Re: Measurement error in longitudinal studies

On Wed, 08 Aug 2007 04:19:22 EDT, Tim De Meyer <Tim.DeMeyer@xxxxxxxx>
wrote:

[snip, previous]

Thanks very much for your reply Rich,

well in fact, the point I'm trying to make is that, when you have *big* differences in the time intervals, the estimated yearly (or monthly, whatever) weight gains are more reliable for large time intervals (say 10 years) than for very small intervals (say 2 years). And it seems from your answer you agree with that.

I think I still have a problem with your use of
terminology. "Reliable" is, for me, a term with
certain technical implications that put a bound on
measured "validity"; I might think a 6 month change
is large and reliable and valid. However, it *is* a
more term with larger variance in a model, than
terms with larger intervals. A ten-year change, on the
other hand, cannot reveal various patterns of shorter
term variation.


You had problems with the derivation of the error term for the yearly weight gain estimate. For me it's only important to know if the error for the yearly weight gains is proportial to 1/time_interval, under the appropriate statistical conditions.

- where 'error' points to the variation in the 'error term'.
It might be so, for certain models and modeling.
That would be something to *check* as much as
the dataset allows.

--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.