When is some number really better than no number? On the optimal choice between non-market valuation methods