LRT: Likelihood ratio test for statistically significant dispersion
Description
Computes the -2loglikelihood value and associated p-value for testing whether Poisson regression is reasonable (H0) versus the COM-Poisson regression (H1).
Usage
LRT(x, y, betahat0, betahat, nuhat, max)
Arguments
x
matrix of size nxp
y
column vector of length n
betahat0
Maximum likelihood estimate for beta vector, obtained under the Poisson distribution
betahat
Maximum likelihood estimate for beta vector, obtained under COM-Poisson distribution
nuhat
Maximum likelihood estimate for nu under COM-Poisson distribution
max
maximum number to use for truncating infinite sums
Value
Test statistic and p-value.
Details
LRT computes the test statistic and p-value for testing H0:nu=1 vs. a two-sided alternative.
Note that the p-value relies on an asymptotic normal distribution and should therefore be used only
with sufficiently large n. n=number of records and p=number of predictors.
References
A Flexible Regression Model for Count Data, by Sellers & Shmueli, http://ssrn.com/abstract=1127359