Is there a statistical test to identify if a variable has gamma and negative-binomial distribution?
My data seems to fit both distributions by comparing it with images from books and statistical softwares. I was thinking on running both models and compare them using a goodness of fit criteria. I am new to non-linear models, so I will need to probably learn new methods of data evaluation.
Yes you might find a Kolmogorov-Smirnov test helpful if you have the CDFs for those distributions. Best wishes, D. Booth PS see the link:
https://www.google.com/search?q=kolmogorov+smirnov+test&ie=utf-8&oe=utf-8&client=firefox-b-1
Minitab software has worked great in this regard. You can test if the data has gamma and negative-binomial distribution with specific parameters.
Hi there,
As David said, you can use Kolmogorov-Smirnov test. I suggest to use "allfitdist" to find out which distribution fits best to your data. You can find useful info in the following link:
https://github.com/dcherian/tools/blob/master/misc/allfitdist.m
Good luck,
Najmeh
Thank you.
I am working with R and Stata, but will look for this functions on MATLAB. Is this a custom package?
Abolfazl Ghoodjani Najmeh Sajedianfard
Didnt know that we can test non-gaussian distributions with KS test. I will look on how to apply it.
David Eugene Booth
You can test any distribution with a K-S test if you have the cdf. See any nonparametric statistics book., e.g. Higgins, James J., Intro. to Modern Nonparametic Statistics, Sect 2.8.3. D. Booth
It's free. You can easily download it. Marcus Vinicius Veber Lopes
Gamma is continious r.v. a negative binomial is a discrete r.v.
So answer to your question is obvious.
One sample is suffice to decide
Yes, I already know this properties. But it is because I am doing some transformations do deal with my variable, and some of them still confusing to me. Igor E Kleiner
The original distribution of the variable is right skewed and inflated with zeroes, so I am studying on how to model it properly. Some readings pointed to a zero inflated negative-binomial or poisson, but my data isnt discrete to apply it. I searched for something like a zero-inflated gamma distribution but with no success.
Statgraphics is very useful to find the best fitting distribution for a column of data. please visit:
https://www.youtube.com/watch?v=O3TzhQCuvTc The Distribution Fitting procedure fits any of 45 probability distributions to a column of numeric data (You can choose the nearest 5 distributions to your data). The distributions will be sorted according to their goodness-of-fit.See this web please:
http://www.statgraphics.com/probability-distributions
Regards
Listen to what Igor E Kleiner told you. Besides that you can conveniently compare types of response distributions by modern Bayesian information criteria, WAIC or LOOIC. Both are available in R packages rstanarm and brms.
Using tests, such as the K-S, is ill-advised for two reasons:
- Badges
- Science topic
- Similar topics
- Cognitive Science
- Thinking