drc (version 2.1-1)

mselect: Model selection

Description

Model selection by comparison of different models using the maximum log likelihood value, Akaike's information criterion (AIC), the estimated residual variance and the p-value from a lack-of-fit test as criteria.

Usage

mselect(object, fctList = NULL, nested = FALSE, sorted = c("IC", "Res var", "Lack of fit", "no"), linreg = FALSE, icfct = AIC)

Arguments

object

an object of class 'drc'.

fctList

a list of dose-response functions to be compared.

nested

logical. TRUE results in F tests between adjacent models (in 'fctList'). Only sensible for nested models.

sorted

character string determining according to which criterion the model fits are ranked.

linreg

logical indicating whether or not additionally polynomial regression models (linear, quadratic, and cubic models) should be fitted (they could be useful for a kind of informal lack-of-test consideration for the models specified, capturing unexpect

icfct

function for supplying the information criterion to be used. AIC and BIC are two options.

Value

• A matrix with one row for each model and one column for each criterion.

Details

The maximum likelihood cannot be used for comparison unless the models are nested. For Akaike's information criterion and the residual variance: the smaller the better and for lack-of-fit test: the larger the better. Note that the residual variance is only available for continuous dose-response data.

Examples

### Example with continuous data
## Fitting initial four-parameter log-logistic model
ryegrass.m1 <- drm(rootl ~ conc, data = ryegrass, fct = LL.4())

## Model selection
mselect(ryegrass.m1, list(LL.3(), LL.5(), W1.3(), W1.4(), W2.4(), baro5()))

## Model selection including linear, quadratic, and cubic regression models
mselect(ryegrass.m1, list(LL.3(), LL.5(), W1.3(), W1.4(), W2.4(), baro5()), linreg = TRUE)

## Comparing nested models
mselect(ryegrass.m1, list(LL.5()), nested = TRUE)

### Example with quantal data
## Fitting initial two-parameter log-logistic model
earthworms.m1 <- drm(number/total~dose, weights=total, 
data = earthworms, fct = LL.2(), type = "binomial")

## Comparing 4 models
mselect(earthworms.m1, list(W1.2(), W2.2(), LL.3()))
# model selection based AIC or p value not working in this example!!!