2

Suppose one wants to assess the goodness of fit between a theoretical distribution and an empirical distribution in graphical manner. The normal Q-Q plot is good if the theoretical distribution is normal. But what if the theoretical distribution is not normal? Can you use a double logarithmic graphical method?

chl
  • 50,972
  • 18
  • 205
  • 364
Thomas James
  • 21
  • 1
  • Remember to register your account, Thomas. You will get system notification, and you will be able to vote on Q&As. – chl Feb 04 '12 at 21:15
  • 1
    We just discussed that on [a very recent question](http://stats.stackexchange.com/q/22258/7224)! – Xi'an Feb 04 '12 at 21:15
  • If you examine more than one distribution (especially if you examine more than 2) and use the empirical CDF to judge the fit, the final precision (variance) of the fitted values will inherit the imprecision of the ECDF, so you will not be gaining anything over just using the ECDF as your final estimator. – Frank Harrell Feb 04 '12 at 21:17

1 Answers1

2

Assuming you are using R, although this method should translate to other software, you can use a qqplot to look at other distributions, see here. That website has a link further down on fitting distributions in R. Again, all methods will be transferable to other software.

Michelle
  • 3,640
  • 1
  • 23
  • 33