SPSSX Discussion

longitudinal data

Classic

List

Threaded

10 messages Options

janm

longitudinal data

Is this a good longitudinal model for two data points (two waves)?

MIXED achievement WITH time
/PRINT = SOLUTION TESTCOV
/METHOD = REML
/FIXED = intercept time
/RANDOM = intercept | SUBJECT(idstudent)
/RANDOM = intercept | SUBJECT(idclass)
/REPEATED = wave | SUBJECT(idstudent*idclass).

Ryan

Re: longitudinal data

No it isn't. Please share the study design, data structure and research question(s) to receive specific feedback.

Ryan

On Sat, Jan 24, 2015 at 1:09 PM, janm <[hidden email]> wrote:

Is this a good longitudinal model for two data points (two waves)?

MIXED achievement WITH time
/PRINT = SOLUTION TESTCOV
/METHOD = REML
/FIXED = intercept time
/RANDOM = intercept | SUBJECT(idstudent)
/RANDOM = intercept | SUBJECT(idclass)
/REPEATED = wave | SUBJECT(idstudent*idclass).

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/longitudinal-data-tp5728473.html
Sent from the SPSSX Discussion mailing list archive at Nabble.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

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

janm

Re: longitudinal data

I have two waves of measurements of the same respondents. I have three levels: time, student and class. I have student and class time invariant covariates.

Why is my syntax wrong?

Maguin, Eugene

Re: longitudinal data

Your current syntax is

MIXED achievement WITH time
/PRINT = SOLUTION TESTCOV
/METHOD = REML
/FIXED = intercept time
/RANDOM = intercept | SUBJECT(idstudent)
/RANDOM = intercept | SUBJECT(idclass)
/REPEATED = wave | SUBJECT(idstudent*idclass).

You can have a three level model: time within students within class or you can have a two level model with time as a repeated factor and the repeated factor nested within class. If you go the three level model route you need to have

MIXED achievement WITH time
/PRINT = SOLUTION TESTCOV
/METHOD = REML
/FIXED = time
/RANDOM = intercept | SUBJECT(idstudent) covtype=id
/RANDOM = intercept | SUBJECT(idclass*idstudent) covtype=id.

I think this is the correct model statement for at two level with time as a repeated factor. But I absolutely may be wrong.

MIXED achievement by time
/PRINT = SOLUTION TESTCOV
/METHOD = REML
/FIXED = time
/RANDOM = intercept | SUBJECT(idclass) covtype(id)
/REPEATED = wave | SUBJECT(idstudent*idclass) covtype(cs).

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of janm
Sent: Tuesday, January 27, 2015 8:55 AM
To: [hidden email]
Subject: Re: longitudinal data

I have two waves of measurements of the same respondents. I have three
levels: time, student and class. I have student and class time invariant covariates.

Why is my syntax wrong?

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/longitudinal-data-tp5728473p5728496.html
Sent from the SPSSX Discussion mailing list archive at Nabble.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

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

Re: longitudinal data

In reply to this post by janm

I still have not received the info I requested but I will try to respond as best I can with what you have given me.

I assume that students are nested in classes. This means that students cannot jump from one class to the next during this study. Certainly, students nested in a class can induce correlation (within-class correlation).

Students are measured repeatedly and these repeated measurements can also induce correlation (within-student correlation).

To account for both sources of dependence in your data, you could incorporate a RANDOM statement to account for within-class correlation and a REPEATED statement to account for the temporal correlation.

MIXED <dependent variable> BY <time variable>
/PRINT = SOLUTION TESTCOV
/METHOD = REML
/FIXED = <time variable>
/RANDOM = intercept | SUBJECT(<classroom variable>)
/REPEATED = <time variable> | SUBJECT(<student variable>) COVTYPE(CS).

I've made all sorts of assumptions that I do not have time to elaborate, but I wanted to give you something to chew on. Write back if you have any *specific* questions.

By the way, it still isn't really clear to me what your research question is! If you are interested in determining whether there is a significant mean change from time 1 to time 2 on the dependent variable after account for the dependencies in your data noted above, then this is accomplished by the syntax above.

I assume that student IDs do not repeat across classes. If they do, then you will need to incorporate the student-by-classroom interaction as the subject identifier on the REPEATED statement.

(There are other parameterizations that are equivalent to the model above, but there's no need to spend time pointing them out right now.)

Ryan

On Tue, Jan 27, 2015 at 8:55 AM, janm <[hidden email]> wrote:

I have two waves of measurements of the same respondents. I have three
levels: time, student and class. I have student and class time invariant
covariates.

Why is my syntax wrong?

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/longitudinal-data-tp5728473p5728496.html

Sent from the SPSSX Discussion mailing list archive at Nabble.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

janm

Re: longitudinal data

Thank you!

the last syntax gives an error:

At least one random subject variable is not contained in the list of repeated subject variables.

janm

Re: longitudinal data

To answer your questions (thank you for your answer):

Yes, students are nested in classes and stick to one class. They were measured two times so far.
My research question focuses on whether different conditions from wave one alter the growth of achievement. I want to take into account that students were tested in a longitudinal setting.
I think that a random intercept is the only possible model (no random slope) because I only have two waves so far.

Maguin, Eugene

Re: longitudinal data

In reply to this post by janm

Please include the syntax that gave an error. Also, if you are using the model that includes the Repeated subcommand, use the syntax that Ryan gave you and not what I gave.

You could have a random slope with only two measures. Doing so means that you will perfectly model the intercept (T1 score) and the T1-T2 change for each person and, therefore, the residual will be 0.0 for each person, meaning the level 1 residual variance will be 0.00. I haven't done this in a multilevel model so I don't know this will work out in the significance tests for those coefficients.

Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of janm
Sent: Tuesday, January 27, 2015 2:14 PM
To: [hidden email]
Subject: Re: longitudinal data

Thank you!

the last syntax gives an error:

At least one random subject variable is not contained in the list of repeated subject variables.

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/longitudinal-data-tp5728473p5728503.html
Sent from the SPSSX Discussion mailing list archive at Nabble.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

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

janm

Re: longitudinal data

This post was updated on .

CONTENTS DELETED

The author has deleted this message.

janm

Re: longitudinal data

It is the syntax of Ryan that is troublesome.
Modeling two waves with a random slope is impossible because the Hessian matrix becomes negative (in spss and mplus).