Zero-Inflated Models with Application to Spatial Count Data D.K.Agarwal,A.E.Gelfand and S.Citron-Pousty, Environmental and Ecological Statistics 2002, vol 9, pp 341-355
Zero inflated poisson comes awfully close to a negative binomial, and makes more "biological" sense (i.e. there are a lot more places with zero counts and there is a process to account for them).
Thanks,
Steve
Brian R Gray wrote:
sounds like you are describing a two-part or hurdle model. a possibly more attractive but complex approach (zero-inflated count distributions) postulates two sources of zeroes: structural and stochastic. this doesn't require working with a zero-truncated count distribution. the downside is that the process defining how zeroes are separated is latent. brian
**************************************************************** Brian Gray, Ph.D. USGS Upper Midwest Environmental Sciences Center 2630 Fanta Reed Road, La Crosse, WI 54602 608-783-7550 ext 19 - Onalaska campus or 608-781-6234 - La Crosse campus fax 608-783-8058 [EMAIL PROTECTED] *****************************************************************
|---------+----------------------------> | | "Edzer J. | | | Pebesma" | | | <[EMAIL PROTECTED]| | | u.nl> | | | | | | 11/29/2003 06:35 | | | AM | | | | |---------+----------------------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| | | | To: Brian R Gray <[EMAIL PROTECTED]> | | cc: [EMAIL PROTECTED], Marcelo Alexandre Bruno <[EMAIL PROTECTED]> | | Subject: Re: AI-GEOSTATS: About gstat and binomial negative family data | >--------------------------------------------------------------------------------------------------------------------------------------------------|
I know of a paper where people split up the process in begin zero or positive (binomial), and the value of the process given that it is positive (Poisson). In fact you're working with a composite pdf, two spatial processes that have to be merged later on. The idea is attractive, but not very easy. If you want the title of the paper, email me. -- Edzer
Brian R Gray wrote:
you could modify the suggested approach by using a generalization of themay
Poisson, the neg binomial assumption you mention. most stat software
allows negative binomial regression. in this case, the variance component
of the Chi-squared resids may be better approximated (than under the
Poisson assumption). as an aside, you may have a zillion zeroes with your
fisheries data. such data may be handled moderately well by the neg bin
assumption you mention. however, they may better be handled under the
assumption that some portion of the zeroes are structural (ie *can't*
generate a positive count) rather than stochastic. I haven't seen spatial
corr assessed under these assumptions in the published lit. regardless,
such "zero inflated" models are often considerably more complicated and
not suit your purposes. brian
****************************************************************
--
* To post a message to the list, send it to [EMAIL PROTECTED]
* As a general service to the users, please remember to post a summary of any useful responses to your questions.
* To unsubscribe, send an email to [EMAIL PROTECTED] with no subject and "unsubscribe ai-geostats" followed by "end" on the next line in the message body. DO NOT SEND Subscribe/Unsubscribe requests to the list
* Support to the list is provided at http://www.ai-geostats.org
-- * To post a message to the list, send it to [EMAIL PROTECTED] * As a general service to the users, please remember to post a summary of any useful responses to your questions. * To unsubscribe, send an email to [EMAIL PROTECTED] with no subject and "unsubscribe ai-geostats" followed by "end" on the next line in the message body. DO NOT SEND Subscribe/Unsubscribe requests to the list * Support to the list is provided at http://www.ai-geostats.org
