The best models come from starting with an explanation of the least common common occurrence in a body of occurrences and then working inwards. People who do the opposite are not really explaining anything, they're just betting on "house" in a gambling den.
But then I suppose the first condition of the former is the last condition of the latter, and thus it ultimately depends on your intuition for what the "least common common occurrence" is. I just think based on experience that the former is tied to stronger intuition I guess.