Abstract:


We suggest a new methodology for the
assessment of the goodness of fit of parametric regression models.
To this end asymptotics of common and new goodness of fit
statistics is considered under violation of the null model. This
can be used to construct asymptotic p-value curves, a new graphical tool
for understanding the quality of fit.


Ideas are illustrated with 
satelite data which are used  for the reconstruction
of the star distribution in the  Milky Way.


Keywords: goodness of fit,  parametric versus nonparametric regression,
central limit theorems, bootstrap, luminosity density, statistical 
inverse problems.