Re: Re: st: interpreting marginal effects of fractional logit with continuous independent variables

[David Hoaglin <dchoaglin@gmail.com>]

Hi, Sandra.

Having only the proportions makes the analysis challenging.  In
particular, if you don't have the denominators (or standard errors
themselves), you can't take the variability of the individual
proportions into account.  I suppose knowledge of some numerators
could risk compromising confidentiality; and if you know the
proportion and the denominator, you essentially have the numerator.
Still, it seems unlikely that all 8500 observations should receive
equal weight.

In pursuing interpretations of those effects, it may not be
appropriate to assume that the data support "all else held at means."
I suggest looking for ways to explore the combinations of values of
those two independent variables, the other covariates, and the factor
variables in the data.  A start might be to ask how much data you
actually have in the region of "predictor space" near the point
defined by the means of "all else."  Then look at how the data support
the combination of "all else at means" and relevant variation in the
values of ple and llti_stand.

David Hoaglin

On Mon, Nov 18, 2013 at 10:28 AM, Sandra Virgo <Sandra.Virgo@lshtm.ac.uk> wrote:

> Hello David - thanks for your response.
>
> I'm using fractional logit because for various reasons the raw data are confidential, so I have been given them in proportion form but have no more information.
>
> I have around 8500 individual observations, and they are data for electoral wards. Apart from the DV the data are in the public domain and were collected via the Census and other administrative sources.
>
> Sandra
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/