st: xtlogit: panel data transformation's recast to double makes model incomputable

[Tom <tommedema@gmail.com>]

Dear list,

I am experiencing a very peculiar problem in STATA. I have a panel
dataset with 8000 groups and 4 million observations. I want to run a
fixed effects panel binary logistic regression (xtlogit).

As you can read in [my original question][1], STATA's maximization
algorithm cannot compute a solution when using `xtlogit`. I have
reduced this problem to occurring when transforming the data to panel
data, because:

1. running `logit depvar indepvars` works just fine
2. `logit depvar indepvars` does not maximize when first running
`xtdata indepvars, fe clear`

Therefore `xtdata` does something that causes the maximizer to fail.
It turns out that when I run `xtdata indepvars, fe clear nodouble` it
does compute. This means that stata's panel function cause the data to
be recast to the `double` format (from `float`), and this increased
precision causes the maximization process to fail.

However, I cannot use `logit` after `xtdata .., nodouble` because it
is not the same as running `xtlogit` (`xtdata` is meant for OLS).
Therefore  I need to run `xtlogit`, but it does not seem to have a
`nodouble` option, causing STATA to recast the "hidden background
panel" data to the `double` format with higher precision, and
therefore the function fails to converge.

So, is it possible to somehow tell STATA to not recast my `float`s to
`double`s when running `xtlogit`, such that the function will
converge?

Is there any other solution to this increased precision issue causing
the function to fail to converge?

Regards,
Tom

  [1]: http://stats.stackexchange.com/questions/54750/binary-panel-logistic-regression-xtlogit-fixed-effects-is-not-converging-in-st
*
*   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/