How to keep omitted categorical variable on clogit model??

[ari raka]

Currently I am doing choice experiment modelling using conditional logistic model. I am treating all of my attributes as categorical variables. Therefore I run this command using Stata14:

clogit most i.benefit i.distribution i.duration i.restriction i.participation, group(respqu)

However, I got this response:

note: 3.distribution omitted because of collinearity
note: 6.duration omitted because of collinearity
note: 1 group (3 obs) dropped because of all positive or
all negative outcomes.

Iteration 0: log likelihood = -2544.1899
Iteration 1: log likelihood = -2463.2558
Iteration 2: log likelihood = -2459.202
Iteration 3: log likelihood = -2459.1952
Iteration 4: log likelihood = -2459.1952

Conditional (fixed-effects) logistic regression

Number of obs = 8,106
LR chi2(10) = 1018.51
Prob > chi2 = 0.0000
Log likelihood = -2459.1952 Pseudo R2 = 0.1716

---------------------------------------------------------------------------------
most | Coef. Std. Err. z P>|z| [95% Conf. Interval]
----------------+----------------------------------------------------------------
benefit |
500 | 1.019902 .0955277 10.68 0.000 .8326708 1.207133
1000 | 1.347354 .094481 14.26 0.000 1.162174 1.532533
1500 | 1.353202 .093054 14.54 0.000 1.170819 1.535584
|
distribution |
1 | .2844393 .0608707 4.67 0.000 .1651349 .4037437
2 | .368224 .0605038 6.09 0.000 .2496387 .4868093
3 | 0 (omitted)
|
duration |
2 | -.0632722 .0603698 -1.05 0.295 -.1815948 .0550505
4 | -.0115473 .0621341 -0.19 0.853 -.133328 .1102333
6 | 0 (omitted)
|
restriction |
1 | -.6369584 .0605884 -10.51 0.000 -.7557095 -.5182074
2 | -.7518877 .0610098 -12.32 0.000 -.8714646 -.6323107
|
1.participation | .4798952 .0444267 10.80 0.000 .3928205 .56697
---------------------------------------------------------------------------------


I found 2 variables from 2 attributes are omitted (distribution 3 and duration 6). When I added status quo variable, the variable is omitted as well, and the response are not changed. I need to keep those 2 variables since all are important. Could somebody give me advice please to solve this problem?? Your advice is appreciated. Many thanks.

 

[Dason]

I don't use stata but it sounds like it's just doing a variation on what all software does when fitting models. If you have a categorical variable with k levels and you fit an intercept then you only need k-1 levels in the model. Adding the "k"th level doesn't add anything to the model and complicates things. You should read up on how models are "parameterized" to get a better feel for what is going on.

As far as the variable that is omitted. It looks like stata is telling you that it is constant - it doesn't change at all. There really isn't anything you can do about that now but it isn't useful for the model. If you don't have at least one observation on a different level then how could you possibly figure out what that variable is actually doing?