fieller: Confidence Limits for Lethal Dose Estimate From Dose-response...

In jhmaindonald/qra: Quantal Response Analysis for Dose-Mortality Data

Confidence Limits for Lethal Dose Estimate From Dose-response Line

Description

This uses Fieller's formula to calculate a confidence interval for a specified mortality proportion, commonly 0.50, or 0.90, or 0.99. Here "dose" is a generic term for any measure of intensity of a treatment that is designed to induce insect death.

Usage

fieller(
  phat,
  b,
  vv,
  df.t = Inf,
  offset = 0,
  logscale = FALSE,
  link = "logit",
  eps = 0,
  type = c("Fieller", "Delta"),
  maxg = 0.99
)

fieller2(
  phat,
  b,
  vv,
  df.t = Inf,
  offset = 0,
  logscale = FALSE,
  link = "fpower",
  lambda = 0,
  eps = 0,
  type = c("Fieller", "Delta"),
  maxg = 0.99
)

Arguments

phat	Mortality proportion
b	Length 2 vector of intercept and slope
vv	Variance-covariance matrix for intercept and slope
df.t	Degrees of freedom for variance-covariance matrix
offset	Offset to be added to intercept. This can be of length 2, in order to return values on the original scale, in the case where b and vv are for values that have been scaled by subtracting offset[1] and dividing by offset[2].
logscale	Should confidence limits be back transformed from log scale?
link	Link function that transforms expected mortalities to the scale of the linear predictor
eps	If eps>0 phat is replaced by \frac{p+\epsilon}{1+2*\epsilon} before applying the transformation.
type	The default is to use Fieller's formula. The Delta (type="Delta") method, which relies on a first order Taylor series approximation to the variance, is provided so that it can be used for comparative purposes. It can be reliably used only in cases where the interval has been shown to be essentially the same as given by type="Fieller"!
maxg	Maximum value of g for which a confidence interval will be calculated. Must be < 1.
lambda	The power \lambda, when using the link="fpower". (This applies to fieller2 only.)

Details

See the internal code for details of the value g. The calculation gives increasing wide confidence intervals as g approaches 1. If g>=1, there are no limits. The default value for df.t is a rough guess at what might be reasonable. For models fitted using lme4::lmer(), abilities in the lmerTest package can be used to determine a suitable degrees of freedom approximation — this does not extend to use with glmer() or glmmTMB.

Value

A vector, with elements

est	Estimate
var	Variance, calculated using the Delta method
lwr	Lower bound of confidence interval
upr	upper bound of confidence interval
g	If g is close to 0 (perhaps g < 0.05), confidence intervals will be similar to those calculated using the Delta method, and the variance can reasonably be used for normal theory inference.

References

Joe Hirschberg & Jenny Lye (2010) A Geometric Comparison of the Delta and Fieller Confidence Intervals, The American Statistician, 64:3, 234-241, DOI: 10.1198/ tast.2010.08130

E C Fieller (1944). A Fundamental Formula in the Statistics of Biological Assay, and Some Applications. Quarterly Journal of Pharmacy and Pharmacology, 17, 117-123.

David J Finney (1978). Statistical Method in Biological Assay (3rd ed.), London, Charles Griffin and Company.

See Also

varRatio

Examples

redDel <- subset(qra::codling1988, Cultivar=="Red Delicious")
redDel.glm <- glm(cbind(dead,total-dead)~ct, data=redDel,
                  family=quasibinomial(link='cloglog'))
vv <- summary(redDel.glm)$cov.scaled
fieller(0.99, b=coef(redDel.glm), vv=vv, link='cloglog')

jhmaindonald/qra documentation built on Nov. 14, 2023, 3:16 p.m.