"H. Noedl" wrote:
> Unfortunately my knowledge of statistics is rather limited to the basics
> (i.e. regression, t-test etc.). I tried by transforming the drug
> concentration into NLog (LN(x)) and the response into probits
> (NORMSINV(x)+5) and doing an ordinary linear regression (y=a+bx) to
> calculate effective concentrations (EC50, EC90, EC99). Unfortunately the
> results are sometimes hardly comparable to the results I get with SPSS
> (generally much lower). I also don't really know how to handle 0 and
> 100% (and I don't know anything about things like maximum likelihood,
> weighting etc.). I downloaded the model description from SPSS. Even
> though very elaborate the model description from the SPSS-website was
> not too much help to a medicine man.
One way of dealing with 0 and 100 in these circumstances is to recode
0's as (say) 0.0001 and 100 as 99.999. I don't whether this would
improve the model though.
Nick.
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
