Statistical Q: Comparing group means of the same people: which test of significance in SPSS?

[Walter Funk]

Since my 1st try last Friday didn't appear on the list, here is my 2nd
try. Sorry if anybody shout get this request twice.
Walter

----------------------------------------------------------------------------

Dear SPSS-community,

we have data from novice drivers out of up to four consecutive surveys
at the beginning of their driving career. We build some indices for each
sampling wave, each index consisting out of a couple of variables.
Assume the indices are metric and show approx. normal distribution. Now
we want

- to compare the mean of the indices between wave 1 and 2, wave 2 and 3,
wave 3 and 4. Here we look at the mean of the sample as a whole or at
the mean of groups (males, females). As far as I understand, looking at
4 repeated cross sections is a trend perspective.

- to compare the individual mean of every single respondent between wave
1 and 2, wave 2 and 3, wave 3 and 4. As far as I understand, this is a
panel perspective.

To compare the individual means between two waves in the panel
perspective we use t-test for paired data.

But what can we do in the trend perspective? The t-test for independent
samples compares group means. The different waves are the groups to be
compared. Would using a t-test for independent samples violate the
assumption of independency, since the same persons are in both waves
that we want to compare? Which alternative test to compare the
significance of mean differences between groups consisting out of the
same people would be appropriate?

Thanks a lot for your help.

Regards
Walter
--
**********************************************************
Dr. Walter Funk
Institut fuer Freie Berufe
an der Friedrich-Alexander-Universitaet Erlangen-Nuernberg
Marienstrasse 2

D-90402 Nuernberg
Germany

Tel: ++49 - (0)911 - 23 565-32
Fax: ++49 - (0)911 - 23 565-50
E-Mail: [hidden email]
http://www.ifb.uni-erlangen.de

Amtsgericht Nürnberg VR 841
**********************************************************

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

Walter,

I think that some or all of your analyses could be combined in to single
analysis step through a repeated measures ANOVA, accessible through the GLM
command. However, I think it would be useful to do some background reading
on the analysis of data with repeated measures. Any intermediate ANOVA stats
text should give this background.

Gene Maguin


>>we have data from novice drivers out of up to four consecutive surveys
at the beginning of their driving career. We build some indices for each
sampling wave, each index consisting out of a couple of variables.
Assume the indices are metric and show approx. normal distribution. Now
we want

- to compare the mean of the indices between wave 1 and 2, wave 2 and 3,
wave 3 and 4. Here we look at the mean of the sample as a whole or at
the mean of groups (males, females). As far as I understand, looking at
4 repeated cross sections is a trend perspective.

- to compare the individual mean of every single respondent between wave
1 and 2, wave 2 and 3, wave 3 and 4. As far as I understand, this is a
panel perspective.

To compare the individual means between two waves in the panel
perspective we use t-test for paired data.

But what can we do in the trend perspective? The t-test for independent
samples compares group means. The different waves are the groups to be
compared. Would using a t-test for independent samples violate the
assumption of independency, since the same persons are in both waves
that we want to compare? Which alternative test to compare the
significance of mean differences between groups consisting out of the
same people would be appropriate?

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