modelCriteriaMSE: Calculate model evaluation criteria (mainly AIC).

In thartbm/SMCL: Utilities for common tasks in the SensoriMotor Control Lab.

Calculate model evaluation criteria (mainly AIC).

Description

This function is part of a set of functions to fit and evaluate the two-rate model of motor learning.

Usage

modelCriteriaMSE(MSE, k, N, n = NA, MSEmean)

Arguments

MSE	numeric: The mean squared error between the model and data. This can be a vector if multiple models are to be evaluated. If this vector has named entries, they will be used as row names in the returned data frame.
k	numeric: The number of parameters in the model. If MSE is a vector k needs to be a vector of the same length.
N	numeric: the number of independent observations (see the function: 'seriesEffectiveSampleSize' for some options. (Not a vector.)
n	numeric: the number of observation. For a two-rate model this would be the number of trials in the sequence. Necessary for calculating AICc, but can be left NA to skip AICc.

Details

This function allows estimating model criteria based on the RMSE. This is useful for models where no method exists to extract the log-likelihood. When log-likelihood can be used, see 'AIC', 'logLik' and 'nobs' from stats.

If possible fit models by maximizing their likelihood.

The function calculates two model evaluation criteria:

"AIC": Akaike's Information Criterion

"AICc": The AIC, but with a correction for small sample sizes.

If there are 2 or more models evaluated, the relative likelihoods of the models, based on each criterion are also returned.

Value

data frame with values for several model evaluation criteria

Examples

# get example data:
data("tworatedata")

# first we baseline it, and get a median for every trial:
baseline <- function(reachvector,blidx) reachvector - mean(reachvector[blidx], na.rm=TRUE)
tworatedata[,4:ncol(tworatedata)] <- apply(tworatedata[,4:ncol(tworatedata)], FUN=baseline, MARGIN=c(2), blidx=c(17:32))
reaches <- apply(tworatedata[4:ncol(tworatedata)], FUN=median, MARGIN=c(1), na.rm=TRUE)

# and we extract the schedule:
schedule <- tworatedata$schedule

# now we can fit the model to the reaches, given the schedule:
par <- twoRateFit(schedule, reaches)
MSE2 <- twoRateMSE(par, schedule, reaches)

# we do the same for the one-rate model:
par <- oneRateFit(schedule, reaches)
MSE1 <- oneRateMSE(par, schedule, reaches)

MSEmean <- mean((schedule-mean(reaches))^2)

# effective N was calculated with "seriesEffectiveSampleSize" as: 6.833

modelCriteriaMSE(MSE=c('one-rate'=MSE1, 'two-rate'=MSE2), k=c(2,4), N=6.833, n=164, MSEmean=MSEmean)

thartbm/SMCL documentation built on Oct. 23, 2022, 5:17 a.m.