The method is based on a maximum-likelihood estimation. The model prediction is the expected variance given the signal, depending on var0 and varR.
getErrorModel(dataexpression, verbose=FALSE)
with columns "slide" (factor, the slide names), "ab" (factor, the antibody/target names), "time" (numeric, the time points), "signal" (numeric, signal values), "var0" (numeric, error parameter for the constant error, equivalent to sigma0^2), "varR" (numeric, error parameter for the relative error, equivalent to sigmaR^2) and other columns depending on the input data.frame
data.frame, standard output from RPPanalyzer's write.Data.
logical, if TRUE, the function prints out additional information and produces a PDF file in the working directory with the signal vs. variance plots.
Daniel Kaschek, Physikalisches Institut, Uni Freiburg. Email: daniel.kaschek@physik.uni-freiburg.de
The empirical variance estimator is \(\chi^2\) distributed with \(n-2\) degrees of freedom, where \(n\) is the number of technical replicates. The estimated error parameters maximize the corresponding log-likelihood function. At the moment, the code assumes \(n=3\). For cases \(n>3\), the error parameters are slightly overestimated, thus, providing a conservative result. The explicit error model is $$\sigma^2(S) = \sigma_0^2 + S^2\sigma_R^2 = var0 + varR S^2$$ where \(S\) is the signal strength.