testassay: Hypothesis testing procedure for assay validation for precision

Description

Does an m:n:q procedure for assay validation for precision. Returns an object of class 'assaytest'. There is are predict and print methods for that class.

Usage

testassay(x, m, n, q = 0.9, model = "normal", constant = "SD",
  data = NULL)

Arguments

The vector of assay values

The vector of values indicating sample membership

The vector of values indicating replicate membership

The confidence level, typically 0.8 or 0.9

model

String specifying the distribution for the assay values. Valid values are "normal" or "lognormal"

constant

String specifying whether the standard deviation is assumed to be constant over the levels ("SD") or the coefficient of variation is assumed constant over the levels ("CV"). The values "sd", "var", or "variance" may be used for "SD", and "cv" may be used for "CV".

data

Data frame or environment in which to look for x

Value

An object of class "assaytest", which is a list of components including a data frame of the relevant statistics calculated on x. Print, summary, predict, and plot methods are available. The list has the following components

sumtabTable summarizing the experiment, includes mean values, SD or CV estimates, and upper confidence limits on those.
UmaxThe maximum of the upper limits on the SD or CV, used in the effective SD interval calculation
nThe number of samples per level
mThe number of levels
qThe confidence level
modelThe assumed model
constantThe parameter assumed to be constant (either 'SD' or 'CV').
alphaThe alpha level, calculated as (1 - q)^(1 / m)
xThe data vector supplied by the user

Details

The m:n:q procedure uses m different samples that have different levels of the true value with n replicates for each sample. The output is a 100q percent upper limit of the bound on the precision parameter when the true values within the range of values for the m samples all follow either a a constant coefficient of variation model or a constant standard deviation model (same as a constant variance model) (see constant argument).

For example, if the 4:4:90 percent procedure using a normal model with a constant variance model returns a bound on the standard deviation (the Umax element of the assaytest class) of 7.9 then under the assumptions we have 90 percent confidence that the true SD is less than 7.9.

The predict method gives effective standard deviation intervals (i.e., 68.27 pct CIs) for the expected response from subsequent observed values from the assay.

References

Fay, MP, Sachs, MC, and Miura, K (2016). A Hypothesis Testing Framework for Validating and Assay for Precision (unpublished manuscript).

Examples

Run this code

# NOT RUN {
# reproduce Table 3 of Fay, Sachs and Miura
I<- gia$parasite=="3D7" & gia$meanAAgia<80
treD7.test<-testassay(x=gia, m=sample, n=assay, q=.9,
  data=subset(gia, parasite=="3D7" & meanAAgia<80))
treD7.test
# get estimated effective standard deviation intervals (68.27 percent CIs)
# for observed values 21.4 and 65.9
# using results from testassay
predict(treD7.test,c(21.4,65.9))

# }

Run the code above in your browser using DataCamp Workspace