In a search for an appropriate model specification, various MNL models with different combinations of generic and alternative specific taste parameters for cost, time, quality and flexibility were analysed. A consistent pattern revealed during this exploratory work was that of surprisingly weak significance in the parameters associated with cost and travel time attributes, compared with generally strong significance in the parameters associated with quality and flexibility. A similar finding was reported in the original analysis performed on these data by Norojono and Young (2003) where, based on their HEV model specification, they noted that the cost and time parameters were typically only significant at the 10% and 20% level respectively. In general terms the explanation for this finding may be either substantive or methodological. That is, either Indonesian shippers truly are relatively insensitive to time and cost (at least over the range of variation presented in the stated preference exercise) or some aspect of the stated preference exercise and/or the model of analysis of the data is generating
this outcome as an artefact. Norojono and Young (2003) did not pursue this point empirically in their paper instead hypothesising that the low significance of these parameters was likely to be due to the limited range of variation presented in the cost and time attributes in the stated preference exercise (20% higher to 10% lower) being insufficient to trigger significant
trading behaviour around times and costs in this sample. Norojono and Young’s hypothesis is plausible but difficult to corroborate directly, since strictly speaking this would require
knowledge of the true underlying population parameter values. An alternative approach which we decided to follow in this paper is to investigate alternative potential explanations for the relative insignificance of the time and cost parameters, to corroborate or refute these potential explanations. To this end, a number of diagnostic analyses were undertaken:
Colinearity: The attribute data were checked for potentially problematic levels of correlation between cost, time and other attributes. The cost-time correlations for small truck, large truck and train were found to be 0.0242, 0.0216 and 0.3565
In a search for an appropriate model specification, various MNL models with different combinations of generic and alternative specific taste parameters for cost, time, quality and flexibility were analysed. A consistent pattern revealed during this exploratory work was that of surprisingly weak significance in the parameters associated with cost and travel time attributes, compared with generally strong significance in the parameters associated with quality and flexibility. A similar finding was reported in the original analysis performed on these data by Norojono and Young (2003) where, based on their HEV model specification, they noted that the cost and time parameters were typically only significant at the 10% and 20% level respectively. In general terms the explanation for this finding may be either substantive or methodological. That is, either Indonesian shippers truly are relatively insensitive to time and cost (at least over the range of variation presented in the stated preference exercise) or some aspect of the stated preference exercise and/or the model of analysis of the data is generatingthis outcome as an artefact. Norojono and Young (2003) did not pursue this point empirically in their paper instead hypothesising that the low significance of these parameters was likely to be due to the limited range of variation presented in the cost and time attributes in the stated preference exercise (20% higher to 10% lower) being insufficient to trigger significanttrading behaviour around times and costs in this sample. Norojono and Young’s hypothesis is plausible but difficult to corroborate directly, since strictly speaking this would requireknowledge of the true underlying population parameter values. An alternative approach which we decided to follow in this paper is to investigate alternative potential explanations for the relative insignificance of the time and cost parameters, to corroborate or refute these potential explanations. To this end, a number of diagnostic analyses were undertaken:Colinearity: The attribute data were checked for potentially problematic levels of correlation between cost, time and other attributes. The cost-time correlations for small truck, large truck and train were found to be 0.0242, 0.0216 and 0.3565
การแปล กรุณารอสักครู่..
