Model estimation. With missing levels of transport quality replaced by imputed values ( ~x ink ), we used maximum likelihood in LIMDEP to simultaneously estimate the linear-in-parameters utility function for each mode. For the binary logit model considering only truck and rail alternatives, the model rejects 38 cases in which the shippers chose the ‘others’ alternative, resulting in 280 remaining observations in the estimation (Table 6.7). The alternative-specific constant for rail is negative, which implies that holding all other attributes constant (e.g., travel time, price), shippers prefer the truck-only alternative over rail-based alternatives. The parameter estimate for the interaction between travel time and the value per tonne implies that the higher the value per tonne of the shipment, the more sensitive the shipper is to travel time. Travel time is one of the highly ranked quality dimensions that also have a large discrepancy between shippers’ ranking of importance and satisfaction (see Section 3.1). Below 70 hours of travel time, we find that rail is more negatively affected than truck by an increase in this interaction term. To explain this, one needs to realise that the type of shipments modelled are primarily in the general cargo segment, which are overall of high-value and thus of relatively high priority. While on average the rail-based solution takes a longer time than truck, it is a higher probability for rail than for truck that a unit increase in lead time or value leads to transport quality below the critical level where the mode isn’t an alternative. Above 70 hours, truck and rail alternatives are equally, negatively affected by increases in travel time; however, it is not significant at the 80% level. It is reasonable to believe that shippers perceive travel time indifferently above a certain level, since it will take a long time no matter what service they choose, and therefore land-based transport with lead time above 70 hours is primarily used for shipments that are less sensitive to increases in lead time. Since shippers pay upfront out-of-pocket costs for transport services regardless of mode, it is reasonable to assume that an increase in unit price will be perceived to be equally onerous across alternatives. Thus, for the truck- and rail-based alternatives, we used a common estimate for the parameter corresponding to the model variable of price over value. We divide price per tonne by value per tonne (thus cancelling the tonnes) to again accoun...
Model estimation
