Morpho (version 2.12)

predictShape.lm: Predict shapes based on linear models calculated from PCscores

Description

Predict shapes based on linear models calculated from PCscores.

Usage

predictShape.lm(fit, datamod, PC, mshape)

Value

predicted

array or matrix containing predicted landmark coordinates

predictedPC

matrix containing predicted PC-scores

Arguments

fit

model of class lm where the PCscores are fitted onto

datamod

a one-sided "model" formula, of the form ~ x1 + x2 + ... + xk, corresponding to the right hand term in the model used in fit. If omitted, the predicted shapes of all specimen are calculated based on the fitted values.

PC

Matrix/vector containing Principal components (rotation matrix) corresponding to PC-scores used in fit.

mshape

matrix of the meanshape's landmarks by which the data was centered before rotation in covariance eigenspace.

Warning

Make sure that the levels of the variables used in datamod correspond exactly to those used in fit. Otherwise model matrix will be calculated erroneous.

Details

This function predicts the landmarks based on models calculated from PCscores.

See Also

model.matrix, lm, formula

Examples

data(boneData)
proc <- procSym(boneLM)
pop <- name2factor(boneLM,which=3)
##easy model with only one factor based on the first four PCs
fit <- lm(proc$PCscores[,1:4] ~ pop)
## get shape for Europeans only
datamod <- ~as.factor(levels(pop))[2]
Eu <- predictShape.lm(fit,datamod, proc$PCs[,1:4],proc$mshape)

## get shape for Europeans and Chinese
datamod <- ~as.factor(levels(pop))
pred <- predictShape.lm(fit,datamod, proc$PCs[,1:4],proc$mshape)
if (FALSE) {
deformGrid3d(pred$predicted[,,1], pred$predicted[,,2], ngrid = 0)
}

## more complicated model

sex <- name2factor(boneLM,which=4)
fit <- lm(proc$PCscores[,1:4] ~ pop*sex)
## predict female for chinese and European
datamod <- ~(as.factor(levels(pop))*rep(as.factor(levels(sex))[1],2))
pred <- predictShape.lm(fit,datamod, proc$PCs[,1:4],proc$mshape)

## predict female and malefor chinese and European
popmod <- factor(c(rep("eu",2),rep("ch",2)))
sexmod <- rep(as.factor(levels(sex)),2)
datamod <- ~(popmod*sexmod)
pred <- predictShape.lm(fit,datamod, proc$PCs[,1:4],proc$mshape)

## add some (randomly generated) numeric covariate
somevalue <- rnorm(80,sd=10)
fit <- lm(proc$PCscores[,1:4] ~ pop+somevalue)
probs <- quantile(somevalue, probs=c(0.05, 0.95))
## make model for European at 5% and 95% quantile
popmod <- rep(factor(levels(pop))[2],2)
datamod <- ~(popmod+probs)
pred <- predictShape.lm(fit,datamod, proc$PCs[,1:4],proc$mshape)