Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

Re: st: compare effect size between dummys and metrics variables in logistic regression

From   Morten Hesse <>
Subject   Re: st: compare effect size between dummys and metrics variables in logistic regression
Date   Mon, 27 Sep 2010 17:10:13 +0200

OK Maarten. Good point, and thank you for that. "Statistical significance" in the absence of "substantial significance" is quite meaningless. I very much endorse that view. And to be honest, I have some difficulty coming up with an example where a nominal variable and a continuous variable can be meaningfully compared.

Den 27-09-2010 13:48, Maarten buis skrev:
--- On Mon, 27/9/10, Morten Hesse wrote:
It may well be that Michael is right. But if I conduct a logstic
regression, and then write:

test [continuous covariate]=[1.dichomotomous covariate]

Then I get a chi-square statistic (with one df) and a
p-value. It looks just like any other result of "test":

In my understanding, this asks the question "is the
difference between the value of the coefficient when
dic_covar=1 significantly different from the coefficient of

If this does not make sense, then why does STATA do it? Is
this not what Joerg would want to use?
It is a substantive question. Stata knows nothing about
substantives, that is what we researchers get paid for. Stata
is only for the boring mechanical part of an analysis.

Consider the following example:

*----------- begin example ------------
sysuse nlsw88, clear
gen black = race == 2 if race<= 2
logit union grade
test = grade
*------------ end example -------------

The coeficient for compares the odds of union membership
of white respondents with the odds of union membership of black

The coeficient of grade compares the odds of membership of a
respondent with the odds of membership of another respondent with
1 more year of education.

Can you compare the difference between black and white with the
difference between x years of schooling and x + 1 years of schooling?
Probably not, but how is Stata supposed to know? Like any other
computer program, Stata just does what you tell it to do. It is up
to you to make sure that what you tell it to do makes sense.

A valid position  when it comes to comparing the effects of the
kind of variables I used in the example above is that such variables
are just different, and you just cannot compare them.

Sometimes you have good reason to want to compare the effects of such
variables (often that is not the case, and than you should just not
do it), and that is where standardization comes in. Standardization
is intended to make the unit of variables comparable. There are various
ways of doing that
<>. Which
one you think is least awkward is a substantive question that depends
on the specific variables you are comparing.

Hope this helps,

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

*   For searches and help try:
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index