Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
st: coldiag and pertub
William Hauser <firstname.lastname@example.org>
st: coldiag and pertub
Sun, 13 Nov 2011 13:28:13 -0500
I'm running a logit model with a single interaction term and factor
variable notation. I'm having trouble understanding the collinearity
The model is of the form:
logit y c.var1##c.var2 i.var3 c.var4
There are 11 variables in the model in total but I have omitted them
above for brevity.
Stata is version 12.
coldiag2 won't run after logit but I did plug in my independent
variables into the command and got a condition index number of 397
which is, as you would expect, driven by the association between a, b,
and the interaction term aXb. I'm not sure what to make of this. My
instinct is that the high condition index is ok because it is the
result of the interaction term but I'm unsure.
More troubling, without the interaction term the index is still 98 but
this appears driven by the association of of var1 with the "constant
term." I'm not sure what the constant term is, the coldiag2 help file
is vague about this. I know the old coldiag program defaulted to no
constant but the newer version includes the constant by default. If I
omit the constant using the "nocons" option then the condition index
is a more reasonable value of 20.
What is the constant term and what do I make of all this?
I've also experimented with the perturb command but can't find any
guidance about how to interpret the results or how to specify the
perturbations. For example, how large should the perturbation be for
the continuous variables? 1 standard deviation? Surely the results
are a function of how large the perturbation is. It's also not clear
to me what's reported in the summary table from the command. Is the
"mean" the average coefficient observed across the 100 iterations? Or
is it the mean variation in the coefficient across the 100 iterations?
The help file for perturb is rather vague and I can't find any
examples using perturb on the UCLA Stata website either.
Any guidance you could offer would be greatly appreciated.
* For searches and help try: