Bookmark and Share

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


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

Re: st: R: Interpretation of logistic regression coefficients


From   "Seed, Paul" <[email protected]>
To   "[email protected]" <[email protected]>
Subject   Re: st: R: Interpretation of logistic regression coefficients
Date   Tue, 29 Jan 2013 10:47:19 +0000

Thank you for this Ronan.  It quite made my morning.  

It used to be that we were very limited in the analyses we could perform, and
clinicians (whatever their actual research question were given as answer 
an adjusted odds ratio meant, which they had to interpret as best they could.

Now we have many more options, and can choose more appropriate methods.
In general, I think it is our duty as medical statisticians to find an 
analysis method that answers the question our clinical colleagues need 
an answer to.  I am increasing preferring risk ratio (or health ratios) 
and risk differences over odds ratios.  (And there seems to be quite a lot 
of interest in Population Attributable Fraction).

Best wishes all.

Paul T Seed, Senior Lecturer in Medical Statistics, 
Division of Women's Health, King's College London
Women's Health Academic Centre, King's Health Partners 
(+44) (0) 20 7188 3642.

**************************************************
Date: Mon, 28 Jan 2013 11:42:28 +0000
From: Ronan Conroy <[email protected]>
Subject: Re: st: R: Interpretation of logistic regression coefficients

On 2013 Ean 27, at 17:34, David Hoaglin wrote:

> Unfortunately, starting with linear regression (page 115), Long and
> Freese give the common but oversimplified and often incorrect
> interpretation of regression coefficients that involves holding the
> other predictors constant ("regardless of the level of the other
> variables in the model").  That interpretation does not reflect how
> multiple linear regression works.

This brings up an interesting point about teaching, understanding and truth. 

The Buddhist concept of 'right speech' asks
- - is it true?
- - is it kind?
- - is it helpful?
- - is now that time to say it?

I feel that a good initial grasp of multiple regression will embolden the student to learn more and to nuance and refine their understanding. A meticulous explanation, on the other hand, can simply confirm what every student knows deep down in their core: that they are not going to understand any of this. 

I tend to introduce multiple regression by putting a question based on two univariate analyses:
- - we know that first-time mothers have a higher risk of requiring a section 
- - and we know that older mothers have a lower risk

The question, then, is twofold:
- - if two mothers were the same age, but one was having their first child, would she be at increased risk?
- - if two mothers were alike in terms of whether or not they were having a first child, but one of them was older, would she be at decreased risk?

My feeling is that this is not doing violence to the concept, in that it paves the way for a more nuanced explanation that is much harder to get your head around. 

Does anyone have a teaching strategy that painlessly[1] introduces a more sophisticated understanding?

[1]Terms and conditions apply. This is statistics. By painless, I mean it won't hurt me.


Ronán Conroy
[email protected]
Associate Professor
Division of Population Health Sciences
Royal College of Surgeons in Ireland
Beaux Lane House
Dublin 2

**************************************************


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


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