[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: interpreting marginal effect after logit
--- Yun Liu <firstname.lastname@example.org> wrote:
> I estimate a logit model and need your help in interpreting the
> results. Please see below for part of the output. Both x1 and x2 are
> statistically significant. But based on the derivatives, one standard
> deviation change in x1 increases the probability by merely 1%
> (0.074*0.129). Similarly one standard deviation change in x2
> increases the probability by merely 2% (5.9*0.003). I think the
> reason is that y is unbalanced and the predicted y is only 0.148 at
> sample mean.
> Are these marginal effects embarrassingly low? What is the common
> practice in presenting such results?
With 10,000 cases you shouldn't be surprised that some of the
significant effects are very small, maybe so small that they are
substantively irrelevant even though statistically significant. Whether
or not it is substantively irrelevant depends on the situation. If you
can tell a story based on those 1 and 2% that is interesting, than it
is not irrelevant, otherwise it is.
Also this is only the effect at the mean values of x, the effects
change with the values of x. An excellent way to get a feel for how
much the effect changes is to graph the predicted probabilities against
one explanatory variable (while keeping the rest at some value (e.g.
mean). Have a look at -prgen- in the -spost- package.
Instead of looking at the marginal effect at mean values of x, some
people like to look at the mean value of the marginal effect. There is
a good reason for that, since the latter is a better description of the
marginal effects, since it is its expected value. Also this value is in
this case likely to be a bit larger. See: -ssc desc margeff-.
Hope this helps,
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
1081 HV Amsterdam
Buitenveldertselaan 3 (Metropolitan), room Z434
+31 20 5986715
Yahoo! Answers - Got a question? Someone out there knows the answer. Try it
* For searches and help try: