Colleagues,
I have a large panel dataset (7.8 million observations, of which 7.2
million are zeroes), and when I try to estimate a zero-inflated, negative
binomial using the glmmADMB package, I get an error message and am at a
loss.
The dataset has been declared as panel using the 'plm' package, and is
called 'pdat' (in use below). The time variable is 'month' (29 months of
data) and the panel identifier is 'studentbin' (~270k studentbins). The
estimation is about counts for student commencements ('commence') as a
function of subsidy rates for each student bin over time, amongst other
independent variables. All up, the .RData file is about ~200Mb of hard
drive space.
Here is the code:
*> fit_zinb <- glmmadmb ( commence ~ subsidy + month + (1|studentbin),
data=pdat, zeroInflation=TRUE, family="nbinom")*
and here is the error message:
Error in II[, ii] = II[, ii] + REmat$codes[[i]] :
number of items to replace is not a multiple of replacement length
In addition: Warning messages:
1: In `[.data.frame`(object, !omit, , drop = FALSE) :
Reached total allocation of 8097Mb: see help(memory.size)
2: In `[.data.frame`(object, !omit, , drop = FALSE) :
Reached total allocation of 8097Mb: see help(memory.size)
3: In glmmadmb ( commence ~ subsidy + month + (1 | studentbin), data =
pdat, :
NAs removed in constructing fixed-effect model frame: you should probably
remove them manually, e.g. with na.omit()
4: In II[, ii] + REmat$codes[[i]] :
longer object length is not a multiple of shorter object length
Any proposed solutions or leads will be most welcome, including alternative
packages which could work for this estimation problem.
Thanks in advance for any help,
Leo Yanes