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 at the end of May, and its replacement, statalist.org is already up and running.


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

Re: st: xtlogit, pa questions


From   Kim Peeters <kimpeeters84@yahoo.com>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   Re: st: xtlogit, pa questions
Date   Thu, 8 Mar 2012 01:29:19 -0800 (PST)

Dear Brendan and David,

Within each cluster the value of the response variable is either
0 or 1 for all occasions. This is not due to any design but just because none
of the participants (one participant = one clusters) in the study experienced a
transition from one state to the other state (no data or study error).

Both the response variable and the RHS variables are measured
yearly. The response variable does not vary but recall that this was not done
by design. The RHS variables do vary for each occasion (i.e. for every year).

The study endeavors to understand the relationship between
the response variable (either 0 or 1) and the independent variables.

Thank you for any help you can provide.

Best regards,
Kim



----- Original Message -----
From: David Hoaglin <dchoaglin@gmail.com>
To: statalist@hsphsun2.harvard.edu
Cc: 
Sent: Tuesday, March 6, 2012 3:52 PM
Subject: Re: st: xtlogit, pa questions

Please clarify the following point:
> 3. Within each cluster, the response variable is
> always the same (either 0 or 1). As such, is a population averaged logit modeling
> approach still
> statistically valid?
Are you saying that, within a cluster, either the value of the
response variable is 0 for all occasions or the value of the response
variable is 1 for all occasions (i.e., the values within a cluster are
not a mixture of 0s and 1s)?

David Hoaglin


----- Original Message -----
From: Brendan Halpin <brendan.halpin@ul.ie>
To: statalist@hsphsun2.harvard.edu
Cc: 
Sent: Tuesday, March 6, 2012 1:43 PM
Subject: Re: st: xtlogit, pa questions

On Tue, Mar 06 2012, Kim Peeters wrote:

> You mention “If they do, it seems to me that cluster membership has a
> stronger effect than the covariates, probably for structural reasons.”
> Does this imply that there exist better models to establish the
> relationship between the response variable and the independent
> variables?

If I understand it correctly, you have multiple X measurements per
person, but only a single Y?  Does the outcome come after all the RHS
measurements? 

If there is, by design, a single outcome per case, then this is not
really clustered data, but rather individual-level data with multiply
observed explanatory variables:

Y_i = f(X_{i1},X_{i2},...X_{iT}) rather than Y_{it} = f(X_{it})


Rather than a clustered model, I'd be looking for a way of simplifying
the RHS variables, e.g., via factor analysis. If they are categorical
you might be able to cluster them using sequence analysis (depending on
how time functions, and on your substantive question).

Regards,

Brendan


----- Original Message -----
From: Kim Peeters <kimpeeters84@yahoo.com>
To: Statalist <statalist@hsphsun2.harvard.edu>
Cc: 
Sent: Tuesday, March 6, 2012 11:34 AM
Subject: st: xtlogit, pa questions

Dear,

I am analyzing a population averaged logit model for panel
data (xtlogit, pa), which is equivalent to a Generalized Estimating Equation
model (xtgee) when the link function is logit, the distribution of the response
variable is binomial and the correlation structure within the response variable
is exchangeable.

I have read numerous papers and text books (including
Generalized Estimating Equations by Hardin and Hilbe (2003)) but I have some unresolved questions:

1. I report the odds ratio for each independent
variable. I suppose that the interpretation of the odds ratio is similar to the
interpretation of odds ratios in the standard logistic regression. Is this
correct?
2. When I adjust my standard errors for clustering,
the obtained semi-robust standard errors are smaller than the standard errors I
obtain when I do not adjust my standard errors for clustering? This appears to
be counter-intuitive? Is this phenomenon valid and how should I interpret this?
3. Within each cluster, the response variable is
always the same (either 0 or 1). As such, is a population averaged logit modeling
approach still
statistically valid?

Thank you for any help you can provide.

Best regards,
Kim


*
*   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