Skip to content

mselect results not consistent with modelFit - LOF only 0 #31

@hreinwald

Description

@hreinwald

Dear drc-Team,

for some reason, mselect() only returns 0 values for the LOF (Lack of fit) when comparing multiple models. However, when I am running modelFit() I get p values clearly above the sign. threshold of 0.05.

To reproduce the error I used the ryegrass dataset as shown below:

> # Fitting a four-parameter log-logistic model
> ryegrass.LL.4 <- drm(rootl ~ conc, data = ryegrass, fct = LL.4())
> modelFit(ryegrass.LL.4) # LOF pvalue of 0.9451 <- this is NOT identical with the results of mselect
Lack-of-fit test

          ModelDf    RSS Df F value p value
ANOVA          17 5.1799                   
DRC model      20 5.4002  3  0.2411  0.8665

> AIC(ryegrass.LL.4) # IC = 41.82703 <- this is identical with the results of mselect
[1] 42.31029

> m.ls = list(LL.5(),LL.3(),W2.4(),W1.4())
> mselect(ryegrass.LL.4, m.ls)
        logLik       IC Lack of fit   Res var
W2.4 -15.91352 41.82703           0 0.2646283
LL.4 -16.15514 42.31029           0 0.2700107
LL.5 -15.87828 43.75656           0 0.2777393
W1.4 -17.46720 44.93439           0 0.3012075
LL.3 -18.60413 45.20827           0 0.3153724

From my understanding of the mselect() and modelFit() function the computation of the Lack of fit values is based on a more general ANOVA model. Hence it should return similar p-values, shouldn't it?

I am using drc package version 3.2.0 and R version 4.2.1 on a Windows System.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions