How are scaled residuals calculated when grouping is used?

[akhileshtayade]

Hello!

I have been unable to understand how DHARMa generates scaled residuals when recalculateResiduals is used with argument group.

Here is a minimal, reproducible example (output not shown) to describe what I am having trouble with:

# DHARMa version 0.4.6
library(DHARMa)

# generate data
testData = createData(sampleSize = 250)
fittedModel <- glmer(observedResponse ~ Environment1 + (1|group) , 
                     family = "poisson", data = testData)

simulationOutput <- simulateResiduals(fittedModel = fittedModel, plot = T)

# there are 10 groups in the data
unique(testData$group)

# residuals are grouped according to the groups in the data
groupedSimulationOutput <- recalculateResiduals(simulationOutput, group = testData$group)

# the first 25 rows belong to the first group
testData[testData$group == 1, ]

# confirms that recalculateResiduals just sums the predicted responses 
# from simulationOutput for each group by default
sum(simulationOutput$fittedPredictedResponse[1:25]) == groupedSimulationOutput$fittedPredictedResponse[1]

However, I did not understand how the scaledResiduals are calculated for the groupedSimulationOutput: how are groupedSimulationOutput$scaledResiduals calculated from simulationOutput$scaledResiduals? I hope I am not missing something obvious.

[florianhartig]

Hello akhileshtayade,

basically, the observed data (simulated and real observed data) is summarised per group according to the function that is provided (default = sum) in , and then the quantile residual is calculated as for a single data point (which is described in the vignette).

Does that help?

Best F

[akhileshtayade]

Yes, that helped! Thank you very much for a super quick response!

I also checked this commit and the code of the function recalculateResiduals which also helped in resolving the doubt. Thank you again!