R help - Nov 2012 - Standard errors for predictions of zero-inflated models

Dear all,

I am using the zeroinfl() function from the pscl package to develop a
zero-inflated Poisson GLM. I would like to calculate the standard errors of
predicted values. I've tried code posted in a previous discussion on this
topic (https://stat.ethz.ch/pipermail/r-help/2008-December/182806.html), and I
don't understand the results. Before I apply this code, I get the predicted
value for the data 'newdat' using the following:

-> predict(bestfit,type="response",newdata=newdat)
109
198.5146

The predicted value is approximately 199. When I apply the previously mentioned
code (predict.zeroinfl), I can now calculate the standard error:

-> predict(bestfit,type="response",se=TRUE,newdata=newdat)
MC iterate 1 of 1000
MC iterate 2 of 1000
...
MC iterate 1000 of 1000

[[1]]
       109
0.00016793

[[2]]
[[2]]$lower
[1] 9.151924e-05

[[2]]$upper
[1] 0.0002945504

[[2]]$se
[1] 5.296472e-05

However, the predicted value is now 1.7x10^-4. The standard error value makes
sense for this predicted value, but I'm not sure why the predicted value has
changed. I would appreciate any assistance.


Cheers,

Laura

	[[alternative HTML version deleted]]