Time series variables for the number of transports requested
during each hour of the year were generated from observations
from 1994 and the first half of 1995. Univariate exponential
regression of yearly run volumes from the years 1986 to 1994 was
used to predict total 1995 run volume. Each observation of the 1994
run volume series was multiplied by the yearly volume ratio
(predicted 1995 divided by observed 1994) to adjust for the
estimated increase in total yearly census: adjusted transports/hour =
observed transports/hour • (74,151/68,433). The adjusted observations
from all 52 full weeks of 1994 (n = 8,736 hours) were used to
build the forecasting models. Data from the first 24 complete weeks
of 1995 (n = 4,072 hours) were not used for forecasting, but were