Max Farrell's tutorial says refit model without outliers k>0.7

[jsocolar]

@red

Obviously doing moment matching doesn’t change the model fit, so I assume you still use the original k values from loo

Yup. If you are using k to diagnose influential observations, then you would still use the originals. Note that

@red

bias the data

is not precise language in this context. I can’t tell if I’m being obnoxiously pedantic or usefully clarifying by pointing this out. These observations with high k probably have outsized influence on the model posterior, but that doesn’t mean that the data are biased due to their inclusion. In some scenarios, high k can also be suggestive of model misspecification. For more, see

A quick note what I infer from p_loo and Pareto k values

This a quick note on different cases why Pareto k values can be large. I’ll try to extend this to a longer explanation and a case study. Any feedback on this draft is appreciated If all Pareto k small, model is likely to be ok (although there can be better models) If high Pareto k values If p_loo << the number of parameters p, then the model is likely to be misspecified. PPC is likely to detect the problem, too. Try using overdispersed model, or add more structural information (nonlinearity, …