compare the overall fit of this model to more restrictive
models.

Ryan

On Thu, Aug 25, 2011 at 12:19 PM, Harris, Betty A. <[hidden email]> wrote:

> Hi,
>
> I'm looking at student early literacy scores over five time period for two
> groups using a repeated measures anova. That's situated.
>
> I'm also running independent t-tests at each time period comparing average
> scores by group.  The t-test also gives me whether the variability in the
> scores is significantly different between the groups at each time period.
> This will allow me to talk about the spread of scores between the groups.
> e.g. I would expect the intervention group to have a sd that gets
> progressive smaller over time on several of the subtests.
>
> The question is.... Is there are way to do a group by time test on the
> standard deviations to see if the intervention group's variability decreases
> more over time?
>
> Thanks,
> Betty Harris
> University of Oklahoma E-TEAM
> http://eteam.ou.edu/
>

=====================
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]

If I measured something like "literacy scores" and expected the improvement
to produce a reduced standard deviation, then I would immediately guess that
the given score is *likely* to be the wrong metric for measuring outcome.

In particular, I would consider, What happens when I subtract, to get a
score from 0 on up of Errors on the test; and then, assuming Poisson as
the underlying distribution, I take the square root.  Would this yield
variances that are much more constant?  - That has worked for me, a
time or three in the past.

These days, it might be more conventional that, after reversing
scores so that Errors are counted (minimum, zero), someone would
use a Poisson link function. 

--
Rich Ulrich

---

Date: Thu, 25 Aug 2011 16:19:27 +0000
From: [hidden email]
Subject: group by time test for variability?
To: [hidden email]

Hi,

I'm looking at student early literacy scores over five time period for two groups using a repeated measures anova. That's situated.

I'm also running independent t-tests at each time period comparing average scores by group.  The t-test also gives me whether the variability in the scores is significantly different between the groups at each time period. This will allow me to talk about the spread of scores between the groups. e.g. I would expect the intervention group to have a sd that gets progressive smaller over time on several of the subtests.

The question is.... Is there are way to do a group by time test on the standard deviations to see if the intervention group's variability decreases more over time?