deleteZeroComponents: Delete random effect terms with zero variance

In cAIC4: Conditional Akaike Information Criterion for 'lme4' and 'nlme'

Description

Is used in the cAIC function if method = "steinian" and family = "gaussian". The function deletes all random effects terms from the call if corresponding variance parameter is estimated to zero and updates the model in merMod.

Usage

deleteZeroComponents(m)

## S3 method for class 'lme'
deleteZeroComponents(m)

## S3 method for class 'merMod'
deleteZeroComponents(m)

Arguments

m	An object of class merMod fitted by lmer of the lme4-package or of class lme.

Details

For merMod class models: Uses the cnms slot of m and the relative covariance factors to rewrite the random effects part of the formula, reduced by those parameters that have an optimum on the boundary. This is necessary to obtain the true conditional corrected Akaike information. For the theoretical justification see Greven and Kneib (2010). The reduced model formula is then updated. The function deleteZeroComponents is then called iteratively to check if in the updated model there are relative covariance factors parameters on the boundary.

For lme class models: ...

Value

An updated object of class merMod or of class lme.

NULL

NULL

WARNINGS

For models called via gamm4 or gamm no automated update is available. Instead a warning with terms to omit from the model is returned.

Author(s)

Benjamin Saefken \& David Ruegamer \& Philipp Baumann

References

Greven, S. and Kneib T. (2010) On the behaviour of marginal and conditional AIC in linear mixed models. Biometrika 97(4), 773-789.

See Also

lme4-package, lmer, getME

Examples

## Currently no data with variance equal to zero...
b <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)

deleteZeroComponents(b)

cAIC4 documentation built on Sept. 22, 2021, 5:07 p.m.