I am looking for ways to address comparing frequencies of rare behaviours. There is a problem of trying to compare behavioural frequencies across groups of animals when the behaviour of interest is distributed as a zero inflated Poisson distribution. Transformations ease the problem but do not solve it. Has anyone dealt with this problem and found a good solution?
Hello Debra, nice name but spelt differently to me and maybe you are Dr Judge. Why not see or email Dr Stan Fenwick as I did hear him give the keynote address at the 1st international public health meet in KL in Malaysia in 2011 and it was great - he was so thorough, detailed and it was an expert presentation. He didn't particularly mention anything about what you've queried, but he is an expert on zoonotic disease so maybe he would know as maybe he has encountered this problem before given he may also know about animal behaviour in addition to disease but I'm not 100% certain about this as the presentation he gave was mostly disease related. If you go to the overseas page on my website maybe the link to the conference is still there in which case you can see his presentation and if it doesn't work or you can't find it email me back. Just an idea, sorry I can't help further. Regards, Deborah
http://sites.google.com/site/deborahhilton/
Hello Debra!
From my point of view, comparison of frequencies, whatever their distributions, should be based on log-ratios of such frequencies or, in general, log-contrasts. This is one of basics of the so-called compositional data analysis (Follow references from J. Aitchison, V. Pawlowsky-Glahn, R. Tolosana-Delgado, K. Hron and others). However, your data are counts zero-inflated-Poisson and a crucial question is how to estimate proportions. A first approach is maximum likelihood estimate of parameters (may be you prefer Bayesian estimation). This require to model the distribution of the data, wich in your case will have 2 parameter per group, and then re-expressed as log-ratios of Poisson-parameters and log-ratios of inflation probabilities. I guess this is the main problem: you need more than one proportion to describe properly your between-groups comparison.
My feeling is that your problem require more detail than that in this message. I would suggest to have a look on compositional data analysis.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics