Can we get by with forms of validation that are weaker than experiments? It depends on what question we're asking, but here are some excuses that I find less than satisfactory.
