Longitudinal analysis for beginners: Random Effects Modelling?

[Jennifer Thompson]

Dear SPSSers,

I'm hoping that someone might be able to comment or advise on some
longitudinal analysis I'm undertaking on some clinical measures of motor
disability.  I've made some headway but would greatly appreciate other
suggestions or feedback.

The measures are two 'novel' measures of motor dysfunction (both of which
are timed tasks, with a score in seconds) and, as a comparison, a standard
clinical motor scale, which is the sum of various individual items.

My dataset consists of 918 seperate observations of 213 patients, seen on
average 4.31 times (but varying between 2 and 10+).  The dataset is 'long'
rather than 'wide'.  The mean duration between assessments was 1.25 years.
At the moment, I would simply like to know what the average rate of change
is, per year, for these clinical measures (although a future aim might be to
see if this rate of change varies according to stage of illness).

So far, I have calculated the time between each assessment  [using LAG] and
converted this from seconds to years.  I used the same method for
calculating the difference in scores on the motor scales between the
assessments.  Next, I divided the motor difference scores by 'time in years'
to come up with 'change per year' for each clinical measure.  I then worked
out the mean change per year for each patient, and then finally calculated a
group mean from this.

So, not a very sophisticated analysis and I apologise if I am making the
eyes of statisticians bleed.  Does anyone have any suggestions of a better,
cleaner way to approach this dataset?  A few papers I've looked at mention
Random Effects Modelling, but they give little detail on the procedure and
most seem to have used an in-house analysis program.

Many thanks for reading,

Jennifer

[peter link]

Jennifer -

I might suggest looking into the MIXED procedure in SPSS.  I would recommend
a book to look at, also.

Singer & Willett, "Applied Longitudinal Data Analysis", Oxford University
Press, 2003.

There is a lot of literature out about this type of modelling.  It goes by a
number of different name - mixed models, multilevel models, to name two.  In
short, this approach better handles repeated measures within individuals,
than a linear regression approach does (incorporates the natural correlation
of repeated observations within individuals).  It also handles missing data
better than say a repeated measures ANOVA.

Peter Link
VA San Diego Healthcare System



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]]On Behalf Of
Jennifer Thompson
Sent: Monday, June 25, 2007 4:29 AM
To: [hidden email]
Subject: Longitudinal analysis for beginners: Random Effects Modelling?


Dear SPSSers,

I'm hoping that someone might be able to comment or advise on some
longitudinal analysis I'm undertaking on some clinical measures of motor
disability.  I've made some headway but would greatly appreciate other
suggestions or feedback.

The measures are two 'novel' measures of motor dysfunction (both of which
are timed tasks, with a score in seconds) and, as a comparison, a standard
clinical motor scale, which is the sum of various individual items.

My dataset consists of 918 seperate observations of 213 patients, seen on
average 4.31 times (but varying between 2 and 10+).  The dataset is 'long'
rather than 'wide'.  The mean duration between assessments was 1.25 years.
At the moment, I would simply like to know what the average rate of change
is, per year, for these clinical measures (although a future aim might be to
see if this rate of change varies according to stage of illness).

So far, I have calculated the time between each assessment  [using LAG] and
converted this from seconds to years.  I used the same method for
calculating the difference in scores on the motor scales between the
assessments.  Next, I divided the motor difference scores by 'time in years'
to come up with 'change per year' for each clinical measure.  I then worked
out the mean change per year for each patient, and then finally calculated a
group mean from this.

So, not a very sophisticated analysis and I apologise if I am making the
eyes of statisticians bleed.  Does anyone have any suggestions of a better,
cleaner way to approach this dataset?  A few papers I've looked at mention
Random Effects Modelling, but they give little detail on the procedure and
most seem to have used an in-house analysis program.

Many thanks for reading,

Jennifer

[David Greenberg]

This appears to be well-suited for a growth curve modeling approach. David Greenberg, Sociology Department, New York University

----- Original Message -----
From: Jennifer Thompson <[hidden email]>
Date: Monday, June 25, 2007 7:29 am
Subject: Longitudinal analysis for beginners: Random Effects Modelling?
To: [hidden email]

> Dear SPSSers,
>
> I'm hoping that someone might be able to comment or advise on some
> longitudinal analysis I'm undertaking on some clinical measures of motor
> disability.  I've made some headway but would greatly appreciate other
> suggestions or feedback.
>
> The measures are two 'novel' measures of motor dysfunction (both of which
> are timed tasks, with a score in seconds) and, as a comparison, a standard
> clinical motor scale, which is the sum of various individual items.
>
> My dataset consists of 918 seperate observations of 213 patients, seen
> on
> average 4.31 times (but varying between 2 and 10+).  The dataset is 'long'
> rather than 'wide'.  The mean duration between assessments was 1.25 years.
> At the moment, I would simply like to know what the average rate of change
> is, per year, for these clinical measures (although a future aim might
> be to
> see if this rate of change varies according to stage of illness).
>
> So far, I have calculated the time between each assessment  [using
> LAG] and
> converted this from seconds to years.  I used the same method for
> calculating the difference in scores on the motor scales between the
> assessments.  Next, I divided the motor difference scores by 'time in
> years'
> to come up with 'change per year' for each clinical measure.  I then worked
> out the mean change per year for each patient, and then finally
> calculated a
> group mean from this.
>
> So, not a very sophisticated analysis and I apologise if I am making the
> eyes of statisticians bleed.  Does anyone have any suggestions of a better,
> cleaner way to approach this dataset?  A few papers I've looked at mention
> Random Effects Modelling, but they give little detail on the procedure
> and
> most seem to have used an in-house analysis program.
>
> Many thanks for reading,
>
> Jennifer