Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: bivariate correlation analsis for longitudinal data


From   "Michael N. Mitchell" <Michael.Norman.Mitchell@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: bivariate correlation analsis for longitudinal data
Date   Mon, 24 May 2010 09:38:04 -0700

Dear Junqing

My apologies for misreading your first email... I only saw the term "bivariate" but overlooked that you used the term "correlation". My standard script kicked in about the steps for doing bivariate regressions as a first step for a random effects model. I have not given much thought to looking at correlations in the context that you describe beyond what you have already done. Perhaps someone else on the list has more experience with this idea than myself.

Thanks for following up and gently noting my misunderstanding!

Michael N. Mitchell
Data Management Using Stata      - http://www.stata.com/bookstore/dmus.html
A Visual Guide to Stata Graphics - http://www.stata.com/bookstore/vgsg.html
Stata tidbit of the week         - http://www.MichaelNormanMitchell.com



On 2010-05-24 8.52 AM, jl591164@albany.edu wrote:
Thanks, Michael. I wonder there is a way to get the generalized Person
between-variable and cross-time correlation coefficients without fitting
regression models.

The regression will give the regression coficient. But I am not sure how
the regression will give the person bivaiare correltaions at each time
point.

Junqing

    >  Greetings Junqing

   If you have not done so already, you probably want to reshape your data
into a "long" format (e.g., see
http://www.ats.ucla.edu/stat/stata/modules/reshapel.htm).

   Having the data in a long format, you can then look at bivariate
relationships between a continous predictor and continuous outcome
like this (using the nlswork dataset as an example).

. webuse nlswork
. xtmixed ln_wage tenure || idcode:

   The above looks at "tenure" predicting "ln_wage", with no random
effects at level 2. If you are interested in assessing the variability
of "tenure" as a random slope, you can do something like this...

. xtmixed ln_wage tenure || idcode: tenure, cov(un)

   As I mentioned in a previous post, I cannot recommend the following
book highly enough for learning and as a reference for longitudinal
analysis...

   Applied Longitudinal Data Analysis: Modeling Change and Event Occurrence
by Judith D. Singer&  John B. Willett, see
http://gseacademic.harvard.edu/alda/

I hope this helps,

Michael N. Mitchell
Data Management Using Stata      -
http://www.stata.com/bookstore/dmus.html
A Visual Guide to Stata Graphics -
http://www.stata.com/bookstore/vgsg.html
Stata tidbit of the week         - http://www.MichaelNormanMitchell.com


On Sun, May 23, 2010 at 11:07 AM,<jl591164@albany.edu>  wrote:
I try to do bivairate correlation analysis for longitudinal data. My
understanding is to use the data in wide format and then do regular
correlation of the x and y vairables, for example correlation of x at
time
1 and y at time 2, x at time 2 and y at time 3, etc., or x and y both at
the same time point. By doing so, do i assume that time effect is the
name
on the correlations at different time point? What is the appropriate way
of conducting bivairate correlation analysis for longitudinal data?

Eventually, i will fit randome intercept models for y on Xs. Thanks a
lot.

Junqing


Dear David

    I think that it might help if we were able to see a picture of the
graph that you have
in mind. Of course, sharing such a picture is not easy via Statalist,
but
you could draw
something and share it using Jing, see

http://www.michaelnormanmitchell.com/stow/capturing-and-sharing-screen-images.html

    I think that the growth model that you have in mind does not involve
any of the
interaction effects (note that age and age squared are not interacted
with
anything) thus
when you look at the trajectory over time, the interactions you refer
to
just get
"averaged out". I think it is possible that you might actually want
more
than one graph,
for example you might like...

    1) a graph with age on the X axis, the predicted outcome on the y
axis,
and separate
lines for maternal IQ groups/treatment groups. This shows the growth
curve
over time for
the six groups. Note that the growth curves will be parallel.

    2) a graph with maternal IQ class on the X axis, separate lines for
treatment groups.
This will focus on the interaction.

    There may be multiple ways to touch the elephant, but I don't think
there is one way to
touch the elephant to get the entire picture. It is also possible that
you
might be
interested in interacting age (and perhaps age squared) with some of
your
other predictors
if you think that the growth trajectory depends on these factors.

    Finally, I would recommend "Applied Longitudinal Data Analysis:
Modeling Change and
Event Occurrence" by Judith D. Singer and John B. Willett to anyone who
is
interested in
these kinds of models. It does not solve this exact problem, but is an
outstanding
reference, see

http://www.ats.ucla.edu/stat/examples/alda.htm

I hope this helps,

Michael N. Mitchell
Data Management Using Stata      -
http://www.stata.com/bookstore/dmus.html
A Visual Guide to Stata Graphics -
http://www.stata.com/bookstore/vgsg.html
Stata tidbit of the week         - http://www.MichaelNormanMitchell.com



On 2010-05-22 5.52 PM, David Torres wrote:
Hello,

I actually came upon the link Nick provided in a previous post
(http://www.ats.ucla.edu/stat/stata/faq/mar_graph/margins_graph.htm)
while I was working on this earlier today. The graph I was able to
produce does seem useful in that it shows the difference in slope
between child IQ scores of the different maternal IQ classes for both
the control and treatment groups. I wonder, though, if I'm missing
something by not showing growth. ???

Since the data include assessments at several time points (8 to be
exact), should I not want a graph of the adjusted means at each time
by
maternal IQ and treatment group assignment? And shouldn't it be
curvilinear given the quadratic? I guess I would like a growth curve
that takes into account significant interaction effects.

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

David Diego Torres, MA(Sociology)
PhD Candidate in Sociology


*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index