I agree with Ronan.
Of course simple correlations can disappear once other variables are taken into account. Here's an example
Amongst 20 year olds I would expect sex to be strongly predictive of both height and weight (males taller and heavier on average than females). I would also expect height to be strongly predictive of weight: on average taller people heavier than smaller people. I would expect exx to be weakly predictive of weight at most once height is in the model. (Indeed, I am unsure give two individuals of the same height but of different sexes as to which of male and female I ought to expect to be heavier.)
Regards
Stephen
Stephen Senn
Professor of Statistics
School of Mathematics and Statistics
Direct line: +44 (0)141 330 5141
Fax: +44 (0)141 330 4814
Private Webpage: http://www.senns.demon.co.uk/home.html
University of Glasgow
15 University Gardens
Glasgow G12 8QW
The University of Glasgow, charity number SC004401
-----Original Message-----
From: Evidence based health (EBH) [mailto:[log in to unmask]] On Behalf Of Ronan Conroy
Sent: 18 August 2011 17:15
To: [log in to unmask]
Subject: Re: Statistics question
On 2011 Lún 18, at 16:37, Cristian Baicus wrote:
> I am asking for help from this group in a statistical problem.
>
> I know that when one performs a multivariable analysis, he must be sure that there is no correlation between the independent variables (and this can be verified by plotting the multicollinearity matrix).
This is not true, and in most instances there will be a correlation of some sort between the independent variables. The real problem is dealing with it, because correlated independent variables mean that the estimated standard errors increase and study power decreases.
Ronán Conroy
[log in to unmask]
Associate Professor
Division of Population Health Sciences
Royal College of Surgeons in Ireland
Beaux Lane House
Dublin 2
|