Denis Lafont
Apr14-09, 03:33 PM
Hi
I am working at the moment on injury criteria. In the biomechanics domain the risk curve used is often a logistic regression. But this modelisation of the data is inconsistent not only because P(0) ≠ 0 but mainly because a bad information for exemple around a probability of 80 % will change largely the 10 % region and in the wrong direction: the 10 % limit decreases when we add a point: no injury for a high value of the parameter, it should of course increase since it is logical to find that the risk decreases when we add information corresponding to a no_injury result.
This is due to the fact that the logistics curve is antisymmetric 1-P(x)=P(2m-x) and that it uses only two parameters. So would you know one (or more) equation(s) not antisymmetric which could be used as "risk cuves".
I tried also normal, lognormal and Weibull equation. Only the Weibull curve with 3 parameters could answer the question but as this last equation presents a discontinuity it cannot be computed automatically very easily.
Thank you for your help. Sorry for my English.
Regards.
I am working at the moment on injury criteria. In the biomechanics domain the risk curve used is often a logistic regression. But this modelisation of the data is inconsistent not only because P(0) ≠ 0 but mainly because a bad information for exemple around a probability of 80 % will change largely the 10 % region and in the wrong direction: the 10 % limit decreases when we add a point: no injury for a high value of the parameter, it should of course increase since it is logical to find that the risk decreases when we add information corresponding to a no_injury result.
This is due to the fact that the logistics curve is antisymmetric 1-P(x)=P(2m-x) and that it uses only two parameters. So would you know one (or more) equation(s) not antisymmetric which could be used as "risk cuves".
I tried also normal, lognormal and Weibull equation. Only the Weibull curve with 3 parameters could answer the question but as this last equation presents a discontinuity it cannot be computed automatically very easily.
Thank you for your help. Sorry for my English.
Regards.