Re: Re: st: -mixlogit- numeric overflow

["Filippo Maria D'Arcangelo" <filippo.darcangelo@unibocconi.it>]

Arne,

sorry for the late reply. Hereafter an extract of our dataset:
+--------------------------------------------------------+
     ID   Set   Alter   Choice   bill    [dummy covariates]
---------------------------------------------------------
individual 1:
100080     1       1        1     10 
100080     1       2        0     50 
100080     1       3        0      0 
100080     2       1        1     50 
100080     2       2        0    100 
--------------------------------------------------------
100080     2       3        0      0 
100080     3       1        0    100 
100080     3       2        0    100 
100080     3       3        1      0 
- + - + - + - + - + - + - + - + - + - + - + - + - + - + 
Other choices from same individual......
- + - + - + - + - + - + - + - + - + - + - + - + - + - +
individual 2:
100081     1       1        0     10 
100081     1       2        1     50 
100081     1       3        0      0 
100081     2       1        0     50 
100081     2       2        1    100 
--------------------------------------
100081     2       3        0      0 
100081     3       1        1    100 
100081     3       2        0    100 
100081     3       3        0      0 
+--------------------------------------+
Other individuals.....


We use Set as our "group" variable, so it seems that we are doing it wrong.... Our stata command is
mixlogit Choice bill, rand(_covariates_) group(Set) id(ID).
Regards,

Filippo



>Filippo
>
>I suspect your "group" variable is incorrectly defined - is it a
>unique identifier for each choice situation as required, or does it
>take the values 1,2,...,8 for each respondent? Also, please show us
>your exact -mixlogit- syntax as requested in the Statalist FAQ.
>
>Arne (author of -mixlogit-)
>
>On 5 November 2013 11:09, Filippo Maria D'Arcangelo
><filippo.darcangelo@unibocconi.it> wrote:
>> Dear statalisters,
>>
>> I'm running a mixed logit following Hole (2007) [http://www.stata-journal.com/sjpdf.html?articlenum=st0133].
>> I have a dataset of 24000 observations, that is 1000 individuals, 8 choice situations and 3 alternatives per situation (i.e. 1000 * 8 * 3 = 24000).
>> As soon as I run -mixlogit- with 7 independent variables (almost equally divided in the [indepvar] and rand(varlist) places) I receive the following error:
>> 3000 (group size) take 1000 (# positives) combinations results in numeric overflow; computations cannot proceed
>>
>> I keep on receiving the error, no matter how I rearrange the covariates (or if I reduce them). I get the same message even if I reduce the number of observations.
>> I even tried to use the option [form] with a vector as a starting point with the mean of the covariates as per Hole (2007).
>> Any suggestion?
>>
>> Filippo Maria D'Arcangelo
*
*   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/