I was thinking one way to apply the test would be to base it on the final results. However, strictly speaking, what we're interested in is elections where the voters felt the election was important. This can actually also be measured by the amount of press coverage of an election (which can be objectively measured, it's just harder, especially in the past.)

However, I do also feel that the closeness of the final result is indeed a reasonable proxy for detecting contested races. Probably things can be learned by examining the undervote (ballots cast which did not include a choice on a particular race.) Contested races will have a low undervote and may actually increase the undervote on lesser races.