@amoeba Ah! That seems perfect!

@Glen_b @gung @usεr11852 To be honest, I did not expect this issue to be so controversial. Among the people who explicitly stated their opinion, we now have 3:3 split.

In the worst case, we will get stuck on this and dummy-variables will remain the synonym of categorical-data, which does not make any sense now that we have categorical-encoding. It's almost like obamacare repeal!

@Glen_b I was a bit surprised to see your support of having dummy-variables as a stand-alone tag. You wrote "mostly out of considerations of how people will actually use the tags -- we have similar situations with a number of tags in the regression-constellation". Do you mean that some of the regression-related tags are strongly overlapping? Yes, and I consider it very suboptimal. Why proliferate such situations?

I also think that gung's comment there is spot on: if there is a synonym mapping, then whatever terminology people prefer, everything gets mapped to one master tag. Looks neat to me.

@amoeba I agree that dummy-variables as a synonym of categorical data is suboptimal. We discussed the regression situation before and I recall disagreeing with you on it. I don't think it's possible, nor even desirable to always avoid overlap in categories; that's just how conceptual categories operate. Even when you can outline a reasonable logical hierarchy for some set of tags, it's not always feasible to expect people to work that way

because of entrenched usage differences and partial understanding that we have to deal with as a practical matter.

That's not to suggest that it's impossible to rationalize tags (clearly you've done a great deal along those lines, to the benefit of all of us), but that in some cases overlapping tags are simply going to be a fact of life. I'd be very happy to have you tell me why I'm wrong on this one though - I'd gladly repudiate my comments if I'm wrong.

Anyone can tell me if this approach is valid?

need to ask about error minimization.