The various models and theii estimated parameters are
presented in Table 1. As the table reveals, all of the
estimated parameters are of the expected sign and are of
reasonable magnitude. The estimates are generally
sign&ant and the R* is quite high, although using time
series data tends to lead to overstatement of these
measures of statistical confidence.
The estimates produced by iterated least-squares on a
log-linear model (Model 3 in Table 1) are generally
similar to those of the Gauss-Newton method. The standard
errors of the regression coefficients for trackage and
output are generally lower for Model 3. The economic
interpretation of this estimate of “a” is that it costs
roughly $110,00 per mile of single track annually to
operate a rapid transit system without any &a&. These
are the “fixed” components of maintenance and other
operating costs and do not include any capital or depreciation
costs.
The coefficient of output (6,) is the elasticity of shortrun
variable costs with respect to increases in output and
is greater than one in most of the models. Thus, our
model implies that variable operating costs rise more
than proportionately with increases of output in the short
run. There are economies of density, however, since the
fraction of costs variable was found to be less than one
for all of the properties. That is, more intensive use of
the fixed facilities reduces unit costs by spreading the
fixed components of maintenance and operating costs
over more units. At the current levels of output for the
properties in the sample, variable costs would not rise