2
$\begingroup$

My question is almost identical to this one. I'm searching for a probability distribution for skewed data that allows for zeros. The purpose is to fit a GLM model. My data on species-area distributions are heavy right skewed. Additional variables are "partitioned area" according to latitude- so area in tropical and temperate zones, thus the data now includes zeros.

So my question is; is there a probability distribution, e.g. similar to the log-normal, but which allows for zeros?

Any pointers would be highly appreciated.

UPDATE

Histograms of my area data

enter image description here

$\endgroup$
4
  • 1
    $\begingroup$ The Gamma distribution is the canonical example, but perhaps you should post a histogram with your data so we can get more of a feel? Gamma regression isn't the most common GLM, but it may be very appropriate here. $\endgroup$
    – user44764
    Jun 13, 2014 at 18:48
  • 4
    $\begingroup$ It's not clear, but you might be looking for zero-inflated models (which you can search about). $\endgroup$
    – whuber
    Jun 13, 2014 at 19:01
  • $\begingroup$ Unfortunately I said Gamma regression forgetting that $0$ is not in the support of the Gamma distribution. @whuber has the right of it $\endgroup$
    – user44764
    Jun 14, 2014 at 1:54
  • $\begingroup$ I got suggested this paper on "The compound poisson-gamma model" which is appropriate for skewed, zero-inflated continuous data. onlinelibrary.wiley.com/doi/10.1111/2041-210X.12122/abstract $\endgroup$
    – jO.
    Jun 16, 2014 at 13:42

1 Answer 1

1
$\begingroup$

The tweedie distribution is in the exponential family, hence can be used in a GLM context, and it allows for the occurrence of zeroes in the response variable. You could see if this suits your purpose.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.