Hi Gemma,
You've got a couple of options.
One is to transform either the dependent or the independent variable
by multiplying by a constant - you might do this before or after the
log transformation. Sometimes there's an obvious choice, if your
units are meters, you can just change them to millimeters, or if they
are years, change them to days (or vice versa).
You might consider adding a constant before you log transform, or
multiplying, this can make a difference as well - it also alters the
effect of the transformation.
Finally, you could just write shift the decimal place three points to
the effect and instead of writing:
B
X 0.0001
Write
B
X (1000s) 0.1
Jeremy
2009/1/30 Gemma Warren <[log in to unmask]>:
> Dear all
>
> I am having a slight issue when it comes to reporting my B values in my
> multiple regression analysis and I was wondering if anyone could help. Due
> to a positive skew in my dependent variables, I ended up having to subject
> this to a log transformation so that it was normally distributed.
> Unfortunately as a result of this, the unstandardised regression
> coefficients have ended up being so small as to be indistinguishable from
> each other unless you go to four or more decimal places.
>
> What is the procedure for reporting the regression coefficients in these
> cases? Do I use an exponential transformation to get the 'true' values? I
> have tried this as well, but I have unfortunately found that because the
> unstandardised coefficients tended towards 0, the exponentially transformed
> ones all hover around 1 and are similarly barely differentiated.
>
> Any help on this would be much appreciated.
>
> Regards
> Gemma Warren
>
>
> --
>
> Department of Psychology
> University of York
> York
> YO10 5DD
> 01904 432867
>
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|