## Longitudinal mixed models with SPSS

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

fnc
Posts: 8
Joined: Wed Apr 23, 2008 12:01 am

### Longitudinal mixed models with SPSS

Hello,

hope someone could help me to solve this problem for which I can't see a way out.

I want to test the effect of three different kinds of therapy on different patients:

- patients corresponds to the first level unit of analysis
- patients are nested into groups (actually, 12 different groups); so, groups are the second level of analysis
- therapy has three levels: therapy A, therapy B, therapy C
- we measure by a test the level of anxiety at four different times i.e. beginning, after 30 days, after 60 days, after 90 days.

Each participant in a group receive the same type of treatment (namely, groups 1, 2,3,4 receive therapy A, groups 5,6,7,8 receive therapy B, 9,10,11,12 receive therapy C). This is necessary because these aregroup therapies.

My interest is to test

- if groups in condition A, B, C have the same values before beginning the therapy
- how the effect of each therapy changes according to time: my hyphtoeses are that therapy A is the most effective and stable along time, therapy B is effective but only after a prolonged period of time, while therapy C is not effective.

The syntax I am writing has something uncorrect, and I think I have used the wrong approach.

Has anyone any suggestion?

Thank you in advance for any kind of help, I am really desperate.
Labor
Posts: 35
Joined: Thu May 08, 2008 10:58 am
I'm at work, can't go further into it, but search for

Peugh & Enders, 2005, Using the SPSS Mixed Procedure to Fit Cross-Sectional and Longitudinal Multilevel Models, Educational and Psychological Measurement 65, p. 717

It's online at SAGE.

http://epm.sagepub.com/cgi/content/abstract/65/5/717
fnc
Posts: 8
Joined: Wed Apr 23, 2008 12:01 am

Thank you for the reference you provided me.

1) I have tried to develop this model... is it correct?

Code: Select all

``````MIXED
results  BY condition time group
/FIXED = condition time condition*time  | SSTYPE(3)
/METHOD = REML
/PRINT = COVB TESTCOV
/RANDOM INTERCEPT group  | COVTYPE(VC)
/REPEATED = time | SUBJECT(subject) COVTYPE(UN) ``````
where group is the specific terapeutic group, condition is the type of therapy and time is the four discrete moments in time when I take my measurements.

2) How can I make comparison among conditions in different time (e.g. comparison among patients between the three conditions at the end of the therapy)? Is there something like simple effects or contrast analysis I can use?

3) In order to understand better regression coefficients meaning, can I subtract from my data the mean of the value of the groups before the first treatment?
Labor
Posts: 35
Joined: Thu May 08, 2008 10:58 am
1/ I will check it out, but I'm not sure about the interaction between time and condition
2/ If it's a linear relationship, you can enter these variables at level 1, otherwise you will have to search for the right function or plot it graphically
3/ As far as I know, centering is mostly based on all data (grand mean) or for every level 2 unit (group mean), not on the mean at a certain moment. However, since it doesn't change your vector (unless in absolute terms, which indeed can make interpretation more logical), you can distract whatever you want.
fnc
Posts: 8
Joined: Wed Apr 23, 2008 12:01 am
Labor wrote:1/ I will check it out, but I'm not sure about the interaction between time and condition
It seems logical to me, but I am not sure about that...
I think that this interaction should be inserted insiede the model at a certain point, buit I am not sure if that is the point.
2/ If it's a linear relationship, you can enter these variables at level 1, otherwise you will have to search for the right function or plot it graphically
Sorry, I did not understand. The relationship is not linear (in the sense that at a certain condition it first increase and then, at the last times, decreases). Maybe plotting is the most correct solution (i do not have very precise assumptions on which kind of trend results could have, since it is mainly an exploratory study).
3/ As far as I know, centering is mostly based on all data (grand mean) or for every level 2 unit (group mean), not on the mean at a certain moment. However, since it doesn't change your vector (unless in absolute terms, which indeed can make interpretation more logical), you can distract whatever you want.
That is what I was thinking; I know that this approach is "strange", but from a logical point of view it seems to be the correct ones (since I can assume that mean at time 1 is the general level of depandent variable for a certain population... mean across groups are indeed very similar, at time 1).
fnc
Posts: 8
Joined: Wed Apr 23, 2008 12:01 am
No one has any idea?
Labor
Posts: 35
Joined: Thu May 08, 2008 10:58 am
Hi, I'm sorry I've let this thing go, I'm a bit busy right now. Perhaps in a few days ...

### Who is online

Users browsing this forum: No registered users and 1 guest