the expected sign and is significant at the 10% level in a
one-tail test. Spending on maintenance does obviously not
respond to variations in traffic load in the same way as in the
Swedish data.6 The parameter estimates for the quality
proxy (speed) are significant and display the same pattern as
INDX for the Swedish data and the parameters for number
of switches are both highly significant, indicating an
increasing cost at a decreasing rate; the magnitude of that
impact is, however, small. Moreover, Finland has a
significantly negative time trend for costs, indicating costsavings
of 10% or more between 1997 and 1998 or 1999,
respectively.
The elasticity of costs relative to track length7 (i.e.
hy
it ¼ ›ln Cit=›yit) is plotted against Gtkm in Fig. 4. The
sharp decline in elasticity for low values of Gtkm is driven
by the non-electrified lines while elasticities fall slowly with
the traffic load on electrified lines. The similar structures of
Figs. 3 and 4 indicate that non-electrified lines are
tantamount to what was referred to as secondary lines in
the Swedish material. These are the lines where it has not
been deemed worthwhile to invest in electrification because
of a low traffic flow.
The mean values of the respective elasticities have been
calculated and are summarized in Table 6. The overall mean
hy is 0.63 which is lower than in the Swedish model.
Because of large standard errors, we cannot conclude
(statistically) that the means are below one. Although the hu
are estimated with low precision, it is clear that they are
below unity with similar implications as in the Swedish
material. In particular, the magnitude of the respective
elasticities is very similar.