OCplus (version 1.46.0)

smooth1d: Smoothing a vector of counts

Description

This function takes a vector of counts and uses a mixed model approach to smooth it. A common use of this is smoothing binned counts of an observed quantity prior to estimating its density nonparametrically through the relative frequencies.

Usage

smooth1d(y, sv2 = 0.1, err = 0.01, verb = TRUE)

Arguments

y

the vector of counts

sv2

the user-specified starting value for the variance of the random effects, see Details.

err

Tolerance for convergence, see Details

verb

logical value indicating whether to print diagnostics.

Value

A list with three components:

fit

the smoothed counts

df

the degrees of freedom used for smoothing at convergence

sv2

the estimated variance at convergence, equivalent to df.

Details

The smoothing assumes that the counts are Poisson from a generalized linear mixed model, where the second differences are normally distributed. Using the extended likelihood approach described in Pawitan (2001) and the initial estimate sv2 for the variance of the random effects, the routine iteratetively optimizes the fixed and random contributions to the extended likelihood, until the estimate for the variance convergences with tolerance err. The result is quite stable within a reasonable range of starting values and tolerances, and the function can be used for fairly automatic smoothing ((i.e. withou fixing a bandwidth parameter).

References

Pawitan Y.(2001) In All Likelihood, Oxford University Press, ch. 18.11

See Also

fdr1d

Examples

# Stupid dummies, obviously
smooth1d(1:10)
smooth1d(1:10, sv2=1)