In article <[EMAIL PROTECTED]>, Michael <[EMAIL PROTECTED]> wrote: >> Reference? What do you mean by "removing >> duplicates" and what is this K-S test about?
>I am testing goodness-of-fit for continuous distributions. >I have been taught to delete any duplicate data points for example if >we have 10,33,44,44,44,44,55,56 I would calculate my sample and >theorectical probabilites then delete all but ONE of the 44's (the >duplicate data points) and the calculate the differences and then find >then take the absolute values of the differences and my D-value is the >largest of those. >The question is which ONE do leave behind? Its a choice that has an >impact on the resulting D-value. >The dataset I am working almost halves in size when I remove the >duplicates, I have a lot of measurments that are the same. I am >testing the goodness-of-fit of the data to see if it's distrobution is >Gamma (or one of its special cases). >I hope the description of what I'm doing makes sense. >What method do you use to test goodness-of-fit for coninuous >distributions? If your hypothesis is that the distribution is continuous and you have observations where some are equal, and this is not merely roundoff error, your hypothesis is wrong. If the duplicates can be considered due to roundoff error, just leave them in; it will not make that much of a difference in the behavior of the test. If the test is used to test the equality of two discrete distributions, or to test a particular discrete distribution, again leave the duplicates in, and the test is somewhat conservative. I have used this myself, and it is much more powerful than such tests as the chi-squared test for reasonable alternatives. The reason is that the chi-squared test ignores order, while a reasonable alternative hypothesis might well have similar differences for nearby points. -- This address is for information only. I do not claim that these views are those of the Statistics Department or of Purdue University. Herman Rubin, Department of Statistics, Purdue University [EMAIL PROTECTED] Phone: (765)494-6054 FAX: (765)494-0558 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
