gnlht: Test nonlinear functions of model parameters

In OnofriAndreaPG/aomisc: Statistical methods for the agricultural sciences

Test nonlinear functions of model parameters

Description

This function calculates linear/nonlinear functions of model parameters and returns an estimate with standard errors.

Usage

## S3 method for class 'numeric'
gnlht(coefs, func, const, vcov.,
            parameterNames, dfr)
## S3 method for class 'lm'
gnlht(model, func, const, ...)
## S3 method for class 'nls'
gnlht(model, func, const, ...)
## S3 method for class 'lme'
gnlht(model, func, const, ...)
## S3 method for class 'nlme'
gnlht(model, func, const, ...)
## S3 method for class 'drc'
gnlht(model, func, const, ...)

Arguments

coefs	a numeric vector
vcov.	a variance-covariance matrix, or a function to calculate it
parameterNames	a character vector with namings for 'coefs'
dfr	degrees of freedom
model	a model object for which there are coef and vcov methods, and for which the named coefficient vector returned by coef is asymptotically normally distributed with asymptotic covariance matrix returned by vcov.
model	a model object for which there are coef and vcov methods, and for which the named coefficient vector returned by coef is asymptotically normally distributed with asymptotic covariance matrix returned by vcov.
funList	a dataframe with one column, listing the quoted strings that are the functions of the parameter estimates to be evaluated
constList	If necessary, a dataframe whose columns are the constants to be used in the calculations above. For each row of this dataframe, the functions above are evaluated

Value

A data.frame

Author(s)

Andrea Onofri

Examples

rm(list = ls())
library(aomisc)
data(metamitron)

#Fit nls grouped model
modNlin <- nls(Conc ~ A[Herbicide] * exp(-k[Herbicide] * Time), 
               start=list(A=rep(100, 4), k=rep(0.06, 4)), 
               data=metamitron)
summary(modNlin)


# Compare parameters #######
funList <- list(~k1 - k2, ~k1 - k3, ~k1 - k4)
gnlht(modNlin, funList)

# Combine parameters
funList <- list(~ -log(0.5)/k1, ~ -log(0.5)/k2,
                ~ -log(0.5)/k3, ~ -log(0.5)/k4)
gnlht(modNlin, funList)

# Combine more flexibly
funList <- list(~ -log(prop)/k1, ~ -log(prop)/k2,
                ~ -log(prop)/k3, ~ -log(prop)/k4)
gnlht(modNlin, funList, const = data.frame(prop = 0.5))

funList <- list(~ -log(prop)/k1, ~ -log(prop)/k2,
                ~ -log(prop)/k3, ~ -log(prop)/k4)
gnlht(modNlin, funList, const = data.frame(prop = c(0.7, 0.5, 0.3)))

# Other possibilities
funList <- list(~ (k2 - k1)/(k1 * k2) * log(prop),
                ~ (k3 - k1)/(k1 * k3) * log(prop), 
                ~ (k4 - k1)/(k1 * k4) * log(prop))
gnlht(modNlin, funList, const = data.frame(prop = c(0.7, 0.5, 0.3)))

# Predictions
funList <- list(~ A1 * exp (- k1 * Time), ~ A2 * exp (- k2 * Time), 
                ~ A3 * exp (- k3 * Time), ~ A4 * exp (- k4 * Time))
propdF <- data.frame(Time = seq(0, 67, 1))
func <- funList
const <- propdF
pred <- gnlht(modNlin, funList, const = propdF)
head(pred)
tail(pred)

# Back transformation of means
library(emmeans)
data(mixture)
head(mixture)
mod <- lm(sqrt(Weight) ~ Treat, data = mixture)
med <- emmeans(mod, ~Treat)
coefs <- as.data.frame(med)[,2]
#parameterNames <- paste("b", 1:4, sep = "")
parameterNames <- as.data.frame(med)$Treat
pn <- parameterNames
# func <- as.list(paste("(1 -", pn, ")/", pn, sep = ""))
# func <- as.list(paste("exp(", pn, ") - 1", sep = ""))
# func <- as.list(paste("(1 -", pn, ")/", pn, sep = ""))
func <- as.list(paste(pn, "^2", sep = ""))
vcov. <- diag(as.data.frame(med)[,3]^2)
sigma <- vcov.
const <- NULL

medieBack <- gnlht(coefs, func,  vcov. = sigma,
                           parameterNames = pn, dfr = 12)
medieBack
emmeans(mod, ~Treat, transform = "response")

OnofriAndreaPG/aomisc documentation built on Feb. 2, 2023, 12:13 p.m.