varSumPredictNlmeGnls: varSumPredictNlmeGnls
In twNlme: Standard errors for basic nlme and gnls model predictions

Description Usage Arguments Details Value Author(s) References Examples

Variance of the sum of predictions taking care of covariances between single predictions.

1	varSumPredictNlmeGnls(object, newdata, pred = FALSE, retComponents = FALSE)

`object`	the model fit object used for predictions, treated by `attachVarPrep`
`newdata`	dataframe of new predictors and covariates
`pred`	if TRUE, the predicted value (sum of predictions) is returned in attribute pred
`retComponents`	if TRUE, the sum of the error components (fixed, random, noise) are returned in attributes "varFix","varRan","varResid"

Variance calculation is based on Taylor series expansion as described in appendix A2 by Wutzler08.

Performance of this function scales with n^2. So do not apply for too many records.

named vector

`pred`	sum of predictions
`sdPred`	standard deviation of pred
`varFix`	variance component due to uncertainty in fixed effects
`varRan`	variance component due to random effects
`varResid`	variance component due to residual variance

Thomas Wutzler

Wutzler, T.; Wirth, C. & Schumacher, J. (2008)

Generic biomass functions for Common beech (Fagus sylvatica L.) in Central Europe - predictions and components of uncertainty.

Canadian Journal of Forest Research, 38, 1661-1675

data(modExampleStem)	# load the model, which has already been prepared for prediction


#-- prediction on with varying number of records


data(Wutzler08BeechStem)


(resNlme <- varSumPredictNlmeGnls(modExampleStem, head( Wutzler08BeechStem, n=10 )))	


(resNlme2 <- varSumPredictNlmeGnls(modExampleStem, head( Wutzler08BeechStem, n=180 )))


# plotting relative error components


barplot(c(sqrt(resNlme[-(1:2)])/resNlme[1], sqrt(resNlme2[-(1:2)])/resNlme2[1]) )


# note how the residual error declines with record number, 


# while the fixed and random error does does not decline