Testing the assumptions behind importance sampling