I've applied the Stata non-parametric test for trend (nptrend) for a two-way table including one ordered categorical data (quintiles of deprivation index).
Such clearly ordered variables are straightforward, but what about variables such as education (where there is a general order, but some categories are not clearly higher or lower than others: apprenticeship vs secondary/high school) or smoking/alcohol (current/ex/never, where an order can be argued - exposure - but not clearly: quitting may or may not resolve different aspects of health detriment).
I appreciate that the answer predominantly depends on the reserach question (eg in the case of smoking/alcohol). From the statistical tests' point of view, are there any recommendations about how overtly ordered a variable need to be, before one should use ordered/trend tests instead of ordinary chi squared? And whether there is preference over different trend tests (eg 1, 2, 3 etc)?