avg: Model averaging

In brechtdv/QMRA: Parametric Models for Quantitative Microbial Risk Assessment

Description

Average model estimates according to Akaike's Information Criterion.

Usage

avg_ea(...)
avg_drm(..., dose)

Arguments

...	Possibly named objects of class "ea" or "drm".
dose	Doses for which model-averaged responses need to be obtained.

Details

Starting from a set of K plausible candidate models, the averaged model estimate \hat{θ} is defined as ∑_{i=1}^{K}{w_i θ_i}. The weights are defined as

w_i = \frac{exp(-0.5 Δ_i)}{∑_{j=1}^{K}{exp(-0.5 Δ_j)}}

with Δ_i = AIC_i - AIC_{min}

The variance of the averaged model estimate is given by

var(\hat{θ}) = [∑_{i=1}^{K}{w_i √{var(θ_i) + (θ_i - \hat{θ})^2}}]^2

Value

An object of S3 class "avg".

Author(s)

[e-mail]

References

• Haas CN, Rose JB, Gerba CP (1999) Quantitative Microbial Risk Assessment. John Wiley & Sons, Inc.

• Burnham KP, Anderson DR (2002) Model Selection and Multimodel Inference. Springer-Verlag New York, Inc.

• Namata H, Aerts M, Faes C, Teunis P (2008). Model averaging in microbial risk assessment using fractional polynomials. Risk analysis, 28(4), 891-905.

Examples

## Exposure assessment from concentration data
gam <- ea_conc(x = x, d = d, data = giardia, model = "gamma")
lno <- ea_conc(x = x, d = d, data = giardia, model = "lognormal")
wei <- ea_conc(x = x, d = d, data = giardia, model = "weibull")
inv <- ea_conc(x = x, d = d, data = giardia, model = "invgauss")

## Model averaging
avg_ea("Gamma" = gam,
       "Log-Normal" = lno,
       "Weibull" = wei,
       "Inverse Gaussian" = inv)


## Fit several dose-response models to the Campylobacter dataset
bp <- drm(infected, total, dose, campy, "betapoisson")
ll <- drm(infected, total, dose, campy, "loglogistic")
lp <- drm(infected, total, dose, campy, "logprobit")
ev <- drm(infected, total, dose, campy, "extremevalue")

## Model averaging
avg_drm("bp" = bp, "ll" = ll, "lp" = lp, "ev" = ev,
        dose = c(1, 10, 100))

brechtdv/QMRA documentation built on May 13, 2019, 5:06 a.m.