If, as is reasonable to expect, ftxed operating costs
increase with the size of the property, we expect oO>
o> as> 0,.
It is also possible to allow the other parameters of eqn
(1) to vary among the rapid rail systems. Unfortunately,
the statistical method was already too strained to m*e
this extensive test on our limited sample.
In addition to problems of heterogeneity of the sample,
we have problems of heteroscedasticity, in that the residuals
(the difIerence between fitted and actual values)
are correlated with T, the scale of the system. Corrections
for heteroscedasticity were made by dividing
through both sides of the equation by F, raised to various
powers. The equation was then reestimated and the
absolute values of the residuals for a number of the
smallest firms and a number of the largest tirms were
compared. We used as a criterion for a “best” correction
of heteroscedasticity that the two groups of residuals
differ least signi6cantly (by a Mann-Whitney U-Test).
The deflator p” was found to be the best correction for
the models reported here.