[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Interaction effect or running seperate models

From   "Clive Nicholas" <>
Subject   Re: st: Interaction effect or running seperate models
Date   Thu, 27 Sep 2007 05:49:13 +0100

Quang Nguyen wrote:

> Suppose, I had a simple model: Y = Ao + A1*X1 + A2*Z + u, where A2 is
> a categorical variable e.g., ethnicity. If I wanted to see how
> different ethnic groups may have different effect on Y, I can use an
> interaction term: X1*Z or I can run seperate equations fro each
> ethnicity. Can you tell me the advantage of each method?

Jaccard (2001: 17) provided a very good reason to prefer the
single-model analysis in the case of logistic regression. I don't
believe it would be any different for OLS regression:

"It is not uncommon for researchers [to calculate] seperate logistic
regression equations for males and females and then examining whether
the logistic coefficient for X is "statistically significant" (i.e.,
has an associated p value less than 0.05) in both analyses. If the
coefficient is statistically significant in one group but not in the
other, then the conclusion is that X is more important for the one
group than for the other. This logic is flawed because the researcher
never performs a formal statistical test of the _difference_ between
the logistic coefficients for the two groups."

You will, of course, note that is there is now a substantial research
literature that has called into question the practice of using
interaction terms in LR models (much of which has been discussed on
this list), but I hope this helps to answer the original question.

Clive Nicholas

[Please DO NOT mail me personally here, but at
<>. Thanks!]

Jaccard J (2001) "Interaction Effects In Logistic Regression", QASS
Series 135, Thousand Oaks, CA: Sage.
*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index