"the random‐effect based test for multinomial logistic regression ... is not invariant to the choice of the reference level" This "can be problematic ... because researchers may draw potentially contradictory conclusions due to different choices of the reference level" DOI 10.1002/gepi.22427