saving individual regression coefficients

[Scott Roesch]

Hello all:

I have a data set where I have repeated measures per participant. At the
partipant level I would like to run a regression using time as a predictor
of the outcome of interest (e.g., generate the slope or individual growth
curve) that I would then like to use in other analyses. This would require
me to do a number of individual-level regressions and save the regression
coefficients to the data file. I have over 800 participants and I was
hoping someone had already automated a procedure to do this. Any ideas
would be greatly appreciated.

Scott

[Swank, Paul R]

You are better off doing that as a mixed model where you can model
predictors of the growth parameters that take into account the
estimation of those growth parameters.

Paul R. Swank, Ph.D. Professor
Director of Reseach
Children's Learning Institute
University of Texas Health Science Center-Houston


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Scott Roesch
Sent: Wednesday, September 19, 2007 4:12 PM
To: [hidden email]
Subject: saving individual regression coefficients

Hello all:

I have a data set where I have repeated measures per participant. At the
partipant level I would like to run a regression using time as a
predictor of the outcome of interest (e.g., generate the slope or
individual growth
curve) that I would then like to use in other analyses. This would
require me to do a number of individual-level regressions and save the
regression coefficients to the data file. I have over 800 participants
and I was hoping someone had already automated a procedure to do this.
Any ideas would be greatly appreciated.

Scott

[Maguin, Eugene]

I'd like to make a limited defense of what Scott seems to be wanting to do.
This idea has been presented in Singer and Willett's book and I'd bet in
Bollen and Curran's book also because they were demonstrating it at the
growth curve modeling workshop they did several years ago for ICPSR. It is
useful only for visualizing and screening the slope and intercept values
(assuming a linear curve). The standard errors are incorrect. Using the
computed values as input for a level 2 regression also has, I think,
incorrect standard errors. The actual analyses should be done, as Paul says,
in a multilevel model. I think that Mplus can do both steps as Mplus has
plotting routines capable of plotting individual growth curves. I don't
think, however, that it can show a distribution of slope and intercept
values. I've never looked for that capability because I've done the initial
analysis in spss.

With respect to using spss for this. The most direct way is to arrange the
data in a 'long' format, split the file by ID, do the regressions, and use
OMS to get the data out. I can't advise you on this. Others can.

When I've done this, I just wrote a little bit of syntax to compute the
slope and intercecpt directly. However, this is a custom job in that it was
written (and has to be) for the specific arrangement of assessments.

Gene Maguin

[ViAnn Beadle]

This description indicates to me that assuming Gene's long format, a
individual growth curve plot could easily be done directly with GPL. Here's
a simple example (two individuals, 10 observations using GPL:

data list list / id (F1.0) value (F1.0) timepoint (F1.0).
begin data
1 5 1
1 6 2
1 5 3
1 7 4
1 8 5
1 8 6
1 9 7
1 8 8
1 9 9
1 10 10
2 3 1
2 6 2
2 4 3
2 6 4
2 7 5
2 6 6
2 8 7
2 7 8
2 9 9
2 10 10
end data.
GGRAPH
  /GRAPHDATASET NAME="graphdataset" VARIABLES=timepoint value id
  MISSING=LISTWISE REPORTMISSING=NO
  /GRAPHSPEC SOURCE=INLINE.
BEGIN GPL
 SOURCE: s=userSource(id("graphdataset"))
 DATA: timepoint=col(source(s), name("timepoint"))
 DATA: value=col(source(s), name("value"))
 DATA: id=col(source(s), name("id"), unit.category())
 GUIDE: axis(dim(1), label("timepoint"))
 GUIDE: axis(dim(2), label("value"))
 ELEMENT: line(position(smooth.linear(timepoint*value)), split(id))
END GPL.

This example runs in Version 15. Note that you can also try some other fits
using GPL.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Gene Maguin
Sent: Thursday, September 20, 2007 7:47 AM
To: [hidden email]
Subject: Re: saving individual regression coefficients

I'd like to make a limited defense of what Scott seems to be wanting to do.
This idea has been presented in Singer and Willett's book and I'd bet in
Bollen and Curran's book also because they were demonstrating it at the
growth curve modeling workshop they did several years ago for ICPSR. It is
useful only for visualizing and screening the slope and intercept values
(assuming a linear curve). The standard errors are incorrect. Using the
computed values as input for a level 2 regression also has, I think,
incorrect standard errors. The actual analyses should be done, as Paul says,
in a multilevel model. I think that Mplus can do both steps as Mplus has
plotting routines capable of plotting individual growth curves. I don't
think, however, that it can show a distribution of slope and intercept
values. I've never looked for that capability because I've done the initial
analysis in spss.

With respect to using spss for this. The most direct way is to arrange the
data in a 'long' format, split the file by ID, do the regressions, and use
OMS to get the data out. I can't advise you on this. Others can.

When I've done this, I just wrote a little bit of syntax to compute the
slope and intercecpt directly. However, this is a custom job in that it was
written (and has to be) for the specific arrangement of assessments.

Gene Maguin