[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: Re: re: analysis of multi-level data
----- Original Message -----
From: "Devendra Amre" <email@example.com>
Sent: Monday, February 03, 2003 3:32 PM
Subject: st: re: analysis of multi-level data
> I had posted this question some time ago. Unfortunately I haven`t as yet
> found an
> adequate solution (consulted quite a few statisticians!!), hence am posting
> it again.
> Essentially the response variable is measured repeatedly at 1 time point
> (say 3 months) on a group of individuals and the same done for another set
> of individuals at another time point (say 6 months). There are only 3 time
> points (3, 6 and 12 months). This is unlike a normal panel data set where
> measures are repeated on the same individuals over time.
> This is like a survey carried out among different populations at different
> times with a repeated measures entity.
> The objective is:
> 1. To compare the rate of change in the response variable within groups
> defined by the X variable.
> Earlier it was kindly suggested to look into the use of 'psuedo panel
> methods'. However in order to implement the latter I understand one will
> need to have many time points (only 3 here). Any suggestions on the
> appropriate methods to implement will be kindly appreciated.
I'm not sure I understand this completely, but it does sound like a pseudo-panel
data problem. Since the number of individuals changes period by period, Deaton
(1985, "Panel data from time series of cross-section", Journal of Econometrics)
suggests aggregating first over those individuals belonging to cohort c and
surveyed at time t and estimate the relationship on cohort means. The problem
now is that the cohort averaged individual effect varies with time. Treating the
cohort effects as random will result in inconsistent estimates while treating
them as fixed effects, the model is no longer identified, unless you treat them
as time invariant. If the cohort means are based on large number of individuals,
the latter assumption can be plausible; therefore you can use the within
estimator. If the number of individuals in each cohort is large, so that the
average cohort size nc = N/C tends to infinity, then the within cohort estimator
will be consistent (Verbeek and Nijman, 1993, "Minimum MSE estimation of a
regression model with fixed effects and a series of cross sections", Journal of
Econometrics; See also Baltagi "Econometric Analysis of Panel Data").
I hope this helps,
* For searches and help try: