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

Re: st: RE: question from statalist

From   Maarten buis <>
Subject   Re: st: RE: question from statalist
Date   Thu, 9 Apr 2009 16:04:45 +0000 (GMT)

--- On Wed, 8/4/09, Paley, Irina  wrote:
> I use your exact code for mlogit, adapted to my dataset, and
> when I run margeff I get that it's invalid syntax. What do
> you mean by leaving out reference category? 

Say you want to controll for gender, than there are two dummies:
male (indicating who is male) and female (indicating who is 
female). These two variables contain superfluous information (if 
you know that someone isn't male than she is probably a female).
Because of that you can't add both variables in a regression,
and you need to leave one of these two out of your model. The 
variable you leave out is called the reference category. The 
same is true for your state dummies, you need to leave one of the
states out. 

As you created your dummies using -xi3- this already happened, so
this is not the problem. However, given that some of the state 
dummies have missing values on the standard errors suggests that
there is still a problem with your model: you just don't have
enough information in your data to add all the state dummies.
The easiest solution would be if you could find nearby states
that are sort of similar to the problematic states, and merge
these into "super states". So, States 50 54 2 30 46 and 72 are 
problematic (they either have a missing value on the standard 
error or they have been dropped outright due to 
multiconlinearity) and asssume that state 50 is close to State 
51, and state 54 to state 3 . Than you can create the "supper
states" by making a copy of the variable state and use -recode-:

gen state2 = state
recode state2 (50=51) (54=3) etc.

And than you use state2 instead of state in your -mlogit- model.

Hope this helps,

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen


*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index