http://spssx-discussion.165.s1.nabble.com/Quantifying-Variables-in-model-tp5714917p5714939.html
I think you will be better served if you forget the stuff
about percent of the variance. That is really an awkward
and ineffective version of "effect size" in most cases, and
especially (I think) for this.
Take a baseline for days with everything favorable. Call that
100, or use the actual numbers.
Then report the increased rate observed during inclement weather,
etc., alone and in combinations. If you have several predictors, you
can report the results of the fitted equation by plugging in particular
values for the predictors.
You sometimes see lines using percentage-of-variance, like saying that
"personality is one-third genetic and two-thirds environment". These
statements are presumptive in considering *some* particular aspect
of personality... which the reader is expected to intuit... and the
results only apply well to samples that have the *same* distribution
of environmental exposures and genetic loadings. Your data on
accidents will have similar limits -- not that "accident" is so fuzzy, but
there are differences between types of roads, regional variations in
how rain typically falls, and so on.
Your description should pay attention to the actual units that *can* be
measured, and don't increase the imprecision by glossing with "percent
of variation" attributed to an abstract label.
--
Rich Ulrich
> Date: Sat, 1 Sep 2012 12:35:12 -0700
> From:
[hidden email]> Subject: Re: Quantifying Variables in model
> To:
[hidden email]>
> yes, Albert-Jan you are right my response variable is number of accidents per
> day,
>
> Thanks
> ...