[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: logistic versus probit
At 09:48 AM 6/14/2005 -0700, Hyojoung Kim wrote:
If by "different regression outcomes" you mean different substantive
conclusions - that is very unusual, and it might make you wonder about your
data and models as much as it does about which method is better.
I have one basic question. It is my understanding that people alternatively
use logistic and/or probit regression analyses for a categorical dependent
variable, although the two methods are based different distributional
assumptions. The rationale, as far as I can tell, is that they rarely make a
substantive difference in practice.
But, what if they result in different regression outcomes? Is there any
formal test for determining which is more appropriate than the other? If
there is, how do you do it in Stata? Or, ff there is no such a formal test,
what is the convention for choosing (if you choose) what to report?
Long (1997 - Regression models for categorical and limited dependent
variables) has a brief discussion of this on p. 83. A lot of it may just
be whatever the convention is in your field. Some people like being able
to interpret odds ratios in logistic regression. Advanced generalizations
may be a factor, e.g. multiple-equation biprobit models or multinomial
logistic regression. Stata 9 has added some advanced probit-related
routines, which might affect the decision.
Rather than worry about fine-line distinctions between logit and probit
results, depending on your circumstances you might want to think about
other alternatives, such as scobit (skewed logit) or cloglog (complimentary
log-log) models. There is a super-brief overview of such things at
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
WWW (personal): http://www.nd.edu/~rwilliam
WWW (department): http://www.nd.edu/~soc
* For searches and help try: