«BASIC HUMAN DECISION MAKING: An Analysis of Route Choice Decisions by Long-Haul Truckers John Holland Knorring Advisor: Professor Alain L. Kornhauser ...»
used in modern time series analysis. One example of a model might be:
Where X1 might be the average population density in the zone, X2 might be the distance to the central business district, X3 might be the average household income, and, X4 might be a seasonality component. While this method has some obvious advantages in that it is much more descriptive of the representative zone, the complexity of the analysis increases enormously when considering a network as large as the United States interstate highway system.
It is important to keep in mind that the trip generation phase is fairly limited. It only produces characteristics of the respective zone and traveler attributes. All characteristics relating to trip destinations, mode choice, and route choice are determined in later stages. The easiest way to conceptualize the trip generation stage is to think of the zones and then determine how many trips will be leaving the respective zone.
1.1.2 Trip Distribution:
The trip distribution phase of the travel demand forecasting process is a one-way stage that is typically executed after the trip generation stage. The inputs to this stage are the outputs from the trip generation phase. Trip distribution typically finds and allocates
a destination for each of the trips generated at source nodes. The most common method for determining trip distribution is the “Gravity Method.” This method assesses the accessibility and attractiveness of destination zones for each source zone and distributes proportionally the trips from the trip source zones to the destination zones.
1.1.3 Modal Split:
The next stage that transportation planners follow is the modal split stage. This stage determines the mode split that the representative travelers will use to get from their respective origins to their destinations. In other words, this stage splits the travelers into groups that will be driving cars, other groups that take busses, groups that take trains, and groups that fly, for example. Once the planner has ascertained the appropriate levels of modal splits, he can then determine the number of trips taken by road from one region to another and thus plan accordingly. Modal splits are determined similarly to trip distribution. Typically a planner will use a diversionary curve or S-shaped logistic curve to determine modal split.
1.1.4 Trip Assignment:
Trip assignment is the fourth step in the traditional four-step approach. This is the stage where planners attempt to determine the routes that representative travelers will take between their respective origin-destination pair. This phase will show how all of the travelers distribute themselves over the links in the network. There are many ways to derive trip assignments. Some models use a shortest path constraint, however this is not
always the optimal solution. A common constraint found in most models is that no single traveler should be able to reduce his travel time by switching to another series of links.
This constraint is known as the Wardrop user equilibrium condition. The result of this condition is that the model must be able to find alternate paths that a driver can take, and it must be able to determine if any travelers will take the alternate route. In other words, the model must examine routes that are longer than the shortest path solution and determine if the time spent on the alternative route is less than the time spent on the shortest path route.
1.2 Treatment of congestion in the four-step approach:
In the four-step approach to modeling travel demands, congestion rarely is a factor in any calculations. The end result of this is that the final solution to the four-step process might not be optimal for all situations. Congestion becomes increasingly important in factors where link load costs and destinations are functions of link demands.
The most obvious example of this is when link loads approach capacity and traffic rates are slowed down significantly. The result is the time spent traversing a link increases to a level where the benefit to following that link would be decreased significantly, thus resulting in a reduced demand for the specified link. The end result is a vicious feedback loop that is very difficult to model and the four-step process becomes a highly complex iterative algorithm. Stability in the models is oftentimes difficult to achieve with congestion.
1.2.1 Costs of Congestion When one thinks about the last time that he was in a traffic jam, most likely bad memories come to mind due to the unnecessary costs involved with sitting in traffic.
Papacostas and Prevedouros assert that there are a number of costs associated with congestion including: loss of productive time, increased fuel usage, increased levels of atmospheric pollution, increased engine and mechanical wear, increased frequency of traffic accidents, and the negative psychological and emotional impact of sitting in traffic culminating in “road rage.”2 In order to minimize these negative byproducts of congestion, the government has introduced legislation and has created guidelines for monitoring efficiency in the highway. The first step in this process began with the passage of the Federal-Aid Highway Act of 1962. According to the U.S. Department of Transportation’s Travel Model Improvement Program, this act was designed to improve the U.S highway network so that users would have access in all weather conditions. Additionally, the act stipulates that cities and urban areas with populations of 50,00 or more must provide a long-range
Transportation Efficiency Act of 1998 (TEA-21), among others, has put the responsibility for long range planning into the hands of the locally established Metropolitan Planning Organizations (MPO) established throughout the United States in areas with populations greater than 50,000 in an effort to increase flexibility in the design and implementation process.3 2
Papacostas, C.S. and P.D. Prevedouros, Transportation Engineering and Planning, 2d ed United States:
Prentice-Hall Inc., 1987, pg 258.
1.3 Thesis Contents This thesis consists of an analysis of truck driver decision-making gathered from a large revealed-preference data set consisting of over 60,000,000 location records. The analysis will be done to determine the effects of availability of alternate routes in decision-making processes.
Chapter II consists of an analysis of current scholarly work in the area of route choice. The chapter will focus on other studies and the current theory on truck driver behavior. This section will contain a historical perspective on route choice modeling and basic decision modeling along with the challenges others faced with empirical studies.
This section will also include a discussion of Daniel Kahneman’s Nobel Prize winning Prospect Theory. Because this thesis is primarily focused on building on the work that Todd Burner ’99 did for his senior thesis, there is an extensive review of his thesis included in this section.
Chapter III consists of the data that is used in this analysis. This section will include commentary on the size and relevance of the data set along with a discussion on the importance of the objectivity and remotely censored data. In addition, methods of collecting the data, basic formatting of the data, and derived characteristics of the data are discussed. Many of the definitions to be used throughout this thesis as well as drawbacks to the data set will also be covered in this chapter.
Chapter IV consists of the methodology for extracting the data. This section includes the summary statistics of the number of trips and number of observations. In
addition, this section contains analysis on average speeds maintained by the truckers.
Lastly, this section covers the algorithm used for selecting the appropriate trucks.
Chapter V consists of the building of the Logit Model. This includes remarks on notable statistics as well as how assumptions in the definitions and calculations may affect the resulting summary statistics.
Chapter VI consists of commentary on model fitting. This section will cover how logistic models are used and what can be gained from their use. Additionally, a number of case studies are examined using the results of the logistic models.
Chapter VII consists of concluding comments about the results of the study along with a section on areas that deserve further study.
In addition to eight chapters, there are two appendices. The first appendix is a central grouping of the output data, models, and statistics. The second appendix is a more in-depth look, including road maps, at the case studies covered in this thesis.
2 Review of Literature Any scholarly research must begin with a review of related work done by others.
It is necessary to review the work of other individuals so that value can be added and research is not simply repeated. First, background information on the current theories, for the sake of the reader, must be examined in order to set a workable foundation for critical analysis. As stated previously, this thesis is primarily an extension of work done by Todd Burner ’99. Therefore, it is essential to review the work that he did and access what is of value in his work to this thesis. Additionally, it is important to seek out areas where further analysis needs to be done to better set a foundation for this thesis.
2.1 Historical perspective of travel demand modeling There has been much scholarly work done concerning travel demands and individual driver route choice decision-making. Most agree that modern travel demand analysis began in the late 1930’s as concern with building and planning for new roads increased in lock step with the increasing number of automobiles. Most of the early work was fairly limited and more focused on theoretical considerations due to the limited computing resources available. The focus of the early work was also on more aggregate levels of travel demand than on disaggregate individual behavior modeling.
After sorting through the vast amount of research work done on traffic demand modeling, one name that keeps reappearing in most of the research on both aggregate and disaggregate travel demand and behavior modeling is professor of Civil and
Environmental Engineering at the Massachusetts Institute of Technology, Moshe BenAkiva. Professor Ben-Akiva has published a wide array of work in the field of transportation. One of his more pertinent studies relating to this thesis is his doctoral thesis Structure of Passenger Travel Demand Models. In his thesis, he covers a wide range of modeling options for transportation demand. The two primary models that he uses are logit models and probit models. The logit models are simple logistic models that attempt to predict aggregate probabilistic behavior of populations. The probit model that he focuses on is much more complex.
In probit models, joint probabilities are used to model choices. These models try to associate a probability to every decision combination that could be made from the time one decides to travel until the time an individual reaches his destination. Because these models attempt to quantify all decisions, the computation complexity is increased by many factors.4 One of the major weaknesses of the Ben-Akiva’s method is that his empirical study is based on household surveys. His study modeled modal choice by individuals when they go shopping. Household surveys do not accurately reflect human decision making for a number of reasons. First, there are data quality issues that arise when surveys are given. The surveyed individual can lie on the survey, or at least deliberately misinform the surveyor. The surveyed individual also might not know exactly which mode they took at some time in the past and they might guess incorrectly which method they took. Next, by surveying households, one has the issue that by observing the system, one may have disturbed the system. In other words, a surveyed individual might change or have changed their behavior because of the data collection that was being 4 Ben-Akiva, Moshe, Structure of Passenger Travel Demand Models, MIT, 1973, pp 14-17.
done. However, given the level of technology for collecting arms length data at that time, Ben-Akiva’s study provides a useful base of study for this thesis. Going forward with this thesis, access to remotely sensed, arms length data will allow for a more accurate analysis to be performed.
2.2 Basic human decision-making: Utility maximization Every day individuals are faced with a series of decisions to make on how to allocate resources. One of the most interesting resources that people allocate is time.
The concept of time is very interesting because of its non-fungible nature; each person gets a fixed number of hours every day and that amount of time is the same for every person. It is impossible to trade hours with another user, and it is also impossible to change the rate at which the resource is used. While some associate time allocation with a simple newsvendor problem with costs of overage and costs of underage, in the more general case of decision-making, time allocation really is dependent on utility maximization. In the context of transportation, time is especially interesting because of the fundamental nature of time in evaluating destination, mode, and route alternatives.5 Most scholarly quantitative work regarding decision making for travel demand forecasting, transportation planning, and congestion management is based on the concept of utility maximization. The process of making route choice decisions is quite complex and there are a number of factors that are hard to quantify. Utility maximization is used to study travel demand forecasting because it is the most plausible method of evaluation. It 5 Lisco, Thomas E., “Common Economics of Travel Time Value”, Transportation Research Board Special Report 149. Washington, D.C., 1974, pp 103-115.