Unexpected fittedPredictedResponse

[kduskin]

Hello,
I am using a GLMM with a binomial link function to model data. The response variable ('propExternal') is a proportion, with the number of observations accounted for through setting the weights parameter. Party and feed are each factors with two levels and account and scrape number signify the account (equivalent to subject) and the day of the observation.

DHARMa is giving unexpected behavior when I plot it. The predicted values from the model are only four different values, repeated over and over.

m2 = glmer(propExternal ~ party*feed + (1|account/scrape_number), weights=tweet, data=df, family=binomial)

simulationOutput <- simulateResiduals(fittedModel = m2, plot = F)
plot(simulationOutput, asFactor=F)
simulationOutput$scaledResiduals

When I plot it using the normal plot function, the fitted/predicted values look like they are more appropriate, but the plot is not interpretable which is why I wanted to use DHARMa.

plot(m2)

Is there any explanation for this? I am fairly new to using GLMMs and this is my first time using DHARMa.

Thank you!

[florianhartig]

Hi @ kduskin,

I suspect you have only 2 factor levels each for party*feed - DHARMa is per default plotting residuals against the fixed effect (marginal) predictions only. See #43 for an explanation.

If you want you can overrule this by plotting DHARMa residuals against the full model predictions by hand, but as discussed in the issue, there is a reason why we chose this default!

Best
Florian