Highland Statistics Ltd. wrote: > On Tue, 13 Nov 2007 20:07:33 +0200, Anon. <[EMAIL PROTECTED]> wrote: > >> Sami Ullah wrote: >>> Hey Ecologers: >>> >>> I have a various variables for running multiple linear regression model >>> using GLM. Some of my predictor variables are non-normally distributed. >>> Using multiple linear regression, I use proc-univariate to check if the >>> residuals in the regression model met the normality criteria, which the >>> model did. >>> >>> Now I am wondering if it is advisable if I can keep the skewed predictor >>> variables in the model or have to go for non-parametric analysis? >>> >>> >> The distribution of the predictor variables is irrelevant, so you can >> happily keep them in. Well, the distribution is almost irrelevant. You >> can get problems if they are co-linear (i.e. highly correlated), or if >> you have outliers (which can have a large influence on the fit). > > Agree. One extra thing....I would argue that normality of explanatory > variables (predictors) is actually bad. It means that most observations > have the same (or similar) value for that explanatory variable, which may > (!) make it more difficult to find a significant effect. Bad experimental > design. Perhaps a histogram shaped like the uniform distribution would be > the best. It means that you have similar number of observations for each > part of you sampled gradient...for that explanatory variable. > Theoretically, the best distribution (in terms of power) is to have a bimodal distribution, with values either at their maximum or minimum. However, this design makes it impossible to check whether the relationship is linear or not. I mention this because I forgot to mention that linearity is assumed, and this is more important. It's also easy to check - plot the residuals against the predictor. If they look curved, then it suggests that the relationship is not linear, and that would be a reason to transform.
Bob -- Bob O'Hara Dept. of Mathematics and Statistics P.O. Box 68 (Gustaf Hällströmin katu 2b) FIN-00014 University of Helsinki Finland Telephone: +358-9-191 51479 Mobile: +358 50 599 0540 Fax: +358-9-191 51400 WWW: http://www.RNI.Helsinki.FI/~boh/ Journal of Negative Results - EEB: http://www.jnr-eeb.org
