[R] Using the zero-inflated binomial in experimental designs

[Ben Bolker]

Ivan Allaman <ivanalaman <at> yahoo.com.br> writes:

> 
> 
> I'm trying to use the inflated binomial distribution of zeros (since 75% of
> the values are zeros) in a randomized block experiment with four
> quantitative treatments (0, 0.5, 1, 1.5), but I'm finding it difficult,
> since the examples available in VGAM packages like for example, leave us
> unsure of how it should be the data.frame for such analysis. Unfortunately
> the function glm does not have an option to place a family of this kind I'm
> about, because if I had, it would be easy, made that my goal is simple, just
> wanting to compare the treatments. For that you have an idea, here is an
> example of my database.
> 
> BLOCK		NIV  	       NT	        MUMI	
> Inicial		0	       18	          0	

[snip]

> 
> where: NIV are the treatments; NT is the total number of piglets born; Mumi
> is the number of mummified piglets NT. Mumi The variable is of interest. If
> someone can tell me some stuff on how I can do these tests in R, similar to
> what I would do using the function glm, I'd be grateful.
> I thank everyone's attention.

something like comparing the likelihoods of

m1 <- vglm(cbind(MUMI,NT-MUMI)~NIV*BLOCK,zibinomial,data=mydata)
m2 <- vglm(cbind(MUMI,NT-MUMI)~NIV+BLOCK,zibinomial,data=mydata)
m3 <- vglm(cbind(MUMI,NT-MUMI)~BLOCK,zibinomial,data=mydata)

I don't know whether the anova() method works for VGLM objects
or not.

  By the way, 75% zeroes doesn't necessarily imply zero-inflation --
perhaps it just means a low incidence?