Open florianhartig opened 6 years ago
Note: gamlss seems to suggest a similar procedure in https://www.rdocumentation.org/packages/gamlss/versions/5.1-4/topics/rqres.plot
D&S comment that they recommend several plots. I think what's easier to do is to do several simulations, and then plot with densities, and adjust weights in all tests or correct for multiple testing.
When re-simulating residuals for fixed data and integer responses, p-values may spread within a certain range, due to the randomization procedure to smoothen out the integer values.
This phenomenon was discussed in https://github.com/florianhartig/DHARMa/issues/37
Particularly in low-data situation, this can result in p-values being quite variable. In this case, it might be desirable to have an option to obtain an aggregate p-value from multiple simulations.
Question is
a) how to best do this
b) what the properties of the resulting p-values are in terms of their distribution / type I error / power