USE OF BOX-COX MODELS FOR TESTING WAGNER HYPOTHESIS - A CRITICAL NOTE

This paper suggests that conventional constant-elasticity regression s pecifications provide a better format for testing Wagner's ''law'' of expanding public sector than a Box-Cox alternative. Four major conside rations are brought out. First, year-to-year variability of elasticity is not relevant to Wagner's hypothesis; what is needed is the ''avera ge'' elasticity for the entire period, and an estimate of that is dire ctly given by the conventional models but not by the Box-Cox formulati ons. Second, calculation of an average from the Box-Cox estimates invo lves an ad hoc procedure, and no statistic is available for formal hyp othesis tests. Third, the non-linear Box-Cox models are more complex a nd computationally more expensive. Most important, a comparison of the Box-Cox estimates with those obtained from conventional models indica tes that the Box-Cox results are less satisfactory in many cases.