[BlueBear]

Ok, using your hypothetical example, let X, Y be age of youngest, oldest partner respectively. We need to estimate parameters a, b and c where

Income = aX + bY + c(X-Y)

I don't think this a multicollinearity problem, but you'll just have a completely redundant variable in your multiple regression model.

If you reshuffle the terms above, you'll just simply get

Income = (a+c)(X) + (b-c)(Y) = AX + BY

So you could do a standard multiple regression model using 3 variables (to get variables a,b,c) and you could also do a similar model using 2 variables to get (A,B). They'll be exactly the same due to a redundant variable.

So in other words, it's a redundant variable problem instead of a multicollinearity problem.

I'm not a stats guru BTW.