Help with GLM repeated measures ANOVA

[Talk Stats]

Hi!

Here is data issue in a nutshell. I am examining whether the birth of a new
infant affects the average affiliation rate between adult female and male
monkeys. After all the data is worked out the table looks like this:

female before after
fl #### ###
rm #### ####
mv #### ####
ll ### ###
mx ### ###

In order to determine whether there is a significant difference in the
average daily affiliation rate before and after the birth of an infant I
have been running a repeated measures ANOVA in SPSS. I name my factore bfaf
(to stand for before and after) and give it two levels. The two levels are
then defined as the before and after column values. I am not getting a
significant result in the output for within-subjects, but there is a
signifcant result between-subjects. I'm not quite sure what this means, but
I think it is that some of the females show a significant change and some
don't. What is a post hoc test that I can run to determine which females
are showing a significant change and which aren't? Am I setting up the test
correctly or should the data be displayed in a different way?

Any and all help will be greatly appreciated! Thank you so much!

____
This question was originally posted at Talk Stats Forums
http://www.talkstats.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

[Maguin, Eugene]

Svetlana,

I think you have two general models to consider. One is a growth curve
model. You'll need to create a set of four contrast variables to represent
the cohort. Then you will regress your growth curve components (i.e.,
intercept, slope, and, maybe, curve) onto these contrast vars, along with
any other demographic or other covariates.

The other model is a repeated measures model. You'll need to use a multiple
group model to get at between group differences in means.

The main problem with the growth curve model is how well you can model the
services trajectory. It may be that this will fail and you'll be forced back
to a repeated measures model. With the repeated measures model, you are
going to need to evaluate and maintain measurement equivalence between
groups. There have been some papers on this. Overall, this is going to be a
hard project because of non-normality problem.

I don't know whether you subscribe to either the multilevel or semnet lists
but there are elements of your analysis that might well fit better on those
lists, as my strong impression is that there is no too much overlap between
those lists and the spss list. Also, and since you are using Mplus, the
mplus archives and Linda Muthen are great resources.

Gene Maguin

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

I don't get where the between factor comes from. It looks to me like have a
one within factor with two levels (before and after) analysis. This reduces
to a paired t-test. What am I missing in your presentation?

Gene Maguin

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

[Marta Garcia-Granero]

Talk Stats escribió:

> Here is data issue in a nutshell. I am examining whether the birth of a new
> infant affects the average affiliation rate between adult female and male
> monkeys. After all the data is worked out the table looks like this:
>
> female before after
> fl #### ###
> rm #### ####
> mv #### ####
> ll ### ###
> mx ### ###
>
> In order to determine whether there is a significant difference in the
> average daily affiliation rate before and after the birth of an infant I
> have been running a repeated measures ANOVA in SPSS.

You have been given advise that a paired t-test would be more than
enough. I agree.

> I name my factore bfaf
> (to stand for before and after) and give it two levels. The two levels are
> then defined as the before and after column values. I am not getting a
> significant result in the output for within-subjects, but there is a
> signifcant result between-subjects. I'm not quite sure what this means, but
> I think it is that some of the females show a significant change and some
> don't.
No, your interpretation is wrong. Between subjects significance only
implies that the different females had in average different affiliation
rates, even before.

> What is a post hoc test that I can run to determine which females
> are showing a significant change and which aren't?

This is data fishing, avoid it. If the result is not significant, it
isn't. Period. Add  a95% confidence interval for the mean difference
between-after (it's standard output for paired t-test). Add  a line
graph showing the individual changes. You don't mention the sample size,
if it is too small then you are lacking power to detect even important
differences.

See this worked example.

* Sample dataset *.
DATA LIST LIST/before after (2 F8.1).
BEGIN DATA
17.4 18.0
18.9 10.0
15.2 11.2
13.3 13.4
20.0 20.2
14.0  9.2
19.9 18.5
19.1 14.1
13.0 10.2
14.8 11.8
13.9  7.8
14.8  6.9
END DATA.
VAR LABEL before 'pIIIp levels before IFN treatment'/
           after 'pIIIp levels after IFN treatment'.

T-TEST
  PAIRS = before  WITH after
  /CRITERIA = CI(.95)
  /MISSING = ANALYSIS.

* Graphing data *.
VARSTOCASES /ID = id
 /MAKE pIIIp 'pIIIp levels' FROM before after
 /INDEX = Time(2).

GRAPH /LINE(MULTIPLE)MEAN(pIIIp) BY Time BY id .


Regards,
Marta García-Granero


--
For miscellaneous statistical stuff, visit:
http://gjyp.nl/marta/

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