[Limdep Nlogit List] Nested logit model: unfeasible results

Sigitas Karpavicius s.karpavicius at unsw.edu.au
Wed Aug 12 17:01:51 EST 2009

Dear all,



I would like to estimate nested logit model (2 levels; one degenerate
branch). My data includes no choice-specific variables, i.e. all
observations are invariant individual characteristics that do not depend on
choice. In addition, if degenerate branch is chosen, some variables are
always equal to 0. If I include any of the latter variables in the lower
level, then in some cases (not all) I get weird results for upper level. For
example, standard errors are very large and p-values are equal to 1:



|Variable| Coefficient  | Standard Error |b/St.Er.|P[|Z|>z]|


 WW8     |    -.00821288     .518906D+10      .000  1.0000

 WW9     |    8.36713407     .331536D+12      .000  1.0000

 WW10    |    7.01247444     .203552D+12      .000  1.0000


I was wondering whether I get such results because of the date set or
because of the degenerate branch, or because of the both. Is there any way I
could get feasible values?


In addition, I would like to ask how to transform a categorical variable
into a set of dummy variables. The categorical variable has more than 100
values and "CREATE ;expand(...)$" does not work. In output window, I get the
message: "Error 711: variable in EXPAND is not an integer from 1 to 100".
Alternatively, I could run "CREATE ;expand(...)$" several times.


Any advice would be appreciated.


Best regards,



Sigitas Karpavicius


More information about the Limdep mailing list