bspln: Calculate Bootstrap Estimates and Confidence Intervals for the Poisson Lognormal Distribution

Description

Uses the parametric bootstrap to estimate the bias and confidence interval of the MLE of the Poisson Lognormal Distribution.

Usage

bspln(x, cutoff=1, m=200, np=2, alpha=0.95, v=NULL,
                   hellinger=FALSE)
bootstrappln(x,cutoff=1,cutabove=1000,
                          m=200,alpha=0.95,guess=c(0.6,1.2), file = "none")

Value

dist: matrix of sample CDFs, one per row.
obsmle: The Poisson Lognormal MLE of the PDF exponent.
bsmles: Vector of bootstrap MLE.
quantiles: Quantiles of the bootstrap MLEs.
pvalue: p-value of the Anderson-Darling statistics relative to the bootstrap MLEs.
obsmands: Observed Anderson-Darling Statistic.
meanmles: Mean of the bootstrap MLEs.

Arguments

x: A vector of counts (one per observation).
cutoff: Calculate estimates conditional on exceeding this value.
m: Number of bootstrap samples to draw.
np: Number of parameters in the model (1 by default).
alpha: Type I error for the confidence interval.
v: Parameter value to use for the bootstrap distribution. By default it is the MLE of the data.
hellinger: Minimize Hellinger distance of the parametric model from the data instead of maximizing the likelihood.
cutabove: Calculate estimates conditional on not exceeding this value.
guess: Initial estimate at the MLE.
file: Name of the file to store the results. By default do not save the results.

References

Jones, J. H. and Handcock, M. S. "An assessment of preferential attachment as a mechanism for human sexual network formation," Proceedings of the Royal Society, B, 2003, 270, 1123-1128.

Examples

Run this code

# Now, simulate a Poisson Lognormal distribution over 100
# observations with expected count 1 and probability of another
# of 0.2

set.seed(1)
s4 <- simpln(n=100, v=c(5,0.2))
table(s4)

#
# Calculate the MLE and an asymptotic confidence
# interval for the parameter.
#

s4est <- aplnmle(s4)
s4est

#
# Use the bootstrap to compute a confidence interval rather than using the 
# asymptotic confidence interval for the parameter.
#

bspln(s4, m=5)