I would guess that this problem is largely discussed in all the statistical
visualisation techniques, which are different to regression approaches in
that they try to visualise the variation of the data. Here, non-normal
variables are asigned 'weights' (masses), which attempt to normalise any
variation in the data.
In correspondence analysis (which is a dimension reduction technique for
categorical data, see PCA for continous variables) the concept of inertia is
introduced and the exact data transformation, goes back to the SVD or
For further reference see Greenacre, M. and Blasius, J. (1994)
'Correspondence Analysis in the Social Sciences', London: Academic Press
The insignificance of significant chi-squares is also discussed in some of
the texts of the book.
Perhaps it would be a suggestion to the pedestrian modelers to use one of
these visualisation techniques instead of a regression approach.
Department of Information Science
phone 01926 748058
----- Original Message -----
From: "Scott Bennett" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Wednesday, June 25, 2003 10:02 PM
Subject: request for information on effects of non-normality on regression
> Recently there was a a string of messages on this list relating to
> pedestrian modeling, and within that string there was some commentary on
> the distorting effects of non-normal variables in regression.
> Can anyone recommend a good review article for this problem and related
> effects? While non-normality creates some straightforward problems,
> particularly in significance testing, I have a (now vague) memory of some
> statistical simulation work that suggested that such effects tend to
> markedly smaller with extremely large samples - say of the kind one might
> find in a precise national survey.
> Thanks in advance for any references that might elaborate on the above.
> Scott Bennett
> Associate Chair
> Department of Political Science
> Faculty of Public Affairs and Management
> Carleton University
> 1125 Colonel By Drive
> Ottawa, Ontario
> K1S 5B6
> Fax (for non-confidential items) = 613-520-4064
> Email = [log in to unmask]