multicollinearity warning

[ashleyjthomas]

I’m getting the following warning and I was wondering if this is something that might be ignored?

Possible multicollinearity between b_Hierarchy_ZeroSum:other_MS_cooper and b_Intercept[2] (r = 0.86), b_Hierarchy_ZeroSum:other_MS_cooper and b_Intercept[3] (r = 0.87), b_Hierarchy_ZeroSum:other_MS_cooper and b_Intercept[4] (r = 0.87). This might lead to inappropriate results. See 'Details' in '?rope’.

This is the model that I am fitting (its likert scale data)

fit_MS_other <- brm( formula = plan_MS ~ Hierarchy_ZeroSum*other_MS_cooper+Identify+LogYears,
data = TEST_data,
family = cumulative("probit"),save_all_pars = TRUE)

[bwiernik]

I wouldn't worry about this. It's just saying there is a strong separation between levels of your factor variables and each of the dependent variable scale points (which is why the intercept thresholds are correlated with the predictors)

@easystats/maintainers we should probably omit the intercept parameters from collinearity checks. Perhaps for ordinal models like this we should check the ordering of the intercepts

[DominiqueMakowski]

Yeah, also we should probably add some links in ROPE's docs to performance::check_model() and collinearity information