Log-likelihood displacements for single observation and single grouping variable

Share:

Description

Functions for log-likelihood displacements for each observation or each level of given factor

Usage

1
2
3
4
5
recalculateLogLik(model, fixef = fixef(model), vcor = VarCorr(model)) 

groupDisp(formula, data, var) 

obsDisp(formula, data, inds=1:nrow(data)) 

Arguments

model

a mixed model of the class mer,

fixef, vcor

model parameters log-likelihood evaluation, if not provided then the estimates extracted from the 'model' parameter will be used

formula

a model formula that will be passes to the nlme function

data

a data frame

var

a name of grouping variable (factor) for which the group log-likelihood displacement will be performed

inds

indexes of observations for which observation log-likelihood displacement will be performed

Details

Likelihood displacement is defined as a difference of likelihoods calculated on full dataset for two models with different sets of parameters. The first model is a model with ML estimates obtained for full dataset, while the second model is a model with ML estimates obtained on dataset without a selected observation or group of observations.

Likelihood displacements are used in model diagnostic.

Note that these functions reestimate coefficients in a set of model may be a time consuming.

The function recalculateLogLik() calculated a log-likelihood for model defined by the object model and model parameters defined in following function arguments.

The functions groupDisp() and obsDisp() calculates how the log-likelihood will decrees if selected groups or selected observations will not be used for parameter estimates. Note that log-likelihood is calculated on full dataset.

Author(s)

Przemyslaw Biecek

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
data(eunomia)
require(lme4)
set.seed(1313)
eunomias <- eunomia[sample(1:2000,100),]
groupDisp(formula = BPRS.T2~ (1|CENTRE13), data=eunomias, var="CENTRE13")
 
obsDisp(formula = BPRS.T2~ (1|CENTRE13), data=eunomias, inds = 1:10)
 
obsDisp(formula = BPRS.T2~ (1|CENTRE13), data=eunomias)