When I saw this, I found myself wondering if this were the documentation bug again or not? It could be a trick question, where the "correct" answer was the only answer provided that would run teh query with the old cardinality estimator, so the poor practice answer (use compatibility level) would be "correct" despite being poor practice for dealing with a single query rather than all queries in a DB. After all, it was just over a fortnight since STeve's last question on this topic and the documentation error was thoroughly aired then, surely he would have correted this one by now?

Anyway, in the end I decided it was probably a repeat of believing teh wrong BOL page and got it right by selecting an option that wouldn't actually work.