Credence goods have the characteristic that a customer does not know which type of treatment he needs, while an expert seller is able to observe the needed treatment. A series of problems arise from the informational asymmetry associated with credence goods. For instance, when a car mechanic repairs your car, he might have an incentive to cheat on you in two dimensions: First, the repair itself might be inefficient by either overtreatment (replacing more parts than needed) or undertreatment (failing to do a full repair). Second, the repair might be sufficient, but the mechanic might charge for more than he has actually done, thereby overcharging you.
Several empirical studies provide compelling evidence that the provision of credence goods can and does, indeed, suffer from serious problems like under- and overtreatment or overcharging. However, the existing papers suffer from the lack of a controlled variation of factors that influence the efficient provision of credence goods. All empirical papers either provide evidence that overtreatment is, in fact, happening, without systematically exploring the conditions leading to overtreatment. Or they vary only one particular aspect that influences the provision of credence goods without controlling for and varying other important factors (like liability or verifiability) (see Wolinsky, 1993; Emons, 1997; Gruber et al., 1999).
We present a controlled laboratory experiment where we carefully vary one by one the factors that have been identified as crucial for the efficient provision of credence goods, i.e. liability and verifiability (Dulleck and Kerschbamer, 2006). We find that liability (to provide sufficient treatment) is crucial for the efficient provision, whereas verifiability (of the treatment provided) does not lead to full efficiency, contrary to the theoretical prediction. We discuss several implications for economic policy and the organization of credence goods markets.