On May 3, 2011, at 10:03 PM, Usha wrote:
Thanks for the help.
I would like to explain my problem.
I have sample of scores from tests which varies form 0 to 35.
Now, i want to find out the best fit distribution for this data. I
need to
order the distributions based on their best fit.
For this i am using the function fitdistr(). [One of the Ref.used :
FITTING
DISTRIBUTIONS WITH R by Vito Ricci. ]
Example:
scores<-sample(0:35,500,replace=T)
normalfit<-fitdistr(scores,"normal")
normalfit
mean sd
16.8460000 10.1361869
( 0.4533041) ( 0.3205344)
normalfit$loglik
[1] -1867.525
kstestnormal<-ks.test(scores,"pnorm",16.8460000, 10.1361869) # for
the
measure of goodness
1) Am i doing the right thing?
No. The most important right thing you are not doing is describing
your goals. Clearly you do _not_ want the best fitting distribution,
since the best fit distribution would be a multinomial distribution
with whatever probabilities would exactly fit the sample.
2) If yes, can't i follow the same procedure for all the distributions
supported by fitdistr? With the start values wherever necessary?
You can do anything you want. But have you considered the power of
this method and the error rates? Is there no science behind this to
guide what is so far an aimless search strategy?
3) Do I have to consider/worry about the warnings that I get?
We cannot force you to heed the warnings.
--
David Winsemius, MD
Heritage Laboratories
West Hartford, CT
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
