and require a modeling framework for combining them. To illustrate this point consider the example of comparing the door to door travel time between two multimodal systems, a high speed rail with auto access to stations and an air transport journey with transit access to the airports. To simply add up the travel time components for each of these multimodal systems and then compare them ignores important differences in the incidence of these various components and their contribution to the disutility of travel. It ignores the different impact of access time and line-haul time. It also ignores the complex question of reliability and the variance in travel time. It therefore misrepresents the relative performance of these two alternatives and can lead to invalid conclusions and possibly wrong forecasts.
Even the most obviously additive measure, out-of-pocket cost, which is uniformly measured in money terms and can be simply added for the components of a multimodal service, can sometimes behave in a non-additive manner when the different elements of cost have different tax implications, or subsidy potential. An example of this could be line haul costs that may be tax-deductible or subsidized by employers but terminal costs (parking or access to station) that may not be. Or different cost elements that can be paid at different times thereby having different impact of the user’s budget. The challenge for dealing with multimodal measures of performance is how to address the issue of non-additive metrics. Some of these are separable in that they may be combinable with some linear function. An example would be line-haul time and access time combined according to incidence parameters that are estimated from a discrete choice model. Others are non-separable in that they combine in nonlinear ways and require complex functions. An example of this can be found in reliability and safety measures that reflect the performance of the weakest link of the multimodal system. Another example comes from the various measures that reflect variance of individual metrics requiring the nonlinear combination of standard deviations. The detailed description of the various measures of level of service is in the following session.
3.3.1 Travel Time The two main components of the cost of traveled that is incurred by individual users are: time and money. We begin with travel time. The measurement of travel time is fairly easy, but the difficulty is in quantifying user’s perception of time especially as it varies between the different components of the total travel time. . On the individual level, the evaluation of perceived travel time is based on the definition in the beginning of this section: access, waiting, transfer, and in-vehicle traveling. IN VEHICLE TRAVEL TIME The weight of in-vehicle travel time of different modes in total perceived travel time depends on the overall level of service provided by the mode, such as level of comfort, sense of security, etc. For example, the perceived travel time of taking a bus is usually longer than driving, for the same time of actual time spent. Therefore, again, these can be interrelated or even overlapping. In addition to adding the different components of time together with appropriate weights, it is not unusual to define weights that reflect some attributes that affect the perception of time, such as comfort, and security. But that would require detailed information as might be obtained from in