ABSTRACT
In many applications, the support region (the region over which a vari ate’s density is nonzero) of the variate being studied must be bordered on the left by the value zero. Thus the superiority of models (2 ) and (3) over model (1 ) can be due either to the better fit of these models over all support points or, alternatively, due to models (2) and (3) being in accord with prior knowledge concerning the zero-bordered support region. Because model (3) is a more complex curve representation than model (2) which, in turn, is more complex than model (1 ), were either the parameter r of model (2) or the parameters 7 1 and 7 2 of model (3)—estimated-rather than constrained to equal zero, then model (3) would be found typically to fit data better than model (2 ), while model (2 ) would almost certainly fit better than model ( 1 ).
