«BASIC HUMAN DECISION MAKING: An Analysis of Route Choice Decisions by Long-Haul Truckers John Holland Knorring Advisor: Professor Alain L. Kornhauser ...»
ORFE Princeton University 2003
BASIC HUMAN DECISION MAKING: An
Analysis of Route Choice Decisions
by Long-Haul Truckers
John Holland Knorring
Advisor: Professor Alain L. Kornhauser
April 14, 2003
Submitted in Partial Fulfillment
of the requirements for the degree of
Bachelor of Science in Engineering
Department of Operations Research and Financial Engineering
Engineering and Management Systems Program Program in Finance Princeton University 3 Knorring ORFE Princeton University 2003 I hereby declare that I am the sole author of this thesis.
I authorize Princeton University to lend this thesis to other institutions or individuals for the purpose of scholarly research.
John H. Knorring I further authorize Princeton University to reproduce this thesis by photocopying or by other means, in total or in part, at the request of other institutions or individuals for the purpose of scholarly research.
John H. Knorring 4 Knorring ORFE Princeton University 2003 To my Mother and Father 5 Knorring ORFE Princeton University 2003 Acknowledgements I would like to thank Professor Alain Kornhauser for his help and guidance throughout this entire thesis process. His unending support and insight made this all possible.
I would also like to thank Rachel He. Without her help in working with the data in the TIDE lab, none of this would have been possible.
Lastly, I would like to thank Andrew C. Redman ’02 for his help and support of this project.
6 Knorring ORFE Princeton University 2003 1 INTRODUCTION
1.1 TRAVEL DEMAND FORECASTING
1.1.1 Trip Generation:
1.1.2 Trip Distribution:
1.1.3 Modal Split:
1.1.4 Trip Assignment:
1.2 TREATMENT OF CONGESTION IN THE FOUR-STEP APPROACH:
1.2.1 Costs of Congestion
1.3 THESIS CONTENTS
2 REVIEW OF LITERATURE
2.1 HISTORICAL PERSPECTIVE OF TRAVEL DEMAND MODELING
2.2 BASIC HUMAN DECISION-MAKING: UTILITY MAXIMIZATION
2.3 QUANTIFYING UTILITY: UTILS
2.4 THE TRUCK DRIVER UTILITY FUNCTION
2.5 FACTORS IN THE UTILITY FUNCTION
2.5.1 Income and Education
2.5.2 Number of Alternate Routes Considered
2.5.3 Risk Aversion
2.5.4 Traffic Information
2.5.5 Time of Day
2.5.7 Additional Factors
2.6 WORK DONE BY TODD BURNER ‘99
2.6.1 Value of work done by Todd Burner
2.6.2 Drawbacks to work performed by Todd Burner ‘99
2.7 PROSPECT THEORY
2.7.1 Profile of Kahneman and Tversky
2.7.2 Forecasting the Future
2.7.3 Asymmetry of Risk
2.7.4 Ambiguity Aversion
2.7.5 Prospect Theory as it Relates to this Study
2.8 OTHER SCHOLARLY TRUCK SPECIFIC STUDIES
3 EXPLANATION OF THE DATA SET
3.1 COLLECTION OF DATA
3.1.1 Time of Arrival Ranging
3.1.3 Global Positioning System
3.1.4 Inertial Navigation Systems
3.1.5 General Comments on Accuracy of Navigation Equipment
3.2 COMPLICATIONS AND DIFFICULTIES WITH DATA EXTRACTION
3.2.1 Origin-Destination Information
3.2.2 Non-Random Sample of Trucks
3.2.3 Random Time Spacing of Observations
3.2.4 Observation Reporting Error and Map Matching Complications............ 65 3.2.5 Stop Determination
3.3 ADDITIONAL DATA CONSIDERATIONS
3.3.1 Observed Traffic on Routes
3.3.2 Socioeconomic Factors
3.3.3 Characteristics of Specific Trucks
4.1 STOP DETERMINATION
4.2 DETERMINATION OF REGIONS TO BE EXAMINED
4.3 TRIP DETERMINATION
4.3.1 Route Determination Heuristic
4.3.2 The Box Algorithm
4.3.3 Drawbacks Associated with the Box Algorithm
4.4 FURTHER DATA REDUCTION USING MINIMUM AND MAXIMUM AVERAGE SPEEDS79 5 BUILDING THE MODEL
5.1 FOUNDATION FOR DISCRETE CHOICE MODELS
5.2 UTILITY MAXIMIZATION
5.3 THE LOGIT MODEL
5.4 DERIVATION OF PERCEIVED SPEEDS
6 COMMENTARY ON RESULTS
6.1 INTERPRETATION OF PERCEIVED SPEED CURVES
6.2 POINTS OF INDIFFERENCE
6.3 INTERPRETATION OF THE PARAMETERS
6.4 PITFALLS IN THE ANALYSIS
6.4.1 Small Sample Size for Specific Cases
8 KnorringORFE Princeton University 2003
6.4.2 Infrequency of Data Collection
6.4.3 Inaccuracy of Stop Definition
6.4.4 Problems Associated with the Box Algorithm
6.4.5 Overall Impact on the Model
7.1 INTERPRETATION OF THE RESULTS
7.2 AREAS FOR FURTHER STUDY
APPENDIX 1: SUMMARY STATISTICS OF TRUCKS ON ROUTES: FIRSTATTEMPT
APPENDIX 2: SUMMARY STATISTICS OF TRUCKS ON ROUTES: SECONDATTEMPT (POST DATA REDUCTION)
APPENDIX 3: C CODE FOR DATA ANALYSIS
APPENDIX 4: GPS BOX COORDINATES
APPENDIX 5: ROUTE MAPS
7.2.2 Cincinnati, OH
7.2.3 Columbus, OH
7.2.4 Houston, TX
7.2.5 Indianapolis, IN
7.2.6 Nashville, TN
7.2.7 7.2.8 Oklahoma City, OK
7.2.9 Richmond, VA
7.2.10 San Antonio, TX
St. Louis, MO
7.2.11 7.2.12 Wilmington, DE
1 Introduction Each and every day, humans make decisions. Though most decisions (such as breathing and blinking) are made subconsciously, humans often take an active role in the decision making process. Humans decide what shoes they are going to wear, what radio station they are going to listen to, and how they are going to get to work each day just to name a few. The processes that are involved in decision-making are quite complex. This thesis is will shed light on some of the more interesting decision-making processes that long-haul truckers utilize when making route choice decisions.
Every day of the year, freight and goods need to be transported around the United States. One of the most popular methods of hauling freight is via trucks. Trucks, in conjunction with a driver, are able to navigate the U.S. highway network to bring goods from where they are supplied to where they are demanded. While this process might seem quite simple, the trucker drivers are constantly faced with decisions. Should he speed up slow down? Should he change lanes or continue on his current path? If he is coming up to a city, should he stay on the current highway and potentially face congestion, or should he change lanes so that he can get on a bypass route around the city that will hopefully save him time? This last question is one that this thesis will explore.
The goal of this thesis is to perform a detailed empirical analysis of fundamental human decision-making processes by examining the behavior of truck drivers as they navigate the United States interstate highway network and to analyze the route choice decisions that they make when a number of plausible alternate routes are available. This thesis uses logistic models to describe route choice behavior when truck drivers are faced
with alternate routes. Previous empirical analyses of this nature have been plagued by data quality problems relating to objectivity issues as well as data collection methods.
However, this thesis relies on data that comes from an objective, remotely sensed revealed-preference data set consisting of over 249,465 trucks over a thirteen-day period.
Because of this, this study will be able to truly perform a detailed empirical analysis.
Once the empirical analysis is completed, this thesis will then use the results of the logistic models and analyze the decisions made by drivers to determine perceived speeds on alternate routes as well as analyze the rationality of decision-making exhibited by the long haul truck drivers. An analysis will then be done to determine how accurately the logistic models predict driver behavior and how to more accurately capture driver preferences and risk aversion. The real value of this thesis comes from the vast data set that will allow thorough enough analysis to finally compare theoretical conjectures to real world results.
One set of locations that has been a popular target for other studies is major cities that have at least two alternatives for traveling through a city. Many cities usually have a route that passes through the downtown area and a route that circumvents the downtown area, oftentimes referred to as a bypass route. When a driver is nearing such a city he is faced with a decision to make. Should he take the downtown route or the bypass route?
Each route has its strengths and weaknesses. The downtown route is oftentimes shorter than the bypass route; however, the downtown route usually has a slower speed than the bypass route due to increased traffic volume or reduced traffic capacity relative to demand on the downtown route. In addition, posted speed limits for traffic in the central business district are generally less on the downtown route than on the complimentary
bypass routes. The end result is that under light to normal traffic conditions on the downtown route, the average time spent is lower than on the bypass route, however, on the whole, the variance in travel times is greater on the downtown route than the bypass.
By investigating the decisions made by the truck drivers when they are faced with a similary, one can begin to piece together the driver’s preferences for distance and time in relation to their risk aversion.
In addition to cities that have alternate routes, due to the size and complexity of the interstate highway network, there are a number of alternate routes between cities.
Interstates 90 and 94, two of the most heavily traveled truck routes, is an example of one such area where truckers have to make a decision on which route to take. It is not obvious why every trucker would not always take the shorter route. Both routes cross areas of the northern plains states that have very little problems with congestion. This thesis, given its large data set, will attempt to explain why many truckers choose a longer route.
1.1 Travel Demand Forecasting Many would argue that present economic activity, as well as future economic growth, in any region is highly dependent on the ability to move goods and services around in an efficient manner. Transportation infrastructure plays an important role in
transportation network are able to handle not only present load factors in a manner that minimizes congestion costs, but also future load demands on the network. However, it is
also important to note that building transportation networks is very costly and time consuming, so building too much capacity into a system should also be avoided. Coming up with forecasts of future link load demands is a very complex process that includes factors such as future population growth, increases in the number of links, as well as alternative transportation methods. Taking these factors into account, planners then need to be able to understand how individuals make routing decisions in order to optimally project future link load demands. The better planners understand route choice decisions, the better their optimization models (and thus solutions) will be, resulting in a more efficient flow of goods and services in the future.
In coming up with projections of future link load demand, planners need to have a common basic methodology to solve for choices that representative travelers will make when faced with travel alternatives while also being limited by budgetary constraints.
This methodology needs to be able to determine accurate estimates for the numbers of trips on the transportation network in addition to being able to determine how the trips will be distributed throughout the network links. The most common method used in textbooks and other similar scholarly work is the four-step travel demand process.
1 The following explanation of the four step travel demand process is adopted from Norbert Oppenheim’s Urban Travel Demand Modeling: From Individual Choices to General Equilibrium and from Moshe BenAkiva’s Structure of Urban Travel Demand Models
This four-step process is commonly used in general traffic planning assessments, and it works well for our database of truck traffic behavior.
In modeling travel demand it is useful to break the entire network into smaller areas that are easier to analyze than the network as a whole. However, it is important though to keep in mind that breaking the network down into areas that are too small will provide little benefit. These smaller, more workable zones will be the basis for this study. The four-step travel demand process breaks each zone in the network into either an origin/source zone or a destination/sink zone. This thesis however, will use a slight variation of this process in that it will also include transit zones that will be used to identify the particular routes trucks take either through the central business district, or via the bypass route. These zones will be instrumental in coming up with workable solutions for the four-step travel demand process.
1.1.1 Trip Generation:
designed to generate the number of trips that originate from a given source zone. One small pitfall to this step is that it says nothing about where the trips will be going.
Planners can calculate and project numbers of trips generated from a zone via a few different methods. The first method for calculating the number of trips generated is to use observed zonal trip rates and extrapolate those numbers onto other similar zones.
While this method is somewhat crude, it does provide a fairly robust solution. In terms of projecting future link demands, a simple growth factor can be applied. The second method being used today, and growing in popularity, is the econometric modeling
method. This method is quite similar to multi-factor modeling and regression commonly