Lane Changing on Freeways

MISSING IMAGE

Material Information

Title:
Lane Changing on Freeways
Physical Description:
1 online resource (159 p.)
Language:
english
Creator:
Hill, Corey A
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Master's ( M.E.)
Degree Grantor:
University of Florida
Degree Disciplines:
Civil Engineering, Civil and Coastal Engineering
Committee Chair:
Elefteriadou, Ageliki L
Committee Members:
Washburn, Scott S
Yin, Yafeng

Subjects

Subjects / Keywords:
cluster -- congested -- discretionary -- lane-changing -- mandatory -- uncongested
Civil and Coastal Engineering -- Dissertations, Academic -- UF
Genre:
Civil Engineering thesis, M.E.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
Lane changing models are a significant component of microscopic traffic simulation. Understanding the details of this fundamental maneuver is important for accurate modeling in simulation. Therefore, lane changing has received much attention. Many studies have focused on the details of the lane change maneuver from external observation based data without regard for the type of driver performing the maneuver. In this thesis, the physical details of freeway lane changing have been related to the   type of driver performing the maneuver.  46 research participants drove an instrumented vehicle and performed a combined total of 726 freeway lane changes. A cluster analysis was performed to categorize each research participant into one of four groups ranging from conservative to aggressive. Then an analysis was done to determine any trends that existed between the different driver types and their lane changing characteristics, specifically lane change duration and gap acceptance characteristics. It was found that, in general, more conservative drivers have greater lane change durations than aggressive drivers. The gap acceptance comparison among driver types did not yield any conclusive trend.  In addition, distributions were fitted to lane change duration and gap acceptance histograms. Also, hypothesis testing was used to determine if significant differences occur for lane change durations and accepted gap sizes for different lane change types and congestion conditions. The results of this thesis suggest that driver types do have a significant role in the details of freeway lane changing. Therefore, this relationship should not be disregarded when developing lane changing models.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Corey A Hill.
Thesis:
Thesis (M.E.)--University of Florida, 2012.
Local:
Adviser: Elefteriadou, Ageliki L.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2013-02-28

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2012
System ID:
UFE0044750:00001


This item is only available as the following downloads:


Full Text

PAGE 1

1 LANE CHANGING ON FREEWAYS By COREY HILL A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DE GREE OF MASTER OF ENGINEERING UNIVERSITY OF FLORIDA 201 2

PAGE 2

2 2012 Corey Hill

PAGE 3

3 To Kim

PAGE 4

4 ACKNOWLEDGMENTS I would like to thank Dr. Lily Elefteriadou for being the most influential individual during my time in the Transportation Graduate School Program at UF Since the beginning she was always kind a nd encouraging in the way she provided me instruction and suggestions. Her knowledge in the field of transportation engineering was valuable and she created an environment in which questions could be asked freely. In addition, she provided me with very use ful resources throughout the work of my thesis and always made the time to meet wi th me amongst her busy schedule. I would also like to thank Dr. Scott Washburn and Dr. Yafeng Yin for serving as part of my m rn for the students has been evident since I began the graduate program. I also enjoyed knowing them as professors who shared their detailed knowledge of specific areas of transportation engineering through their courses. They, along with the rest of the f aculty, have made the transportation graduate school at UF an esteemed program. A special note of thanks should be made to Clark Letter, a colleague and friend. He made trips to Orlando during the data collection phase of the research enjoyable and was al ways generous in allowing me to bounce ideas off of him concerning details of my thesis. Thank you to my family and friends who were very supportive throughout the process of writing my thesis. They were always there to encourage me to persevere and keep a balanced perspective throughout the process.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ ........ 12 ABSTRACT ................................ ................................ ................................ ................... 14 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 16 Problem Statement ................................ ................................ ................................ 16 Research Objectives ................................ ................................ ............................... 18 Determine significant differences between different lane change types during congested and uncongested conditions. ................................ ............ 18 Determine lane change duration and lag gap acceptance distributions. .......... 19 Classify driver s based upon their desired speed and the number of discretionary lane changes completed per mile. ................................ ........... 19 Investigate the relationship between different driver types and their lane changing charact eristics. ................................ ................................ ............... 20 Thesis Outline ................................ ................................ ................................ ......... 20 2 LITERATURE REVIEW ................................ ................................ .......................... 23 Lane Chang ing Models ................................ ................................ ........................... 24 Lane Change Duration ................................ ................................ ............................ 32 Lane Changing Models used in Simulation ................................ ............................. 35 AIMSUN ................................ ................................ ................................ ........... 35 CORSIM ................................ ................................ ................................ ........... 36 SITRAS ................................ ................................ ................................ ............ 39 Gap Acceptance ................................ ................................ ................................ ..... 42 Summary ................................ ................................ ................................ ................ 48 3 DATA COLLECTION ................................ ................................ .............................. 57 Instrumented Vehicle ................................ ................................ .............................. 5 7 Research Locations ................................ ................................ ................................ 58 I 4 Orlando, FL ................................ ................................ ................................ 58 I 95 Jacksonville, FL ................................ ................................ ......................... 58 Research Participants ................................ ................................ ............................. 59 Data Extraction ................................ ................................ ................................ ....... 60 Lane Changes ................................ ................................ ................................ .. 61 Desired Speed ................................ ................................ ................................ .. 62

PAGE 6

6 Gap Acceptance ................................ ................................ ............................... 62 Qualitative Video Data Observations ................................ ................................ ...... 63 Double Lane Changes ................................ ................................ ...................... 63 Explicit Target Lane Choice ................................ ................................ .............. 64 Avoiding On ramp Merging Vehicles ................................ ................................ 64 Interactions with Heavy Vehicles ................................ ................................ ...... 64 Summary ................................ ................................ ................................ ................ 65 4 DATA ANALYSIS AT AN AGGREGATE LEVEL ................................ .................... 70 Hypothesis Testing ................................ ................................ ................................ 70 Distributions of Lane Change Durations and Gaps ................................ ................. 71 Distributions of Lane Change Durations Conclusions ................................ ...... 72 Distributions of Lead and Lag Gaps Conclusions ................................ ............. 73 Summary ................................ ................................ ................................ ................ 74 5 DATA ANALYSIS BY DRIVER TYPE ................................ ................................ ..... 79 Driver Classification Overview ................................ ................................ ................ 79 Cluster Analysis ................................ ................................ ................................ ...... 80 K means Clustering ................................ ................................ .......................... 80 Within Cluster Sum of Squares ................................ ................................ ........ 81 The Hartigan Index ................................ ................................ ........................... 81 Cluster Analysis for Orlando and Jacksonville Combined Data .............................. 82 K means Clustering for Combined Data ................................ ........................... 83 Interpreting the Results of the Combined Data Cluster Analysis ...................... 84 Cl uster Analysis for Orlando and Jacksonville Separated Data .............................. 85 K means Clustering for Separated Data ................................ ........................... 85 Determining the Number of Driver Groups ................................ ....................... 86 Survey Results to Cluster Analysis Comparison ................................ .............. 86 Analysis of the Driver Groups ................................ ................................ ........... 87 Lane change duration by driver type ................................ .......................... 87 Gap acceptance size by driver type ................................ ........................... 88 Summary ................................ ................................ ................................ ................ 90 6 CONCLUSIONS ................................ ................................ ................................ ... 107 Research Conclusions ................................ ................................ .......................... 107 Recommendations ................................ ................................ ................................ 109 Future Research ................................ ................................ ................................ ... 110 APPENDIX A RESEARCH PARTICIPANT FORMS ................................ ................................ ... 112 B MEASU RING FREEWAY TRAFFIC STREAM GAPS USING VIDEO DATA ........ 115

PAGE 7

7 C LANE CHANGE DURATION AND GAP ACCEPTANCE DESCRIPTIVE STATISTICS ................................ ................................ ................................ ......... 117 D HYP OTHESIS TESTING ................................ ................................ ...................... 125 Uncongested DLC Durations vs. Congested DLC Durations ................................ 125 Uncongested MLC (merge only) vs. Congested MLC (me rge only) ...................... 127 Uncongested MLC (no merge) vs. Congested MLC (no merge) ........................... 129 Uncongested MLC (no merge) vs. Uncongested DLC ................................ .......... 131 Congested MLC (no merge) vs. Congested DLC ................................ ................. 133 Right DLC vs. Left DLC ................................ ................................ ......................... 135 Uncongested Lag Gap vs. Congested Lag Gap ................................ ................... 137 Summary of Hypothesis Testing Findings ................................ ............................. 138 E LANE CHANGE DURATION AND GAP A CCEPTANCE STATISTICAL SUMMARIES ................................ ................................ ................................ ........ 139 Uncongested Discretionary Lane Change Duration Statistical Summary ............. 139 Congested Discretiona ry Lane Change Duration Statistical Summary ................. 140 All Discretionary Lane Change Durations Statistical Summary ............................. 141 All Lane Change Durations Statistical Summary ................................ .................. 142 Discretionary Lane Change Lag Gap Statistical Summary ................................ ... 143 Discretionary Lane Change Lead Gap S tatistical Summary ................................ 144 Lag Gap Statistical Summary ................................ ................................ ............... 145 Lead Gap Statistical Summary ................................ ................................ ............. 146 Uncongested Lag Gap Statistical Summary ................................ ......................... 147 Congested Lag Gap Statistical Summary ................................ ............................. 148 F DRIVER GROUPS DE SCRIPTIVE STATISTICS ................................ ................. 149 G MEAN LANE CHANGE DURATION COMPARISONS OF DRIVER GROUPS .... 151 H MEAN LAG GAP ACCEPTED IN CONGESTED CONDI TIONS COMPARISONS OF DRIVER GROUPS ................................ .............................. 153 LIST OF REFERENCES ................................ ................................ ............................. 155 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 159

PAGE 8

8 LIST OF TABLES Table page 2 1 Summary of Literature on Lane Changing Models ................................ ............. 54 2 2 Summary of Lane Change Duration Literature ................................ ................... 55 2 3 Summary of Lane Change Gap Acceptance Literature ................................ ...... 56 4 1 Fitted Distribution for Different Lane Change Duration Data Gr oups .................. 75 4 2 Fitted Distribution for Different Lane Change Gap Data Groups ........................ 75 5 1 Summary of Orlando and Jacksonville Combined Data Cluster Analysis Centroids ................................ ................................ ................................ ............ 93 5 2 Within Cluster Sum of Squares Results (Combined Data) ................................ 93 5 3 Hartigan Index Results (Com bined Data) ................................ ........................... 94 5 4 Summary of Orlando Cluster Analysis Centroids ................................ ................ 96 5 5 Summary of Jacksonville Cluster Analysis Centroids ................................ ......... 96 5 6 Within Cluster Sum of Squares Results (Separated Data) ................................ 97 5 7 Hartigan Index Results (Separated Data) ................................ ........................... 99 5 8 Naming Convention of the 4 Groups ................................ ................................ 101 5 9 Number of Participants in each Group ................................ .............................. 101 5 10 Comparison of Survey Responses, Field Observations, and Cluster Analysis Results for I 4 Orlando Data ................................ ................................ ............. 102 5 11 Comparison of Survey Responses, Field Observations, and Cluster Anal ysis for I 95 Jacksonville Data ................................ ................................ ................. 103 5 12 Mean Lane Change Durations ................................ ................................ .......... 105 5 13 Mean Lag Gap Accepted in Congested Conditions ................................ .......... 106 C 1 Discretionary Lane Changes ................................ ................................ ............ 117 C 2 Mandatory Lane Changes ................................ ................................ ................ 117 C 3 Mandatory Lane Changes (merging maneuvers excluded) .............................. 118

PAGE 9

9 C 4 Mandatory Lane Changes (merging maneuvers only) ................................ ...... 118 C 5 U ncongested Discretionary and Mandatory Lane Changes ............................. 119 C 6 Uncongested Discretionary Lane Changes ................................ ...................... 119 C 7 Uncongested Mandato ry Lane Changes ................................ .......................... 120 C 8 Uncongested Mandatory Lane Changes (merging maneuvers only) ................ 120 C 9 Uncongested Mandatory Lane Changes (merging maneuvers excluded) ........ 121 C 10 Congested Discretionary and Mandatory Lane Changes ................................ 121 C 11 Congested Discretionary Lan e Changes ................................ .......................... 122 C 12 Congested Mandatory Lane Changes ................................ .............................. 122 C 13 Congested Mandatory Lane Changes (merging maneuvers only) .................... 123 C 14 Congested Mandatory Lane Changes (merging maneuvers excluded) ............ 123 C 15 Left and Right Uncongested Discretionary Lane Change Durations Comparison ................................ ................................ ................................ ...... 1 24 D 1 Uncongested DLC Durations vs. Congested DLC Durations Hypothesis Testing ................................ ................................ ................................ .............. 125 D 2 MLC (merge only) Durations vs Congested MLC (merge only) Durations Hypothesis Testing ................................ ................................ ........................... 127 D 3 Uncongested MLC (no merge) Durations vs Congested MLC (no merge) Durations Hypothesis Testing ................................ ................................ ........... 129 D 4 Uncongested MLC (no merge) Durations vs. Uncongested DLC Durations Hypothesis Testing ................................ ................................ ........................... 131 D 5 Congested MLC (no merge) Durations vs. Cong ested DLC Durations Hypothesis Testing ................................ ................................ ........................... 133 D 6 Right DLC Durations vs Left DLC Durations Hypothesis Testing ..................... 135 D 7 Uncongeste d Lag Gap vs. Congested Lag Gap Hypothesis Testing ................ 137 E 1 Uncongested DLC Durations Descriptive Statistics ................................ .......... 139 E 2 Uncongested DLC Durations Fitted Normal Parameter Estimates ................... 139 E 3 Uncongested DLC Durations Goodness of Fit Test ................................ ......... 139

PAGE 10

10 E 4 Uncongested DL C Durations Descriptive Statistics ................................ .......... 140 E 5 Congested DLC Durations Fitted Normal Parameter Estimates ....................... 140 E 6 Congested DLC Dura tions Goodness of Fit Test ................................ ............. 140 E 7 Uncongested and Congested DLC Durations Descriptive Statistics ................. 141 E 8 Uncongested and Cong ested DLC Durations Fitted Normal Parameters ......... 141 E 9 Uncongested and Congested DLC Durations Goodness of Fit Test ................. 141 E 10 Unc ongested and Congested DLC and MLC Durations Descriptive Statistics 142 E 11 Uncongested and Congested DLC and MLC Durations Fitted Johnson Su Parameters ................................ ................................ ................................ ....... 142 E 12 Uncongested and Congested DLC and MLC Goodness of Fit Test ................. 142 E 13 Uncongested and Congested DLC Lag Gap Descriptive Statistics .................. 143 E 14 Uncongested and Congested DLC Lag Gap Fitted Gamma Parameters ......... 143 E 15 Uncongested and Congested DLC Lag Gap Goodness of Fit Test .................. 143 E 16 Uncongested and Congested DLC Lead Gap Descriptive Statistics ................ 144 E 17 Uncongested and Congested DLC Lead Gap Fitted Gamma Parame ters ....... 144 E 18 Uncongested and Congested DLC Lead Gap Goodness of Fit Test ................ 144 E 19 Uncongested and Congested DLC and MLC Lag Gap Descriptive Statistics .. 145 E 20 Uncongested and Congested DLC and MLC Lag Gap Fitted Gamma Parameters ................................ ................................ ................................ ....... 145 E 21 Uncongested and Congested DLC and MLC Lag Gap Goodness of Fit Test 145 E 22 Uncongested and Congested DLC and MLC Lead Gap Descriptive Statistics 146 E 23 Uncongested and Congested DLC and MLC Lead Gap Fitted Johnson SI Parameters ................................ ................................ ................................ ....... 146 E 24 Uncongested and Congested DLC and MLC Lead Gap Goodness of Fit Test 146 E 25 Uncongested DLC and MLC Lag Gap Descriptive Statistics ........................... 147 E 26 Uncongested DLC and MLC Lag Gap Fitted Gamma Parameters ................... 147

PAGE 11

11 E 27 Uncongested DLC and MLC Lag Gap Goodness of Fit Test .......................... 147 E 28 Congested DLC and MLC Lead Gap Descriptive Statistics .............................. 148 E 29 Congested DLC and MLC Lead Gap Fitted Johnson SI Parameters ................ 148 E 30 Congested DLC and MLC Lead Gap Goodness of Fit Test ............................ 148 F 1 Group 1 Very Conservative Drivers Summary of Statistics ............................ 149 F 2 Group 2 Somewhat Conservative Drivers Summary of Statistics ................... 149 F 3 Group 3 Somewhat Aggressive Drivers Summary of Statistics ...................... 150 F 4 Group 4 Very Aggressive Drivers Summary of Stat istics ............................... 150 G 1 Difference of Means for all Pairs of Driver Groups ................................ ........... 152 G 2 Least Significant Difference Threshold Matrix ................................ .................. 152 G 3 Connecting Letters Report ................................ ................................ ............... 152 G 4 Ordered Differences Report ................................ ................................ ............. 152 H 1 Difference of Means for all Pairs of Driver Groups for Lag Gaps ..................... 154 H 2 Least Significant Difference Threshold Matrix for Lag Gaps ............................ 154 H 3 Connecting Letters Report for Lag Gaps ................................ ......................... 154 H 4 Ordered Differences Report for Lag Gaps ................................ ....................... 154

PAGE 12

12 LIST OF FIGURES Figure page 1 1 Flow of Work Framework ................................ ................................ .................... 22 2 1 Gap Acceptance Terminology ................................ ................................ ............ 49 2 2 Gipps Lane Changing Model Decision Process Flowchart ................................ 50 2 3 Ahmed et al. Lane Changing Model Structure ................................ .................... 51 2 4 Lane Change Str ucture with Explicit Target Lane Choice ................................ .. 52 2 5 Lane Changing Rate and Traffic Density Relationship (Kesting et al., 2007) ..... 52 2 6 Structure of a General Lane Change Process ................................ .................... 53 2 7 Distribution of Accepted and Rejected Lag Gaps ........................ 53 3 1 Instr umented 2006 Honda Pilot SUV ................................ ................................ .. 66 3 2 Orlando Study Route: Interstate 4 from Conroy Rd. (A) to Maitland Blvd. (B) (11 miles) ................................ ................................ ................................ ............ 67 3 3 Jacksonville Study AM route: Interstate 95 from Philips Highway (A) to Bowden Road (B) (6.5 miles) ................................ ................................ .............. 68 3 4 Jacksonville Study PM route: Interstate 95 from Baymeadows Road (A) to Bowden Roa d (B) (3.8 miles) ................................ ................................ .............. 69 4 1 Lane Change Duration Distribution for this paper (N = 726, Johnson Su distribution) ................................ ................................ ................................ ......... 76 4 2 Lane Chang e Duration Distribution by Toledo and Zohar 2007 (N = 1,790) ....... 77 4 3 Lane Change Duration Distribution by Hetrick 1997 (N = 282) ........................... 78 5 1 Plot ................................ ................................ ................................ ..................... 92 5 2 Within Cluster Sum of Squares vs. Number of Clusters Plot (Combined Data) .. 94 5 3 Cluster Analysis Results for the Combined Data ................................ ................ 95 5 4 Within Cluster Sum of Squares vs. Number of Clusters Plot (Orlando) .............. 98 5 5 Within Cluster Sum of Squares vs. Number of Clusters Plot (Jacksonville) ....... 98

PAGE 13

13 5 6 Grouped Orlando Participants based on DLC/mile and Desired Speed ........... 100 5 7 Grouped Jacksonville Participants based on DLC/mile and Desired Speed ..... 100 5 8 Mean Lane Change Duration by Driv er Groups ................................ ............... 105 5 9 Mean Lag Gap Accepted in congested Conditions by Driver Groups ............... 106 B 1 Determining the Fugitive Point for th e A) front and B) rear camera. ................. 115 E 1 Uncongested DLC Durations Histogram ................................ ........................... 139 E 2 Congested DLC Durations Histogram ................................ .............................. 140 E 3 Uncongested and Congested DLC Durations Histogram ................................ 141 E 3 Uncongested and Congested DLC and MLC Durations Histogram .................. 142 E 4 Uncongested and Congested DLC Lag Gap Histogram ................................ ... 143 E 5 Uncongested and Congested DLC Lead Gap Histogram ................................ 144 E 6 Uncongested and Congested DLC and MLC Lag Gap Histogram .................... 145 E 7 Uncongested and Congested DLC and MLC Lead Gap Histogram .................. 146 E 8 Uncongested DLC and MLC Lag Gap Histogram ................................ ............. 147 E 9 Congested DLC and MLC Lead Gap Histogram ................................ ............... 148 G 1 Multiple Means Comparison of Lane Change Durations for Driver Groups ...... 151 H 1 Multiple Means Comparison of Lag Gaps Accepted in Congested Conditions for Driver Groups ................................ ................................ .............................. 153

PAGE 14

14 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Engineering LANE CHANGING ON FREEWAYS By Corey Hi ll August 2012 Chair: Angeliki Elefteriadou Major: Civil Engineering Lane changing models are a significant component of microscopic traffic simulation. Understanding the details of this fundamental maneuver is important for accurate modeling in simulat ion. Therefore, lane changing has received much attention. Many studies have focused on the details of the lane change maneuver from external observation based data without regard for the type of driver performing the maneuver. In this thesis, the physical details of freeway lane changing have been related to the type of driver performing the maneuver. 46 research participants drove an instrumented vehicle and performed a combined total of 726 freeway lane changes. A cluster analysis was performed to categ orize each research participant into one of four groups ranging from conservative to aggressive. Then an analysis was done to determine any trends that existed between the different driver types and their lane changing characteristics, specifically lane ch ange duration and gap acceptance characteristics It was found that, in general, more conservative drivers have greater lane change durations than aggressive drivers. The gap acceptance comparison among driver types did not yield any conclusive trend.

PAGE 15

15 In a ddition, d istributions were fitted to lane change duration and gap acceptance histograms. Also, hypothesis testing was used to determine if significant differences occur for lane change durations and accepted gap sizes for different lane change types and c ongestion conditions. The results of this thesis suggest that driver types do have a significant role in the details of freeway lane chang ing Therefore this relationship should not be disregarded when developing lane changing models.

PAGE 16

16 CHAPTER 1 INTRODUC TION Problem Statement Lane changing is an integral part of freeway driving. Every driver participates in the lane change maneuver at some point in their freeway driving experience. From the engineering perspective, these routine maneuvers must be thorough ly understood as they have a significant effect on traffic flow (Toledo et al. 2003; Coifman et al. 2005; Laval and Daganzo 2006; Toledo and Zohar 2007; Moridpour et al. 2010). Lane change maneuvers are a key component to understanding breakdowns in the fl ow of traffic. Therefore, lane changing on freeways deserves close attention. Understanding the details involved in a lane change maneuver and the relationship between different driver types and lane changing is important in understanding the overall perfo rmance of the freeway. Understanding lane changing is also a key component to developing an effective microscopic simulator. According to Pursula (1999), the use of computer simulation as a ertation titled "Simulation of Freeway Traffic on a General purpose Discrete Variable Computer". Today, microscopic traffic simulators have become increasingly popular. Simulators have developed into an effective tool that is used for a wide range of trans portation applications (Ben Akiva et al. 2006; Hidas 2002). Car following, lane changing, and gap acceptance are the three basic components of any microscopic simulator. Car following is less com plicated to model because there are fewer steps involved. La ne changing and gap acceptance, on the other hand, depend heavily upon driver characteristics and the amount of risk they are willing to

PAGE 17

17 take. This makes lane changing and gap acceptance more difficult to accurately replicate in a microscopic simulator. Th e variability of driver types and their corresponding driving behavior introduces an obstacle in attempting to replicate the movement of vehicles on a freeway. A sound lane changing model should account for this variability of drivers. Part of this study a ttempts to draw conclusions about the relationship between different driver types and lane change behavior. This will aid in constructing a reasonable lane change algorithm that captures the variability associated with drivers. In return, t his will ultimat ely allow for more accurate m icroscopic simulation of vehicular movement along a freeway. A difficulty in the study of lane change maneuvers is the data collection process. Lane changing is a complex maneuver involving more than just the lane changing vehi the current lane and target lane must be known. In this study, data collection was performed via an instrumented vehicle. Vehicle location, gap sizes, lane change duratio ns, and speeds are all readily available through video data. This comprehensive data collection method was not available to many previous researchers. It allows for greater insight into lane changes, with the potential of leading to more comprehensive lan e changing algorithms. Gap acceptance plays a primary role in most lane change models (Bham 2008). It is a major part of many lane change maneuvers and becomes very important during congested freeway conditions. As congestion increases the number of opport unities for lane changes decreases. This creates a situation in which drivers are forced to search for suitable gaps in the traffic stream to complete a lane change. Many microscopic

PAGE 18

18 simulators execute lane changing through gap acceptance. Typically, a pre specified critical gap threshold value is determined and the feasibility of a desired lane change will depend upon the size of the gap in the target lane. Determining a reasonable value for a critical gap has proven to be a challenge as the ability to col lect values of rejected gaps is not immediately observable (Bham 2008). Despite the critical nature of freeway lane changing it has only recently received increased attention (Toledo and Zohar 2007; Kesting et al. 2007; Sun and Elefteriadou 2010). In light of the importance of lane changing in relation to freeway traffic flow and microscopic simulators, this thesis will provide a critical evaluation of lane changing on freeways through field data from an instrumented vehicle. The purpose is to analyze lane changing from an aggregate level and to determine if any relationship exists between different driver types and their lane changing characteristics. Research Objectives The purpose of this research is to gain insight on freeway driving in relation to lane changing. Specifically, this purpose comprises the following objectives: Determine significant differences between different lane change types during congested and uncongested conditions. Not all lane change maneuvers are the same or occur in the same traf fic conditions. Therefore, lane changes should not all be modeled in the same fashion. The purpose of this objective is to attempt to explain significant d ifferences between different lane change types and the congestion conditions that they occur in. The congestion condition is based upon the speed that the driver is travelling and by observation from the video data. Each lane change is classified as one of the following lane change types:

PAGE 19

19 Discretionary lane change, DLC Mandatory lane change, MLC ( non merg ing maneuvers only ) Mandatory lane change, MLC (merging maneuvers only ) Mandatory lane change, MLC ( non merging and merging maneuvers combined ) Each maneuver within these lane change types are then separated according to whether they occur in congested or uncongested conditions. Once all of the data has been sorted into the respective categories, hypothesis testing is used to compare the means and draw conclusions con cerning significant differences between the differing categories. Determine lane change du ration and lag gap acceptan ce distributions. Lane change duration and gap acceptance histograms are made for each of the lane change type s and congestion conditions outlined in the previous objective. Distributions are fitted to the se histograms and a good ness of fit test is performed for each fitted distribution. The omission of lane changing duration from microscopic simulation may have significant effect on simulation outputs (Toledo and Zohar 2007). Presenting the fitted distribution of a large data set of lane change durations as well as showing the distribution of gaps accepted should be considered in the development of lane changing models. The distributions provide an overarching view of the data as they explai n the pattern that they follow. Classify drivers based upon their desired speed and the number of discretionary lane changes completed per mile. This objective is aimed at understanding the relationship between the desired speed of an individual and the number of discretionary lane changes they complete. The desired speed is defined as the speed at which the driver travels when driving

PAGE 20

20 under free flowing and not car following conditions. A plot of the desired speed and number of discretionary lane changes completed is provided to show this relati onship. Once this relationship is established, it is used in a cluster analysis. This results in the classification of each driver into a group of similar drivers with similar characteristics. Investigate the relationship between different driver types an d their lane changing characteristics. The human factor of driving introduces a great amount of variation when trying to develop models. Previous studies have investigated lane change maneuvers without regard for the type of driver performing the maneuver. This objective attempts to draw conclusions concerning the relationship between different driver types and their lane changing characteristics such as lane change duration and gap acceptance behavior. The data collected are analyzed in terms of the drive r groups established from the previous objective. A multiple means comparison statistical analysis is performed to compare the data obtained for each driver group. Thesis Outline The gen eral flow of this research consists of obtaining research participants gaining information of each participant, collecting the data via the instrumented Honda Pilot, extracting the data of interest from the raw collected data, analyzing the data at an aggregate level analyzing the data from the perspective of different dri ver types, and drawing conclusions from the analyzed data. The end goal of this process is to provide more understanding of lane changing on freeways. Specifically, to determine relationships between different driver types and lane change behavior. Many st udies have been done concerning the specifics of lane changing as observed in the field. However, much fewer studies have attempted to draw conclusions between lane

PAGE 21

21 changing and the type of driver performing the lane change maneuver Figure 1 1 provides th e frame work of the flow of work that is followed to accomplish these goals. In Chapter 2 a literature review concerning lane changing on freeways is provided. The literature review presents a critical review of the history of lane changing as it summarize s previous studies and highlights current weaknesses. Chapter 3 outlines the specifics of the data collection process that was used In Chapter 4 the data from two sources, Orlando and Jacksonville, are analyzed at an aggregate level Chapter 5 discusses the cluster a nalysis process of this data which involves grouping drivers according to their desired speed and discretionary lane changes made per mile. Then the data is analyzed in relation to each driver group. Chapter 6 summarizes the results and conclu sions drawn from the data analysis process.

PAGE 22

22 Figure 1 1. Flow of Work Framework

PAGE 23

23 CHAPTER 2 LITERATURE REVIEW This review focuses on evaluating the state of the art on lane changing in the field of transportation engineering. Previous studies are reviewed to determine what models have been developed for lane changing. Different algorithms have been implemented in various simulation software and many previous field experiments have been conducted concerning the analysis of lane changing on freeways. Reviewing these cases will provide an overall perspective of lane changing on freeways. Knowing what has already been thoroughly studied and what is still outstanding has aid ed in the formulation of this research. Conclusions shall b e summarized in order to identify the strengths and limitations of the current state of lane changing in the transportation engineering field. Many terms will be used throughout this paper to refer to situations common to lane changing and gap acceptance o n freeways. For clarity the following terms have been defined and shown in Figure 2 1 Subject Vehicle: The vehicle that is attempting to move from the current lane into the target lane by accepting the gap between the lead and lag vehicle. Lead Vehicle: The vehicle that is at the front of the available gap. This vehicle is the least affected by the gap acceptance process. Lag or Following Vehicle: This vehicle is at the rear of the available gap. The terms lag or following are used interchangeably through out the paper. This vehicle is much more affected by the movement of the subject vehicle into the available gap. Available Gap: The space between the lead vehicle and the lag vehicle. This is the space that the subject vehicle is attempting to occupy safel y. Critical Gap: The minimum spacing between the lead and lag vehicle that the subject vehicle will accept. A gap less than the critical gap will theoretically be rejected by the subject vehicle. Lag Gap: The spacing between the lag vehicle and the subject vehicle. Lead Gap: The spacing between the lead vehicle and the subject vehicle.

PAGE 24

24 Current Lane: The lane that the subject vehicle currently occupies and desires to move out of. Target Lane: The lane that the subject vehicle desires to occupy provided that the available gap between the lead and lag vehicle is sufficient. The following paragraphs are organized into five sections: Lane Changing Models, Lane Change Duration, Lane Changing Models in Simulation Software, Gap Acceptance, and Summary. The first sec tion reviews different lane changing models that have been developed. Specifically, presenting the structure of the models and what distinguishes them from other models. The second section provides an overview of previous studies that specifically focused on lane change duration. The next section discusses various microscopic simulators in relation to their lane changing models. The fourth part reviews gap acceptance studies. The final section provides a summary of the major findings and conclusions. Lane Changing Models Gipps (1986) developed one of the first lane changing m odels to be applied to a micro during a lane change maneuver. The model is specifically intended for urba n arterials; however, the concepts behind the model can be related to lane changing on freeways as well. Gipps explains that there are three general questions behind every lane changing process. Is it possible to change lanes? Is it necessary to change lan es? Is it desirable to change lanes? The way in which the driver answers these questions will determine whether a lane change occurs or not. He continues to outline a more detailed framework of the questions that are asked and must be answered by the drive r before a lane change can be completed. This framework is disp layed in a flowchart in Figure 2 2 which illustrates the series of questions that Gipps included as relevant to the lane

PAGE 25

25 changing process. Gipps includes in his model the selection of lanes, th e feasibility of the maneuver, the presence of a desired turning movement, the urgency of the maneuver, the lane type, the relative advantages between the present and target lane, the presence of heavy vehicles, and the safety of the maneuver. The decision differentiates between preferred lane and target lane. The preferred lane refers to the lane adjacent to the current lane that is on the side of the desired turning movement. The targe t lane is the lane that the driver is thinking of changing into. Initially the target lane is the preferred lane, however, if the preferred lane is impossible to change into then the target lane will be adjusted. Once the preferred an d target lanes are ass igned the next decision considers the feasibility of the lane change. The feasibility is based on if the target lane is available and whether or not the subject vehicle is aligned appropriately with the gap in the target lane. If the lane change is feasib le then it must be determined if an upcoming intended turn is within the pre established threshold value or not. If the target lane is preferred, the maneuver is feasible, the intended turn is close, and the target lane is not blocked by any obstructions t hen the vehicle will determine if it is safe to change lanes and complete the lane change. The model also incorporates transit vehicles, heavy vehicles, and the presence of obstructions in the target lane as seen in Figure 2 2 For arterials there may be transit only lanes which would clearly be a factor in the lane changing decision process. Also if the presence of a heavy vehicle in the present lane limits the speed of the subject vehicle more than in the target lane then the lane change will be desirabl e. The presence of obstructions refers to anything that would block the target lane making it not

PAGE 26

26 possible to be utilized. For example, there could be parked cars along a curb which are blocking a lane or an incident may have occurred creating an obstructi on. While these scenarios are not relevant in dealing with lane changing on freeways Gipps does say that the model can be modified in order to relate better to different environments in which the lane changes are occurring. So rather than a transit lane t he flowchart could incorporate an HOV lane. Likewise rather than blocked lanes due to parked vehicles near a curb this could be adapted to more relevant freeway scenarios like: an incident, a work zone, or a variable message sign warning drivers of a clos ed lane ahead. A limitation of this lane changing model is the assumptions that are made in the lane changing decision process. There are questions throughout the flowchart that are not directly observable and therefore make it difficult to verify their ac curacy. For example, it is generalized that all drivers choose a target lane first and then upon answering that question begin to analyze if the maneuver is feasible by assessing the available gap. However, it could be that these two questions are intercha ngeable in that a driver observes a feasible lane change opportunity first and then decides that that is reflect the exact process every driver goes through when changing l anes. Furthermore, it is difficult to assess the model with field data as the physical lane change itself is the only observable part of the process. Ahmed et al. (1996) developed a lane changing model, shown in Figure 2 3 based on three steps during a la ne change process: decision to consider a lane change, choice of left or right lane, and accepting a gap in the desired lane. Another main distinction made in the structure of the model is mandatory lane change versus

PAGE 27

27 discretionary lane change. They specif y that factors necessitating a mandatory lane change include lane use regulations, incidents, and upcoming exit ramp. Discretionary lane changing is performed due to speed differentials, deceleration of the lead vehicle, presence of a heavy vehicle, and oc cupying a lane adjacent to a ramp. The model attempts to specifically include the decision process that occurs within the driver and not just the physical lane change itself. The developed model was evaluated by using a data set collected by FHWA (Smith 19 85). It was concluded that the developed lane changing model successfully incorporated the decision process and randomness of driver behavior that occurs during a lane change process. Hidas (2005) collected data from video recording and analyzed lane chang ing maneuvers. He observed the interactions with respect to gap acceptance between the vehicle making the lane change, the lead vehicle in the target lane, and the lag vehicle in the target lane. Based on these observations, he similarly proposed, as other s have, that lane changes be classified as free, forced, or cooperative. The feasibility of a lane change is based on the acceptability of the gap between the lead and following vehicle. Hidas explains that the minimum acceptable gap is not a constant valu e but it depends on the driver type and whether the lane change is free, forced, or cooperative. Ben Akiva et al. (2006) observed that most models classify lane changes as either mandatory or discretionary. Separating the two types of lane changes implie s that there are never any compromises between the two. They explain, by way of example, that a vehicle on a freeway that desires to exit via an off ramp will undergo a mandatory lane change once within a pre specified threshold distance to the exit. Upon completion of the mandatory lane change the vehicle will remain in that lane no matter how slow the

PAGE 28

28 leading vehicle is traveling. In reality, a vehicle may overtake a vehicle traveling significantly slower even if they are within the threshold. This exampl e shows one of the limitations that result from treating mandatory and discretionary lane changes as completely separate entities. Therefore, they proposed that an integration of these two types of lane changes into a single utility model provides a more r ealistic result. They conclude that variables like obstacles, lane closures, incidents, speed advantages, and ease of driving can all be combined into a single utility model rather than segregated into mandatory or discretionary. Returning to the previousl y discussed example, according to an integrated model the probability of changing lanes would gradually increase as the distance to the exit ramp decreased. A model where mandatory and discretionary lane changes are separated would result in an abrupt and permanent lane change once a certain threshold distance to the exit ramp was crossed without any further consideration of discretionary lane changing variables. Another limitation with many lane changing models that Ben Akiva et al. (2006) observed was th e absence of an explicit target lane choice. Many models only analyze the characteristics of the current lane and the immediate adjacent lanes to determine the result of a lane changing maneuver. However, it is observed in the field that drivers choose a t arget lane and may change to an adjacent lane with less desirable conditions in order to eventually arrive at their target lane where more desirable conditions exist. For example, if a vehicle desires to be in an HOV lane, they may temporarily change into a slower moving lane in order to eventually arrive at the desired HOV lane. Rather than being constrained to choosing between left, right, or current lane only the more advanced model can choose from all available lanes and asses which one is most

PAGE 29

29 desirabl e. Upon concluding the most advantageous lane the vehicle will then proceed to make the appropriate maneuvers even if it means temporarily changing into a less desirable lane. Figure 2 4 is an example of a lane change structure that utilizes the concept of the target lane being explicitly chosen. The scenario in Figure 4 represents a four lane freeway in which the vehicle is currently in lane 2. In order to compare an explicit target lane choice model to a direction of lane change choice model a study was conducted on a section of I 80 in Emeryville, California using MITSIMLab. The results of the two models were compared to what was observed in the field. The section consisted of 6 lanes in which the left most lane is an HOV lane. Overall, the research show ed that the model with the explicit target lane choice better modeled what was actually observed in the field. Specifically interesting was the observation that the model with a direction of lane change choice structure was under p redicting the usage of th e HOV lane. Essentially since the model was only analyzing the current, left, and right lanes fewer vehicles were getting to the HOV lane than what was observed. Many lane change models approach the concept of feasibility by using gap acceptance parameters However, if a lane change is feasible or not does not solely depend on the size of the gap available for the maneuver. There are situations in which available gaps simply do not exist due to congested conditions. Therefore, other instruments are needed f or lane changing when an available gap does not exist. Certain models contain a courtesy or forced lane change component in which a vehicle may change lanes despite there being insufficient gap. With courtesy lane changes the lag vehicle in the lane changi

PAGE 30

30 the lane change. A forced lane change occurs when the lane changing vehicle forces into the target lane which compels the lagging vehicle to accommodate by slowing down. Lee (2006) compared a lane changing model that used gap acceptance as the only option through which a change could occur with a model that allowed for courtesy and forced lane changes. The study was done along a section of US101 in Los Angeles, California using MITSIMLab. Data was collected in order to show at what distance until the end of a merging lane vehicles change lanes. The observed data yielded that 44% of vehicles change lanes within 100 meters to the end of the lane. The model capturing gap acceptance, courtesy, and force d lane changes predicted that 47% of the lane changes occur within 100 meters from the end of the ramp whereas the model only including the gap acceptance parameters predicted 81%. The model including courtesy and forced lane changes in addition to the sta ndard gap acceptance parameters was substantially closer to the observed data and it was concluded that it is a better predictor of lane changing behavior. Kesting et al. (2007) proposed a general lane changing model. The model takes a different approach from others as its main driving factor is minimizing overall braking induced by lane change (MOBIL). The model is driven by safety constraints and the incentive to change lanes. Like many other lane changing models this one incorporates gap acceptance mode ls. However, what is different about this model is that it uses gap acceptance indirectly rather than directly. The minimization of overall braking during a lane change is the criterion by which the lane change is executed. The accelerations depend upon th e ratio between the effective desired minimum gap and the actual gap.

PAGE 31

31 The lane change maneuver is executed through gap acceptance indirectly as the accelerations depend upon the size of the gaps during the maneuver. One noted advantage to making the criter ia of a lane change depend upon the safe braking decelerations rather than directly upon gap acceptance models is that crashes are eliminated. Another portion of this study involved a lane changing rate. It was noted that the lane changing rate upon a free way is related to the amount of suitable gaps available. Initially, as the density increases on the freeway the lane changing rate will be seen to increase. This is because a higher density means more vehicles that are able to change lanes. However, if the density continues to increase there will come a point where the lane change rate will reach a maximum and then begin to decline. Gap acceptance explains why this trend occurs. At first even though density increases there are still suitable gaps that drive rs can readily accept. But once congestion reaches a certain point, the number of available gaps diminish, which directly affects the number of lane changes that can be made. The results of multiple simulation runs were done in order to see how the lane c hange rate and density are related according to the proposed MOBIL model. One example of the relationship between lane changing rate and availability of acceptable gaps can be seen in Figure 2 5 The plot shows the simulation results from two different len gth sections of freeway (5.5 km and 7.5 km). A politeness factor is also used in this model. Figure 2 5 is for a politeness factor of 0 which implies that the lane changing drivers make selfish decisions not considering the state of the overall traffic sit uation. As the density

PAGE 32

32 increases to higher values the number of suitable gaps decreases which causes the number of lane changes made to decrease. Sun and Elefteriadou (2010) developed an urban arterial lane changing model that incorporates data concerning driver behavior and driver background information. A focus group study was conducted in order to obtain information about driver characteristics and overall perceptions concerning lane changing in urban arterial scenarios. Each driver was classified into o ne of four driver types based on the background of each driver and their stated driver behavior. The other form of data collected was from an instrumented vehicle equipped with video recording. This data was used to analyze every lane change made by each d river. A total of 205 potential, 199 attempted, and 601 completed lane changes were identified and analyzed. From the focus group and the in vehicle data a lane changing model was developed that incorporated driver behavior. This model was then implemented into CORSIM. The simulation results suggests that the new model more accurately represents the interlane travel time differences, lane use, and cumulative number of lane changes than the previously used model. Lane Change Duration Worrall and Bullen (197 0) performed an empirical analysis of lane changing on multilane freeways. Data for the study were collected at 30 different freeway locations in Chicago. The focus was to investigate the patterns and frequency of lane changing maneuvers related to durati ons and the acceptance and rejection of gaps. The range of lane change durations was from 2.3 to 4.1 seconds with a median of 3.2 seconds. Finnegan and Green (1990) reviewed five articles that researched lane changes and reaction time. They would use this data to create a model to predict traffic flow.

PAGE 33

33 From the compiled data it was determined that lane changes took between 4.9 and 7.6 seconds, including visual search time. The studies also stated that more experienced drivers had shorter reaction times asso ciated with looking at mirrors when compared to novice and older drivers. Hetrick (1997) used researchers to collect lane change data from within an instrumented vehicle. 16 participants drove an instrumented vehicle. One part of the study involved the ana lysis of lane change durations. A total of 282 lane changes were made by the participants who drove both urban streets and freeways. The distribution of lane change durations ranged from 3.41 to 13.62 seconds. The mean lane change was stated to be 6 second s. The study further states that the younger participants accounted for the shorter lane change times while the older drivers accounted for the upper side of the spectrum. Salvucci and Liu (2002) performed a study using simulators to see how participants w ill react to lane changes. Several graphs were constructed to show lane change over time as well as eye movements and turn signal usa ge. It was determined that 2 to 3 seconds before a lane change occurs, the participants moved away from the target lane bef ore moving into the target lane. This does not seem intuitive and they reported that given the smaller scope of the study and the use of a driving simulator rather than a real vehicle, the results should be taken as preliminary observations. Overall, o f th e 11 participants tested, it was found that the average duration of a lane change was 5.14 seconds. Lee et al. (2003) used instrumented vehicles and 16 participants to understand lane change frequency and duration based on multiple considerations. The mean lane

PAGE 34

34 change duration was 9.07 seconds for the entire data set of 8,667 lane changes. Of these, 83% were single lane changes with a mean duration of 6.28 seconds. It was reported that lane changes to the left were longer than to the right. Mean left lane c hanges were 11.11 seconds. While, mean right lane changes were 6.62 seconds. It is suggested that this difference between left and right lane change durations is probably due to their being more double lane changes to the left than to the right. Tijerina e t al. (2005) studied the eye glance behavior of drivers during the lane change decision phase. A total of 549 lane changes were analyzed in order to determine eye glance behavior leading up to a lane change. The participants drove an instrumented vehicle a ccompanied by a researcher. It was found that for left to right lane changes, the probability of a glance to the center mirror was substantially higher than a glance to the right side mirror. For a lane change to the left the probability of a glance to th e center mirror was significantly less than for a lane change to the right. Other data that was reported included the duration of each lane change. For the highway lane changes, the duration ranged from 3.5 seconds to 8.5 seconds with a mean of 5.8 seconds Toledo et al. (2007) performed a study on the duration of lane changes. The study was done on the eastbound section of I 80 in Emeryville, California. They used video cameras mounted on a roadside high rise building for data collection This section of f reeway included an on ramp, off ramp, and an HOV lane. A total of 1,709 completed lane change maneuvers were identified and the duration of each was extracted from the video recordings. The results yielded a lane change duration range of 1 to 13 seconds wi th a mean of 4.6 seconds. They noted that many microscopic simulation lane

PAGE 35

35 changing models consider the maneuver as an instantaneous event. The study is significant because it demonstrates that lane change durations are not instantaneous and that simulati on software should account for this in their lane changing models. The output of simulation software may be inaccurate if the lane change model used assumes an instantaneous maneuver as opposed to accounting for the duration of the maneuver. A summary of t he lane change du ration literature is provided at the end of this chapter. The lane change duration results obtained from this research will be compared to what has been found in these previous studies. Lane Changing Models used in Simulation The following section provides a general overview of three microscopic traffic simulators: AIMSUN, CORSIM, and SITRAS. The specific details concerning lane change duration and gap acceptance was included where possible. However, these specifics were not completely ex pl ained by the sources obtained. Many simulators attempt to incorporate different driver types into the modeling process; however, they do not base the differences among driver types on field data. AIMSUN The lane changing model used in AIMSUN is a developm ent of the Gipps lane changing model. The model explains the lane changing decision process as determining the necessity, desirability, and the feasibility of a lane change maneuver. The necessity of a lane change refers to those maneuvers that are determi ned by the route of the vehicle. This is similar to a mandatory lane change discussed previously in other models. The desirability of a lane change is similar to a discretionary lane change. A lane change may not be absolutely necessary but if it would pro vide a speed or queue

PAGE 36

36 advantage then a driver will likely undertake the maneuver. The feasibility of the lane change refers to whether or not the surrounding conditions of the vehicle allow for a safe lane change maneuver, like if there is available gap or not. AIMSUN assesses the necessity, desirability, and feasibility of a lane change every time a vehicle is updated. In order to make the assessment, factors like the distance to an exit ramp, speed and queue lengths of the current and surrounding lanes, a vailable gap size, and braking ratios are considered. Another dimension of the AIMSUN lane changing model is the use of different zones to provide a more accurate representation of driving behavior. The model uses three zones that determine the type of ma neuvers a vehicle will perform. Zone 1 is the furthest distance from an exit ramp. It is in this zone where lane changing decisions are based on the surrounding conditions like speed and queue lengths. A mandatory lane change does not need to be made in z one 1. Zone 2 is closer to the exit ramp but there is still sufficient distance for vehicles to make the appropriate maneuvers. In this zone lane changing decisions are controlled by the need to exit soon. Vehicles can be seen to be changing lanes in order to get closer to the side of the road where the exit ramp is located. Finally, in zone 3 there is very little distance left to the exit ramp and lane changing decisions are solely based on the need to get in the appropriate lane to complete the exiting ma neuver. It is in this zone that vehicles are compelled to reach there desired destination and may force a lane changing maneuver causing other vehicles to slow down in order to provide sufficient gap. CORSIM CORSIM incorporates three types of lane changing scenarios: mandatory lane change, discretionary lane change, and anticipatory lane change. The simulator

PAGE 37

37 classifies a lane change as mandatory for the following four situations: 1) the vehicle is on an acceleration auxiliary lane and must merge into the m ainline traffic, 2) the current lane will be dropped downstream and the vehicle has passed an advanced sign the four scenarios that require a mandatory lane change are similar in that the degree of risk associated with the maneuver increases as the distance to t he end of the lane change opportunity decreases. The risk function in CORSIM estimates the amount of risk associated with a certain lane change maneuver. This function involves a pre specified minimum and maximum risk, along with the square root of the rem aining distance to the event requiring a lane change. Discretionary lane changes in CORSIM consider motivation, advantage, and urgency of a lane change in addition to the amount of risk. To quantify the degree of desire a driver possesses to change lanes, characteristics are taken into consideration. CORSIM represents different driver types by using a random number between 1 and 10 where 1 is most conservative and 10 is most aggressive. Using this random number a computed as a percent. After dete rmining if any motivation for a lane change exists, CORSIM then proceeds to determine if there is any advantage in executing a lane change. To do this,

PAGE 38

38 CORISM uses a Lead Factor and a Putative Factor. The Lead Factor determines the disadvantage associated with staying in the current lane behind the lead vehicle. The Putative Factor determines the advantage to moving into the target lane. A Putative Factor is computed for adjacent lanes and the one providing the largest advantage is chosen as the target lane The difference between the Putative Factor and the Lead Factor is the overall Advantage factor which is compared to a pre specified threshold exceeding the threshold will atte mpt to make the discretionary lane change. A third component of the discretionary lane change process quantifies the urgency of the lane change. An Urgency Factor is computed to capture how drivers who desired to make a lane change but were unable to do s o are more urgent when future lane change opportunities come up. An Impatient Factor is also calculated as a function of driver type. The final lane change type incorporated in CORSIM is the anticipatory lane change. This attempts to capture the effect of an on ramp on a vehicle already in the from an on ramp and decide to make a lane change to provide space for the merging vehicles. To do this CORSIM has an advanced warn ing sign 1500 feet upstream of the on ramp. Once the warning sign is reached the logic determines if there is any advantage in performing an anticipatory lane change. This is done using Lead and Putative Factors similar to how it is done for a discretionar y lane change. In the freeway portion of CORSIM, FRESIM, the default value for the time to complete a lane change maneuver is 2 seconds. This value can be modified by the

PAGE 39

39 user. However, this default value is lower than what has been reported by various fi eld studies concerning lane change duration, including the study of this paper. CORSIM makes the assumption that all drivers have equivalent gap acceptance characteristics In reality, drivers behave differently concerning the size of gap they will accept or reject during the lane change maneuver. Furthermore, CORSIM uses a direction of lane change model when determining the target lane, meaning that only the adjacent lanes are evaluated. This may result in an underproduction of the number of lane changes that occur. This is because drivers can observe favorable conditions in a lane that is not immediately a djacent to their current lane. In this case, they will readily move into an adjacent lane that has undesirable characteristics in order to eventually ar rive at the target lane with desirable characteristics. A direction of lane change ty pe model, like CORSIM, cannot replicate this driver behavior because only the adjacent lanes are considered for the potential target lane. SITRAS Hidas (2002) presents the details of a lane changing algorithm that is incorporated in a specific simulation model. The simulation model is Simulation of Intelligent Transport Systems (SITRAS). SITRAS is a microscopic simulation model that has been developed at the University of N ew South Wales starting in 1995. Hidas focuses on how this simulation model incorporates lane changing and evaluates if the software adequately simulates the lane change maneuver. He addresses how lane changing is modeled in SITRAS and discusses lane chang ing observations made in the field. One part of the lane changing process outlined in SITRAS is whether the lane change is feasible or not. In SITRAS, answering this question depends upon the status

PAGE 40

40 of the gap in the target lane. If the target lane gap is sufficient to accommodate the completion of a safe lane change, then the maneuver is deemed feasible. The gap being considered by the lane changing vehicle is the distance between the lead and following vehicle in the target lane. If the gap is not suffici ent to accommodate the lane changing vehicle then the lead and following vehicle will be required to adjust their speeds. The degree to which the subject or follower vehicle must change their speed is what is used to determine if a gap is sufficient or not The concept is that if a gap is insufficient then either the subject vehicle must decelerate to move behind the lead vehicle or the following vehicle must decelerate to accommodate the subject vehicle. Frequently both of these situations can be seen to o ccur following model, does not meet the threshold value then the gap is insufficient. When a gap is insufficient and a lane change is mandatory, SITRAS uses a forced lane change algorithm. The algorithm attempts to simulate driver courtesy in which a vehicle will adjust their speed in order to create a sufficient gap. This situation is unique to mandatory lane changes and is more commonly seen in highly congested freew ay conditions. The size of a critical gap is based on the speed and acceleration of the subject vehicle in relation to the lead and following vehicle in the target lane. Observations from the field have shown that drivers may accept very small gaps when th e relative speed between subject vehicle and lead/following vehicles is close to zero. A weakness being addressed in SITRAS involves a difference in the simulation behavior and the behavior of drivers as observed in the field. During simulation a

PAGE 41

41 vehicle may make a lane change into a smaller than desired gap. When this occurs it causes the platoon of vehicles in the target lane to be disrupted. The car following model incorporated in SITRAS reacts to this smaller gap by causing the following vehicle to dec elerate significantly. This sudden reduction in speed by the following vehicle then causes the average speed of the target lane to decrease substantially. Hidas explains that when this scenario is compared to field observations it is shown to be deficient. Observations show that drivers regularly accept the risk associated with accepting a smaller gap. When this occurs the subject vehicle once positioned in the small gap will gradually decelerate until they reach a comfortable amount of spacing between thei r vehicle and the lead vehicle. Then the following vehicle, provided that they have sufficient vision of the vehicles ahead of them and can therefore anticipate that the subject vehicle will not brake suddenly, will understand what the subject vehicle is a ttempting to do and will respond by gradually decelerating as well. So the field observations show that small gaps are usable to drivers. It also shows that a sudden, substantial reduction in speed does not necessarily occur every time as suggested by the simulator. Overall, many test runs have been done with this software and the results have been analyzed in attempts of making the simulation match expected speed flow relationships and curves based on the US Highway Capacity Manual. The gap acceptance func tion in relation to the lane changing process has been given much attention. Understanding how drivers utilize freeway gaps in a traffic stream is important in having accurate simulation results.

PAGE 42

42 Gap Acceptance Gap acceptance and lane changing are two key components of every micro simulator. These two components are interrelated as one cannot be studied in isolation of the other. A simplified structure of a lane change process has been diagrammed in the flowchart seen in Figure 2 6 to depict the interaction between these two components. Note that the means by which the change from the current lane to either the right or left lane is through gap acceptance. Models frequently execute lane changes through a gap acceptance model. The second level of the flow ch art in Figure 2 6 is where the left lane or right lane gap is evaluated. It is at this stage of the gap acceptance process that one begins to see variation among gap acceptance models. Many factors must be considered when attempting to evaluate whether a g ap should be accepted or rejected. Some gap acceptance models evaluate the available gap by using critical gaps. Others incorporate the amount of braking that will be required if the subject vehicle accepts the gap. This degree of deceleration induced by t he acceptance of the gap is then compared to a pre specified threshold value. Examining the relative speed of the vehicles affected by a gap being accepted is another way to evaluate whether a gap should be accepted or rejected. This variability shows that the flowchart above in Figure 2 6 is a general structure. In reality, many gap acceptance models can become quite involved. The critical gap is an important part of many gap acceptance models. The critical gap is the minimum size gap that a driver is wil ling to accept. The difficulty with this is that every driver and situation does not have the same critical gap I t depends on the vast diversity in driver types and lane change situations. Much research has been done

PAGE 43

43 in different ways to determine a distr ibution for critical gaps. The following is a brief review of different studies directly related to gap acceptance on freeways. Raff (1950) performed one of the early studies related to gap acceptance. His study involved collecting data of accepted and rej ected lag gaps at 4 different side street stop controlled intersections. While his study was not related to lane changing on freeways directly, his fundamental method for determining the critical gap is still used by many researchers for various applicatio ns. Raff determined critical gap by plotting the number of lags against the length of the lag in seconds. The critical lag was the intersection of the rejected lag curve and the accepted lag curve. An example of this is provided in Figure 2 7 A rejected g ap can be directly observed by the researcher for a stop cont rolled side street situation Unfortunately, for a study on freeway lane changing gap acceptance, rejected gaps are much more difficult to detect. Ramsey and Routledge (1973) formed histograms o f all accepted and rejected gaps. These histograms were then utilized to estimate the probability of critical gaps across a population of drivers. Many other early studies have been done to determine the distribution of critical gap s. Herman and Weiss (196 1), Blun den et al. (1962), Drew et al. (1967), Miller (1972), and Daganzo (1981) used exponential, gamma, lognormal, normal, and normal distribution respectively to capture the variation seen in critical gaps. According to Ahmed et al. (1996) individuals do not evaluate all gaps equally. The number of gaps rejected by an individual should be taken into consideration when determining what size gap an individual is willing to accept. In other words, a driver tends to be more apt to accept a smaller gap if th ey have missed opportunities of

PAGE 44

44 previous available gaps. A gap acceptance model should capture this concept by incorporating how many gaps have been considered by the driver. Brackstone et al. (1998) developed a potential model for gap acceptance for freew ay lane change maneuvers and tested it in a microscopic simulation program. They note the importance of understanding lane change gap acceptance as it relates to the overall performance of a freeway. They say that there is evidence that the lane changing p rocess may be the cause of the largest number of crashes on a freeway. As the number of vehicle s using a freeway increases, the number of acceptable gaps decreases. Drivers who are seeking to change lanes will therefore have fewer gaps available to choose from. As a result, they will be more likely to take a higher risk which can potentially cause a crash or a breakdown. In order to better understand lane change gap acceptance, Brackstone et al. performed an experiment involving an instrumented vehicle. Th e vehicle is equipped with the ability to record speed, location, and video footage of the surrounding environment. The data collected from this instrumented vehicle was used to measure the variability of gap acceptance behavior among drivers. During the e xperiment a participant drove the vehicle while the researcher was in the vehicle with them. The experiment took place on a three lane freeway. At random intervals the participant was asked whether they were considering a lane change or not. If they were c onsidering a lane change then the position of the vehicle and surrounding vehicles were marked and it was determined whether the gap was rejected or accepted. Two drivers participated in the experiment which yielded a total of 174 rejected gaps and 147 acc epted gaps. This data along with other data related to the lane change process was used to formulate a

PAGE 45

45 lane change gap acceptance model. This model which is tailored for a three lane freeway was implemented in the simulation model, FLOWSIM. The results of the implemented model show that it is possible to generate a lane change gap acceptance model, through in vehicle data collection, that can accurately model driver behavior. Toledo et al. (2003) presents a gap acceptance model as part of an integrated lan e changing model. The intent was to integrate mandatory lane changes with discretionary lane changes rather than isolate them. The gap acceptance model presented is used as the model through which lane changes are executed. In their gap acceptance model th e subject, lead, and lag vehicles are involved. The conceptual process consist of the driver of the subject vehicle evaluating the positions and the speeds of the lead and lag vehicle in the target lane. Upon processing this information the driver then dec ides whether the gap between the lead and lag vehicle is enough for a successful lane change to be completed. They further explain that most gap acceptance models are binary choice models in which the driver, when presented with an available gap, must cho ose to either accept or reject it. This decision is made by the gap acceptance model through the use of critical gaps. In attempt to replicate the diversity of critical gaps among drivers they are modeled as random variables. Furthermore, the critical gaps are assumed to follow a lognormal distribution as this ensures that all the generated values will be nonnegative. So if the available gap is greater than the randomly generated critical gap then the gap is accepted, however, if the available gap is less t han this critical gap then it is rejected. This simple binary choice structure is what every lane change process must be executed through in their model.

PAGE 46

46 Overall Toledo et al. concluded that the gap acceptance model sufficiently modeled driver behavior. U sing random variables for lead and lag critical gaps assisted in capturing differences in driver types. Also it was determined that using lead and lag critical gaps for the threshold of gap acceptance or rejection is one viable option. A limitation seen in this gap acceptance model is that other factors may come into play in the rejection of an available gap other than a critical gap threshold. These other factors that contribute to gap rejection are noted later in this paper in the section discussing s work on mandatory lane change behavior on multilane freeways. In analyzing gap acceptance Hidas (2005) noted that the lead vehicle is essentially passive during the process. The subject vehicle and the following vehicle are the key components in analyzi ng the process in which a gap is accepted or an unacceptable gap is created into an acceptable one. Hidas evaluated the willingness of the following vehicle to slow down in order to provide a sufficient gap. Theoretically speaking, this willingness should to minimize their total travel time. However, as observed in the field, vehicles do slow down to provide sufficient gap to a lane changing vehicle. This willingness of the following vehicle depe nds on aggressiveness, driver experience, the emotional or and what the downstream traffic conditions are. These are factors that should be taken into consideration w hen constructing a gap acceptance algorithm for simulation software. Bham (2008) performed an analysis of freeway gap acceptance using vehicle trajectory data in order to determine effective statistical distributions for time gaps that

PAGE 47

47 could potentially be used to improve microscopic simulation models. The distribution of critical gaps that is used in simulation models is important in order to produce results that closely resemble the observed field data. He highlights that it is important to get an appropr iate value of a rejected gap in order to have an accurate value for a critical gap. For an unsignalized intersection, a vehicle is at a complete stop and is waiting for a gap in the traffic stream to accept. A rejected gap is observable in this case as th e vehicle either stays in a stopped position or proceeds to accept the gap. On a freeway, however, determining if a gap is rejected is not as straightforward. Since on a freeway the driver is attempting to identify a suitable gap while in motion it cannot be determined when the driver makes the decision to either accept or reject an available gap. Furthermore, Bham notes that drivers may reject a gap on a freeway for a number of reasons. Many gap acceptance models base the decision to reject or accept a ga p on the comparison between the available gap and the critical gap. However, just because an available gap exceeds the critical gap does not necessarily mean that a driver will always accept it. It can be observed in the field that drivers reject gaps for various reasons. Some of these reasons include: avoiding collisions between the lead and following vehicle, inappropriate alignment with the gap, significant difference in relative speed, driver misperception of a gap, etc. Maximum likelihood estimation was used to estimate the leading and trailing critical gaps from the field data. Data sets from a NGSIM project on I 80 in Emeryville, California were analyzed. There were 976 total lane changes analyzed. Of these, 58% were for uncongested conditions and t he remaining 42% for congested conditions. It was concluded that mean rejected gaps and median rejected gaps are not the best

PAGE 48

48 estimation for critical gaps when dealing with freeway conditions. Rather, the largest rejected gap less than the accepted gaps is recommended as an estimation of critical gaps. Also a Kolmogorov Smirnov (KS) test was used to test the goodness of fit for the gamma and lognormal distribution. A distribution function was calculated for both the field data and the model. The null hypot hesis was that the field data and the model are equivalent. Since the KS value observed exceeded the KS value from the table it was concluded that a gamma distribution is the best fit for both the leading and trailing time gaps. It is suggested that a gamm a distribution be used in microscopic traffic simulation models as it accurately represents driver behavior of gap acceptance. This study is limited in that it focused solely on mandatory lane changes. Summary The literature review has provided a broad vie w of what has been accomplished concerning lane changing models for freeways. Studies involving different lane change models and gap acceptance models have been reviewed. Many of the studies use gap acceptance models to execute the lane changing process. M ajority of the studies also cite the importance of including the various types of lane changes that exist: discretionary, mandatory, merging, cooperative, and forced. The vast number of studies shows the importance of this topic. It also shows the complexi ties and differing approaches that have been taken to formulate lane changing models. There are many similarities and differences among each model. To demonstrate this, a chronological summary of the various studies discussed in the literature review has b een provided in Table 2 1, Table 2 2, and Table 2 3

PAGE 49

49 Figure 2 1. Gap Acceptance Terminology

PAGE 50

50 Figu re 2 2. Gipps Lane Changing Model D ecision P rocess F lowchart

PAGE 51

51 Figure 2 3. Ahmed et al. Lane Chan ging Model Structure

PAGE 52

52 Figure 2 4. Lane Change Structure with Explicit Target Lane Choice Figure 2 5 Lane Changing R ate and Traffic D ensity Relationship (Kesting et al., 2007)

PAGE 53

53 Figure 2 6 Structure of a General Lane Change Process Figure 2 7. Distribution of Accepted and Rejected Lag Gaps

PAGE 54

54 Table 2 1. Summary of Literature on Lane Changing Models Author(s) Year Notes Gipps 1986 One of the first lane changing models Dri and being in the proper lane for a turning movement Ahmed et al. 1996 Based lane change model on three steps: decision to consider a lane change, choice of left or right lane, and accept ing a gap in the desired lane. Distinguishes between MLC versus DLC Hidas 2002 Implemented lane changing model in the micro simulator SITRAS Distinguishes between MLC versus DLC Incorporates cooperative lane changing in MLC Toledo 2003 Used random varia bles for lead and lag critical gaps to capture differences in driver types. Used lognormal distribution for randomly generated critical gaps to ensure they will all be non negative Hidas 2005 Proposed that lane changes be classified as free, forced, or co operative. Ben Akiva et al. 2006 Suggests the integration of MLC and DLC Suggests an explicit target lane choice not just a direction of lane change choice. Lee 2006 Used MITSIMLab to compare a lane changing model that uses courtesy and forced lane chang es with one that does not Concluded that the model incorporating courtesy and forced lane changes better represented observed field data Kesting et al. 2007 Proposed a lane change model that is executed by minimizing the overall braking induced by a lane change As the traffic density increases, the number of suitable gaps decrease, which causes the lane change rate to decrease. Bham 2008 Notes that drivers may reject a gap for various reasons in addition to gap size: inappropriate alignment with the gap, avoiding collisions between the lead and following vehicle, significant difference in relative speed, and driver misperception of a gap Concluded that mean rejected gaps and median rejected gaps are not the best estimation for critical gaps when dealing wi th freeway conditions. Rather, the largest rejected gap that is less than the accepted gaps is recommended as an estimation of critical gaps. Sun & Elefteriadou 2010 Incorporated driver behavior and driver characteristics into a lane change model using da ta obtained via a focus group and in vehicle experiment. Implemented the model into CORSIM and found that the interlane travel time differences, lane use, and cumulative number of lane changes were modeled more accurately

PAGE 55

55 Table 2 2 Summary of Lane C hange Duration Literature Author(s) Year Range (seconds) Mean /Median (seconds) Notes Worrall & Bullen 1970 2.3 to 4.1 Median = 3.2 Underestimated due to poor image resolution Finnegan & Green 1990 4.9 to 7.6 Median = 6.3 Included visual search time Hetr ick 1997 3.4 to 13.6 Mean = 6.0 16 participants drove an instrumented vehicle accompanied by a researcher along urban streets and freeways Tijerina et al. 2005 3.5 to 8.5 Mean = 5.8 Participants drove an instrumented vehicle along a highway accompanied by a researcher Tijerina et al. 2005 3.5 to 6.5 Mean = 5.0 Participants drove an instrumented vehicle along urban streets accompanied by a researcher Salvucci & Liu 2002 Mean = 5.14 11 participants were asked to use a driving simulator to drive through a multilane highway Lee et al. 2003 Mean = 6.28 16 participants used instrumented vehicles for 10 day each to go about their daily errands. In vehicle observers were not needed as the instrumented vehicles automatically gathered lane changing. 8,667 lane changes were recorded. Toledo & Zohar 2007 1 to 13 Mean = 4.6 Duration of 1,709 lane change maneuvers was extracted from video recordings along I 80 in Emeryville, California.

PAGE 56

56 Table 2 3 Summary of Lane Change Gap Acceptance Literature Author (s) Year Accepted Gap Distribution Critical Gap Distribution Notes Herman and Weiss 1961 Exponential Did not capture the effect of previously rejected gaps on critical gaps. Blun den et al. 1962 Gamma Initially set out to model a critical gap distribu tion but actually modeled an accepted gap distribution instead. Drew et al. 1967 Lognormal Miller 1972 Normal Daganzo 1981 Normal Toledo et al. 2003 Lognormal Using random variables for lead and lag crit ical gaps assisted in capturing differences in driver types. Bham 2008 Gamma Gamma This study focused solely on mandatory lane changes. Used KS test to test the goodness of fit for both gamma and lognormal distribution.

PAGE 57

57 CHAPTER 3 DATA COLLECTION Part o f the data used in this Variable Speed Limit (VSL) Best Management Practice FDOT Contract BDK77 TWO977 11 (UF Project 88592) This project focused on improving the VSL system that is currently used on Interstat e 4 in Orlando, Florida. Another data set used in this study is from a former University of Florida phD student, Alexandra K ondyli. She collected data via the same instrumented vehicle along Interstate 95 in Jacksonville, Florida to develop her dissertatio Ramp Merges. The following chapter discusses the data collection process. Details concerning the instrumented vehicle, the research locations, the research participants, the data extraction methods, and qualitati ve observations made from the video data are given. Instrumented Vehicle A primary data collection tool used in the FDOT VSL project and the I 95 freeway ramp merging dissertation was the University of Florida Transportation Research t S UV. This vehicle provides continuous, real time collection of data which can easily be reviewed and analyzed later. This is made possible via the HTDR400 Honeywell Mobile Digital Data Recorder. The system consists of four on board digital cameras that are strategically positioned to collect video data of activities surrounding the vehicle. There is a digital camera in the front, rear, left, and right side of the vehicle. The vehicle is equipped with GPS which provides continuous collection of speed and loca tion of the vehicle. The system also indicates when the driver performs basic driving functions such as braking, accelerating, or using the turning signal.

PAGE 58

58 Upon collecting the data it is then downloaded to a laptop that is equipped with BusView software. This software allows for the video data to be viewed and manipulated. Once the data is downloaded into the BusView format it can be accessed at a later time for further analysis. Figure 3 1 has been included to clearly portray the setup of the instrumente d vehicle. The instrumented vehicle provides the research participants with a normal vehicle to drive. Although the vehicle is equipped with technology unfamiliar to the participants, they are essentially unaware of anything unusual and are able to drive t he vehicle as they would any other vehicle. Research Locations I 4 Orlando, FL The freeway used for this portion of the research is I 4 in Orlando, Florida. Figure 3 2 shows the section of freeway between Conroy R oa d and Maitland B oulevard approximately 1 1 miles. Research participants, discussed in the next section, drove 11 miles one direction and then turned around and drove 11 miles back to the starting location for a total of approximately 22 miles of freeway driving. This section is favorable for stud ying lane changing as it is a major interstate that runs right through downtown Orlando. It has both regularly occurring congested areas and uncongested areas. There are also many on ramps and exit ramps along these 11 miles of freeway which creates scenar ios for both mandatory and discretionary lane changes. Overall, by using this corridor participants will be placed in a driving environment that captures many different situations within a relatively short distance of freeway. I 95 Jacksonville, FL The f reeway used for this research was I 95 in Jacksonville, Florida. This study focused on developing a freeway merging algorithm based upon the data collected by

PAGE 59

59 the same instrumented vehicle used for the O rlando study. Research participants, details provided in the next section, were accompanied by a researcher. The participants were instructed to drive along a pre specified route. Figure 3 3 shows the AM route from Philips Highway to Bowden Road, approximately 6.5 miles. M orning participant s would drive this route and then return to the starting point. They would complete this course twice. Figure 3 4 shows the PM route from Baymeadows Roa d to Bowden Road, approximately 3.8 miles. Afternoon participants would complete this round trip course three times. These routes were chosen in order to fulfill a pre specified criterion of the freeway merging study (Kondyli and Elefteriadou, 2009). Research Participants The FDOT VSL study involved 15 participants. Participants of this previous study drove the instrumented v ehicle along a specified stretch of freeway on Interstate 4 in Orlando, Florida. Each participant was accompanied by two researchers who made observations during the experiment. A total of 15 participants performed the driving task. Before beginning the d riving portion of the experiment each participant was required to complete two forms. The first form was the Prescreening Questionnaire for Freeway Driving Research. The purpose of this form was to gain information about the driver such as their gender, ag Background Survey which is aimed at obtaining information concerning the driver performance characteristics of e ach participant (see Appendix A for actual forms used). Once these preliminary forms were completed the participant drove the specified section of I 4 having no knowledge of what the research consisted of.

PAGE 60

60 dy involved 31 participants. The procedure of the in vehicle experiments were similar to those done in Orlando. The participants of this study drove the instrumented vehicle along a specified path of Interstate 95 in Jacksonville, Florida. Each research pa rticipant was accompanied by at least one researcher who made observations during the study. Similar forms to the Background Survey were used for each experiment. In this study t he participants were told to drive how they normally would and they were asked to narrate their driving by talking out loud through their thought process of every maneuver made. This helped the researcher to gain more insight as to why the driver performed certain lane changing or merging maneuvers. For the Orlando and Jacksonville experiment, each participant drove during the morning or afternoon peak periods. Since the data was collected in this fashion, congested conditions occurred multiple times durin g the experiments. This was helpful in observing differences in lane changing behavior between congested and uncongested conditions. The background information obtained from the preliminary discretionary lane changes made per mile was used to identify each participant as a certain driver type. This process was done via a cluster analysis procedure outlined in Chapter 5 Data E xtraction The data collection phase of the research was completed as part of the FDOT VSL Jacksonville. The data from Orlando consists of 15 videos of participants that have driven the instrumented vehicle the pre specified 22 mile round t rip route along I 4. The

PAGE 61

61 data set from Jacksonville consists of 31 videos of participants that have driven the instrumented vehicle along the pre specified route along I 95. The next phase of the research involved extracting the desired information from th e large quantity of collected raw data. The information of interest from the collected data of each participant includes: quantity of lane changes made, duration of each lane change made, the desired speed, and accepted gap sizes. Lane C hanges The number o f discretionary and mandatory lane changes made was extracted from the video for each participant. Furthermore, the speed and location of the vehicle during each lane change was recorded along with whether the participant accelerated, decelerated, or remai ned at a constant speed during the lane change. This provided insight to the motivation behind the lane change. It was also noted, where appropriate, whether it appears that the driver is changing lanes for a speed advantage, to avoid an on ramp merging si tuation, to avoid interaction with a semi truck, or to move into position for a turning movement. In addition to quantifying how many lane changes were made by each participant and what type each lane change is classified as, the duration of every lane ch ange was measured from the videos. The definition of the duration of a lane change used in this thesis is: the time it takes from the point the vehicle first moves laterally until the vehicle is centered in the destination lane (Olsen et al. 2002). The dur ation was measured by watching the videos and using a stopwatch to time the start and end of each lane change This data was then used to evaluate if there is any correlation between the time it takes individuals to complete a lane change maneuver and the driver type.

PAGE 62

62 Desired S peed The desired speed of each participant was based upon the speed they chose to drive when under free flowing and not car following conditions (Kondyli and Elefteriadou 2009). To further confirm the desired speed, the speed of the v ehicle was extracted from the video data in uniform increments of distance. From this information a smooth speed plot over the distance the vehicle has travelled was constructed. The average speed at which the participant desires to drive when unobstructed by surrounding vehicles can be confirmed from the speed distance plots. These plots show the trajectory of the vehicle over the entire distance of the trip. This also helped in segregating the data into congested and uncongested conditions as the drop in speed was evident in the plots. Gap A cceptance behavior: accepted gaps and rejected gaps. A gap is the distance between the lag and lead vehicle. If the subject vehicle desires to cha nge lanes then the driver will evaluate the size of an adjacent available gap. Then the driver of the subject vehicle will either accept the available gap and change lanes, or reject it and remain in the current lane. An accepted gap is easier to record th an a rejected gap. It is directly observable to the researcher when a driver accepts a gap. However, rejected gaps are often latent as they are not readily observable and there may be no indication that a driver rejected a gap. For this research, rejected gaps were recorded when a participant used the turning signal, indicating the desire to change lanes, but then remains in the current lane. Furthermore, the researcher observed signs that the driver gave that indicated they desire to change lanes. At times the driver verbalized their frustration of not being able

PAGE 63

63 to get over into the other lane or they would constantly look over their shoulder and in their mirrors evaluating the gap. A gap was classified as rejected if the driver performed these behaviors a nd the n remained in the current lane. Using these definitions of accepted and rejected gaps; the size of every gap for each lane change was extracted from the videos. This was accomplished by calibrating the digital cameras to a known distance. Once the di gital cameras were properly calibrated, the distance from the digital camera to any object within the image of the video can be obtained. See Appendix B for a detailed description of how the gaps were measured from the video data. The lead gap measurement is from the front digital camera to the back of the lead vehicle. The lag gap measurement is from the rear camera to the front of the lag vehicle. These two measurements have been made for every lane change completed. By obtaining the accepted gap sizes fr om the videos and rejected gap sizes an estimation of the critical gap can be calculated for each driver. However, as discussed in more detail in Chapter 5 and 6, the number of rejected gaps observed was not sufficient to determine the critical gaps of the drivers. Qualitative Video Data Observations Many qualitative observations were made while in the instrumented vehicle with the research participants during the experiments and while viewing all of the video data. The following observations provide inform ation on some specific driver behaviors related to lane changing on freeways. These behaviors should be considered when developing freeway lane changing models. Double Lane Changes A double lane change in this paper refers to when a driver begins moving la terally and continues this movement until they are centered in the lane that is two lanes over

PAGE 64

64 from the initial lane. From the vi deo data, nine double lane changes and one triple lane change was observed. The majority of these maneuvers occurred as the dri ver was merging onto the freeway in very light traffic. Rather than merging into the right lane they would continue their lateral movement until they were in their desired lane. Explicit Target Lane Choice Ben Akiva et al. (2006) states that a limitation w ith many lane changing models is the absence of an explicit target lane choice feature Many models only analyze the immediate adjacent lanes to determine if the lane change should occur or not. However, it was observed in the field that drivers do not eva luate the immediate adjacent lanes only but evaluate all lanes. In this research there were two occurrences where a driver verbalized that the favorable conditions existed two lanes over. They further expressed that they were willing to accept less than fa vorable conditions in the adjacent lane in order to eventually occupy the more desirable far left lane. Lane changing models that disregard this behavior may under predict lane changes as the simulated vehicle will be constrained by the conditions in the i mmediate adjacent lanes only. Avoiding On ramp Merging Vehicles There were14 observations of drivers changing lanes to provide space for on ramp merging vehicles. Some drivers would remain in their lane and stated that it is the responsibility of the on ra mp merging driver to adjust to the freeway vehicles. However, majority of drivers adjusted for on ramp merging vehicles by changing lanes or adjusting their speed. Interactions with Heavy Vehicles There were multiple instances in which the driver verbalize d their discomfort with being near heavy vehicles and that they try to speed up to get away from them. Other

PAGE 65

65 than discomfort many drivers stated that they did not desire to be behind heavy vehicles as they tend to drive slower. Another observation concer ning the interaction of passenger cars with heavy vehicles was the u tiliz ation of the space gap in front of heavy vehicles. There were two instances when a driver stated that they often use the extra space that is i n front of heavy vehicles when changing l anes if such an opportunity presents itself. While this behavior may not be the safest, as heavy vehicles need more braking distance which is why they maintain a greater following distance in the first place, it was an observation nonetheless. Home Lane o f Driver return to and attempt to remain in while on the freeway. One research participant verbalized that they like being in the middle lane, for example This participant further explained that they will change lanes when necessary but that they always attempt to return to the middle lane. Similar behavior was observed in other drivers This suggests that n additional factor when describing indivi duals lane change behavior observation worth noting as it could be further researched and possibly incorporated into lane change algorithms. Summary In this chapter the data collection pr ocess was covered. In summary, video data from I 4 in Orlando and I 95 in Jacksonville were used. The video data were the result of research participants driving along a pre specified route in either the Orlando or Jacksonville study location. Surveys were used to gather background information on each research participant.

PAGE 66

66 lane changing, desired speed, and gap acceptance behavior. Specifically l ane change durations and accepted gap sizes were measur ed from the videos. Lane change durations were measured using a stopwatch from the point the vehicle first moves laterally until the vehicle is centered in the destination lane. Accepted gap sizes were measured using the camera constants and the known heig ht of the camera from the ground. The details of this process are shown in Appendix B. Lastly; specific observations of driver behavior during freeway lane changing were discussed. Collecting the large volume of data and getting it into a usable form to b e analyzed was the goal of this section of the research process. Figure 3 1 Instrumented 2006 Honda Pilot SUV

PAGE 67

67 Figure 3 2 Orlando Study Route: Interstate 4 from Conroy Rd. (A) to Maitland Blvd. (B) (11 miles)

PAGE 68

68 Figure 3 3 Jacksonville Study AM route: Interstate 95 from Philips Highway (A) to Bowden Road (B) (6.5 miles)

PAGE 69

69 Figure 3 4 Jacksonville Study P M route: Interstate 95 from Baymeadows Road (A) to Bowden Road (B) (3.8 miles)

PAGE 70

70 CHAPTER 4 DATA ANALYSIS AT AN AGGREGATE LEVEL T h is chapter ana lyzes the data at an aggregate level focusing on lane change types and traffic conditions A total of 726 lane changes were identified. Of these 321 ( 44.2 %) were discretionary and 405 ( 55.8 %) were mandatory. Also 639 ( 88.0 %) occurred in un congested condit ions and 87 ( 12.0 %) occurred in congested conditions. H ypothesis tests are conducted to determine whether any significant differences exist between quantifiable characteristics of lane changing such as durations and gap acceptance sizes ; for different lan e change types and congestion conditions Also, fitted distributions of lane change durations and lag gaps for different lane change types and different congestion conditions are explored and compared to the literature. The aim of the analys is in this sect ion is to obtain results from the data as a whole, without relating it to the driver performing the lane change maneuvers. The data analysis by driver type is discussed in Chapter 5. All analysis was done using JMP 8 statistical software. Hypothesis Testin g The specifics of the collected data, durations and accepted gap sizes, for each lane change type and traffic condition is provided in Appendix C. A t wo s ample t t est was done on various combinations of this lane changing data. The purpose of the hypothes is testing for multiple scenari os was to determine if any significant difference exists between the means of two groups. Identifying differences between quantifiable characteristics of lane changing is useful when attempting to develop models. All tests we re conducted using a t wo s ample t t est at a 5% level of significance ( : Each of the scenarios was tested for Orlando data, Jacksonville data, and the two data set s combined. For simplicity the summary of the main conclusions of the

PAGE 71

71 hypothesis test ing process ha s been summarized. The following statements are base d upon the t wo sample t t est for the Orlando and Jacksonville combined data set at 5% level of significance ( = 0.05). Mandatory lane changes take longer to complete in congested (7.72 sec) conditions than uncongested (5.29 sec) conditions The average size of accepted lag gaps are smaller in congested (45.66 ft) conditions than uncongested (87.26 ft) conditions There was no significant difference between the durations of the uncongested disc retionary lane changes to the left (5.25 sec) compared to the right (5 .24 sec) There was no significant difference between discretionary lane c hange durations in congested (5.58 sec) and uncongested (5.25 sec) conditions There was no significant difference between mandatory (excluding merging maneuvers) and discretionary lane change durations in congested (MLC: 7.27 sec, DLC: 5.58 sec) or uncongested (MLC: 5. 39 sec, DLC: 5.25 sec) conditions The first three results listed above seem reasonable. However, the last two results were surprising. It was expected that discretionary lane changes would take longer to complete in congested conditions. Another expected result was that mandatory lane changes would take longer than discretionary lane changes as the drive r would perform the maneuver with more urgency. However, the results of the hypothesis test suggest that there is no significant difference in both of these scenarios. The detailed results of all the hypothesis tests are provided in Appendix D. Distributio ns of Lane Change Durations and Gaps Distributions were fit to the lane change duration and gap acceptance histograms The purpose of this process was to analyze the data from a broad perspective and understand what trends exist. The statistical software f its the following distributions to the histogram and chooses the best fit :

PAGE 72

72 Normal Lognormal Weibull Extreme Value Exponential Gamma Beta Johnson Su Johnson Sb Johnson Sl Different combinations of the data concerning lane change type, DLC or MLC, and whe ther traffic conditions were uncongested or congested were used. The following categories of data were analyzed and fitted to a distribution. Groups of lane change duration d ata that were f itted to a d istribution : Uncongested DLC Durations Congested DLC Du rations Uncongested and Congested DLC Durations Uncongested and Congested DLC and MLC Durations Groups of a ccepted g aps d ata that were f itted to a d istribution : Uncongested and Congested DLC Lag Gaps Uncongested and Congested DLC Lead Gaps Uncongested and Congested DLC and MLC Lag Gaps Uncongested and Congested DLC and MLC Lead Gaps Uncongested Lag Gaps Congested Lag Gaps T he distributions that were fitted to each group of data for durations and gaps can be seen in Table 4 1 and Table 4 2 respectively A complete summary of statistics for all of these data groups are provided in Appendix E. Distributions of Lane Change Durations Conclusions In general, the results suggest that a Normal distribution be used to represent discretionary lane change durations. All of the duration distributions were fitted to a

PAGE 73

73 This group of data contains all 726 lane changes observed and is shown in Figure 4 1 This distribution closel y resembles the lognormal distribution found by Toledo and Zohar (2007) shown in Figure 4 2 It also is similar in shape to the histogram presented by Hetrick (1997) in Figure 4 3 Both this study and the study done by Toledo and Zohar (2007) and Hetrick ( 1997) suggest that lane changes take between 1 to 14 seconds with the mean approximately occurring around 5 or 6 seconds. The histograms for discretionary lane change durations were all fitted to normal distributions. The histogram containing all lane chan ges both discretionary and mandatory, was fitted to a Johnson Su curve and was very close to being fitted to a lognormal curve. The difference between the discretionary lane change duration distributions and the distribution for all lane changes, both dis cretionary and mandatory, can be explained by there being occurrences of congested MLC merging maneuvers, which had the highest duration. These occurrences contributed to the histogram having a longer tail to th e right. The Johnson Su distribution is an u nbounded, infinite, asymmetric, continuous type distribution with four parameters The software selected this curve as the best fit for the data. For the purposes of simulation it may be a cumbersome distribution to incorporate in a lane changing model. Th erefore, a lognormal distribution could be used to yield similar results. Distributions of Lead and Lag Gaps Conclusions For the analysis of the lead and lag gaps it appears that a Gamma distribution should be used to represent lag gaps E ach of the groupe d data were fitted to a Gamma

PAGE 74

74 curve. Both data groups of lead gaps were fitted to the J ohnso n Sl distribution. Much of the literature reports on lag gap acceptance values only and not on lead gaps. Blunden et al. (1962) and Bham (2008) both obtained result s that fitted a Gamma distribution to their lag gap acceptance data. Blunden et al. (1962 ) had the initial objective of obtaining (2008) scope involved mandatory lane chan ges only. Overall, the results obtained in this study are supported by the literature. Summary This chapter specific ally discussed the data a s a whole without considering the effect of driver types Hypothesis testing was used to identify any significant d ifferences between lane change durations and gap acceptance sizes. The details of the hypothesis testing results are provided in Appendix D For lane change durations, discretionary and mandatory lane changes were tested for differences as well as duratio ns for congested and uncongested conditions. Lag gap acceptance sizes for congested and uncongested conditions were tested. Detailed statistical summaries of all the distribution fittings are provided in Appendix E. In general, the results obtained from t his portion of the data analysis are supported by previous studies discussed in the literature review.

PAGE 75

75 Table 4 1. Fitted Distribution for Different Lane Change Duration Data Groups Data Group Fitted Distribution Mean (sec) Std Dev Uncongested DLC Du rations Normal 5.25 0.97 Congested DLC Durations Normal 5.58 1.29 Uncongested and Congested DLC Durations Normal 5.28 1.00 Uncongested and Congested DLC and MLC Durations Johnson Su 5.48 1.35 Table 4 2. Fitted Distribution for Different Lane Change Ga p Data Groups Data Group Fitted Distribution Mean (ft) Std Dev Uncongested and Congested DLC Lag Gaps Gamma 90.49 46.50 Uncongested and Congested DLC Lead Gaps Johnson SI 135.97 168.22 Uncongested and Congested DLC and MLC Lag Gaps Gamma 81.04 45.79 Un congested and Congested DLC and MLC Lead Gaps Johnson SI 113.73 140.57 Uncongested Lag Gaps Gamma 87.26 45.16 Congested Lag Gaps Gamma 45.66 30.96

PAGE 76

76 Figure 4 1. Lane Change Duration Distribution for this paper (N = 726, Johnson Su distributi on)

PAGE 77

77 Figure 4 2 Lane Change Duration Distribution by Toledo and Zohar 2007 (N = 1,790)

PAGE 78

78 Figure 4 3. Lane Change Duration Distribution by Hetrick 1997 (N = 282)

PAGE 79

79 CHAPTER 5 DATA ANALYSIS BY DRI VER TYPE Previous studies have analyzed the mechani cs of lane changing by analyzing large amounts of data collected from high mounted video cameras over freeways (Toledo et al. 2007). Other studies have used inductor loops to study lane changing at a macroscopic level a ttempting to determine the frequency of lane changes and relate it to density (Knoop et al. 2011). Many have used instrumented vehicles with on board cameras to study lane changes (Hetrick 1997; Brackstone et al. 1998; Lee et al. 2003; Tijerina et al. 2005). However, there are few studies th at have specifically attempted to relate the mechanics of lane changing to the type of driver that is performing the maneuver. The aim of this section is to evaluate lane changing by classifying drivers into different categories. Once the drivers are class ified the collected lane changing data is analyzed from the perspective of the different driver types. Driver Classification Overview A main portion of the research is the classification of drivers. The classification was based upon qualitative and quanti tative data. First, the qualitative data consists of survey responses from each participant concerning their driving behavior. The in Appendix A. Second the quantitative data is based on the number of discretio nary lane changes made per mile and the desired speed of the driver when in free flow conditions, not impeded by surrounding vehicles. This quantitative data was used to perform a cluster analysis in order to group the drivers into their classifications. O nce the cluster analysis was completed the classification from the quantitative data was compared to the qualitative data obtained from the survey.

PAGE 80

80 The following sections are organized such that the cluster analysis method is described, the first attempt of using the Orlando and Jacksonville data combined is discussed, the results of using the separated data is presented, a comparison is made to the qualitative data, and the data by driver groups is analyzed. Cluster Analysis Cluster analy sis is used to ca tegorize drivers into groups. The general process for creating groups within a data set involves performing a k means cluster analysis for multiple k values and then determining which value is most appropriate based upon the within cluster sum of squares a nd the Hartigan index. Each of these steps is described in detail in the following sections. Then the results for the Orlando and Jacksonville data combined and separated are presented. K means Clustering K means clustering is a cluster analysis method for grouping data based upon the distance between the individual data points. This approach requires that the number of clusters be specified first. Once the number of clusters is specified, an iterative process begins to determine the centroids of each clust er. The centroids are the average of the to the centroids, the iterations stop and the clusters are established. Upon finish ing the k means clustering process, the software provides the number of points within each cluster and the centroid point of each cluster. The main limitation of k means clustering is that the number of clusters must be determined before the process begins. Since the number of clusters is not known in t he case of this research, it must be determined how many clusters to divide the data into. This can be accomplished by performing an iterative k means clustering process for

PAGE 81

81 different k values. The analysis is run multiple times for different number of clu sters. Once this is completed, the process to determine how many clusters is best can be accomplished by evaluating the within cluster sum of squares and the Hartigan index. Within Cluster Sum of Squares The within cluster sum of squares is a measure of t he squared deviation of the data within a cluster to the centroid of that cluster. Therefore, as the number of clusters increases, the within cluster sum of squares will decrease. Conceptually, if the number of clusters was equal to the number of data poin ts then the within cluster sum of squares would be zero. Equation 3 1 is used to calculate the wi thin cluster sum of squares (3 1) Where: W(k) is the within cluster sum of squares, k is the number of clusters, x j is the data point from the speed ratio and discretionary lane changes per mile, C i i is the centroid point of cluster i. The results of Equation 3 1 can be organized in graphical form. P lots of the number of clusters, k, against the within cluster sum of squares, W(k) provide criteria for making the choice of the most appropriate numb er of cluster s. The number of clusters where adding another cluster will not greatly improve the within clu ster variance should be chosen. The Hartigan Index To further confirm the number of clusters that the data should be broken into, the Hartigan Index can be calculated. The Ha rtigan index (Tibshirani et al. 2001 ; Sun and

PAGE 82

8 2 Elefteriadou 2010) quantifies the amount of dissimilarity that will be removed by adding an additional cluster. When a large decrease occurs by adding an additional cluster, the Hartiga n index is suggesting that the dissimilarity is significantly removed. While a small decrease indicates that the dissimilarity has not been removed as much. The equation for the Hartigan index is shown in Equation 3 2. (3 2) Where: H(k) is the Hartigan Index, k is the number of clusters, n is the sample size of the data or the number of objects to be clustered, W(k)is the within cluster sum of squares for k clusters, and W(k+1) is the within cluster sum of squares for k+1 clusters A l arge decrease in th e Hartigan index when the number of clusters is increased i ndicates that the dissimilarity is largely removed The large decrease indicates how many clusters should be chosen for the specific data being analyzed. Cluster Analysis for Orlando and Jacksonvil le Combined Data Figure 5 number of discretionary lane changes per mile. The speed ratio is defined as the desired speed of the individual divided by the posted speed limit. This ratio was used in orde r to 4 in Orlando where the speed limit is 55 mph. The rest of the data was taken from a previous study done along I 95 in Jacksonville where the speed limit is 65 mph. There is an obvious tren d in the relationship between speed and discretionary lane changes. In general, as the desired speed of an individual increases; the number of discretionary lane changes they make also increases. This trend was expected. It is intuitive that the faster an individual

PAGE 83

83 desires to drive the more likely they will make frequent discretionary lane changes in attempt to maintain their desired speed. K means Clustering for Combined Data The data displayed in Figure 5 1 was analyzed using statistical software. The k means clustering approach was used to classify the combined 46 participants into groups based upon their desired speed and discretionary lane changes. T his was done by performing the k means clustering process for 1, 2, 3, 4, and 5 clusters. The software output for all five k means analyses is summarized in Table 5 1 Once the k mea ns clustering is completed for k = 1, 2, 3, 4, and 5, the within cluster sum of squares and Hartigan index were evaluated to determine how many clusters is best. The within clu ster sum of squares results are shown in Table 5 2 .These results were plotted as seen in Figure 5 2. This plot was then used to determine the most appropriate number of clusters to use. The number of clusters where adding another cluster will not greatly i mprove the within cluster variance shou ld be chosen. In this case, Figure 5 2 suggests that two clusters is the most appropriate to use as adding a third cluster does not have much significance on the within group deviation. To further confirm that two gr oups are appropriate, the Hartigan index was calculated. These results are shown in Table 5 3 Notice the large decrease in the Hartigan index when the number of clusters is increased from one to two. This indicates that the dissimilarity is largely remove d when the data is split from one cluster into two clusters. Overall, the k means cluster analysis recommends that the combined Orlando and Jacksonville data be grouped into two clusters. The final results of the process for the combined data are shown in Figure 5 3.

PAGE 84

84 Interpreting the Results of the Combined Data Cluster Analysis Figure 5 3 represents the final result of the cluster analysis for the combined Orlando and Jacksonville data set. A cluster analysis was performed for 1, 2, 3, 4 and 5 clusters. Th en a plot of the within cluster sum of squares v ersu s the number of clusters was formed. This plot along with the Hartigan index suggested that two groups be used for the combined data. Comparing Figure 5 1 with Figure 5 3 suggests that the Orlando and Ja cksonville data should not be combined. An important observation concerning Figure 5 1 is the difference between the Orlando data and the Jacksonville data. The Orlando data is at the higher end of the spectrum while the Jacksonville data is towards the lo wer end. Essentially, the cluster analysis of the combined data resulted in most of the Orlando data being in the more aggressive group and most of the Jacksonville data being in the more conservative group. This implies that the two research study sites a re intrinsically different and should not be analyzed as a combined data set The Orlando data and Jacksonville data differences can be explained by the physical sites at which the research was conducted. Orlando data was collected from I 4 and Jacksonvill e data from I 95. The speed limits, traffic conditions, and even the focus of the research that the drivers were participating in were all different between each of these sites. The drivers in Orlando tend to have a higher speed ratio because the speed lim it is lower than the speed limit in Jacksonville. Furthermore, the Jacksonville study was focused on freeway on ramp merges a nd did not have as long of freeway driving segments as Orlando did. These differences explain the distinction in the Orlando and Ja cksonville data as depicted in Figure 5 1 Due to these differences it

PAGE 85

85 was decided to perform a cluster analysis for the Orlando data and the Jacksonville data separately. This process is discussed in the following sections. Cluster Analysis for Orlando an d Jacksonville Separated Data K means Clustering for Separated Data Since the combined data set did not seem to produce reliable results, t he k means clustering approach was used to classify the 15 participants of Orlando and 31 participants of Jacksonvill e separately. The same process used for the combined data was applied to the Orlando and Jacksonville data separately. The k means clustering process was done for 1, 2, 3, 4, and 5 clusters The software output for all five k means analyses is summarized i n Table 5 4 and Table 5 5 for Orlando and Jacksonville data respectively. The within cluster sum of squares results are shown in Table 5 6. These results were plotted against the number of clusters and can be seen in Figure 5 4 and Figure 5 5 Figure 5 4 for Orlando suggests that three clusters are the most appropriate. Figure 5 5 for Jacksonville suggests that four clusters are the most appropriate To further confirm the number of clusters that the data should be broken into, the Hartigan Index was calc ulated. The results of the Hartigan index for k = 1, 2, 3, 4, and 5 have been provided in Table 5 7 Notice the large decrease in the Hartigan index when the number of clusters is increased from two to three for the Orlando case and three to four for the J acksonville case. This indicates that the dissimilarity is largely removed in the Orlando case and Jacksonville case when the data is split from two clusters into three clusters and from three clusters into four clusters respectively.

PAGE 86

86 Determining the Numbe r of Driver Groups Based upon the results of the within cluster sum of squares plots and the Hartigan index for the separated data, it is recommended that the appropriate number of clusters for the Orlando data and the Jacksonville data is three clusters a nd four clusters, respectively. However, in order for both the Orlando data and the Jacksonville data to have the same number of driver groups the Orla ndo data has been grouped into four clusters despite the cluster analysis results suggesting three cluste rs. This will allow for the data between the two sites to be more readily compared. Also, it is better to create more groups than suggested by the cluster analysis rather than less. Therefore, the Orlando data grouping was brought up to four clusters to ma tch Jacksonville data. Figure 5 6 and Figure 5 7 display the result of the driver type grouping process based upon Table 5 8 shows the description given to each group. Group 1, Group 2, G roup 3, and Group 4 have been assigned the labels Very Conservative, Somewhat Conservative, Somewhat Aggressive, and Very Aggressive respectively These names were chosen in order to remain consistent with the survey response choices. Table 5 9 shows the number of participants that fall within each group. Survey Results to Cluster Analysis Comparison The next step in the driver classification process is to compare the qualitative data to the quantitative data. The quantitative process involved applying the k means clustering analysis to the relationship between the discretionary lane changes made per mile and the desired speed. Then looking at the results of the k means clustering analysis from the perspective of the within cluster sum of squares and the Ha rtigan index to determine the most appropriate number of clusters. The qualitative side of the

PAGE 87

87 that the quantitative and qualitative data can be compared. Table 5 10 a nd Table 5 11 provide a comparison of the survey responses, field observations, and the results of the cluster analysis. The comparison shows that some of the groupings according to the cluster analysis do not match up with what the driver reported on the survey. It is not expected that there will be a perfect match between these two. Individuals can have an incorrect perception of their own driving behavior and may be reporting how they view themselves in relation to their peer group rather than the entire driver population. Overall, there was a 33.3% and 38.7% consistency between the cluster analysis results and the survey responses for the I 4 Orlando group and the I 95 Jacksonville group e of driver do you Analysis of the Driver Groups Once the data was divided into different driver groups a multiple mean comparison was done to test if the groups are significantly different from each other in terms of their lane change durations and size of gap accepted Lane change duration by driver type The mean lane change durations for each group are summarized in Table 5 12 The software output for the multiple mean comparison s of the lane change durations for each driver group is provided in Appendix G This analysis shows that Group 1 and Group 2 mean lane change duration is significantly different to all other driver groups. Group 3 and Group 4 mean lane change durations are only signific antly different to Group 1 and Group 2 but not to each other. A plot of the mean lane change duration for each driver group is shown in Figure 5 8 For Orlando data the overall trend is that the

PAGE 88

88 mean lane change duration decreases as the driver group numb er increases. In other words, the most conservative driver group has the highest mean lane change duration and the most aggressive driver group has the lowest. It should be noted that Orlando Group 4 consists of one driver who completed 27 lane changes. Th erefore, the variation of having different drivers does not exist for that group. The Jacksonville data follows a similar trend to the Orlando data except that Group 3 mean lane change duration is slightly higher than Group 4. However, Group 1 and Group 2 are still higher than both Group 3 and Group 4 for the Jacksonville data. This data reinforces that the grouping of drivers into 4 groups is reasonable as the results of the multiple mean comparisons and the mean lane change duration plot show that the gro ups have different lane change characteristics compared to each other. Gap acceptance size by driver type The average gap acceptance size of each driver type group was analyzed using a multiple mean comparison. The purpose of this analysis was to determine if any relationship existed between the different driver groups and the size of gap they are willing to accept as part of the lane change maneuver. It was expected that more aggressive drivers would accept s maller gap s than more conservative drivers. Howe ver, the data did not show any conclusive trend. This is because the size of available gaps depends on the number and configuration of vehicles at the time the driver desires to change lanes. Unlike lane change durations, it does not depend on the type of driver performing the lane change maneuver. This explains why there is no trend between the average size of gap and the driver type group. The average gap accepted during congested conditions by the different driver groups was also analyzed. The software o utput for the multiple mean comparisons of

PAGE 89

89 the lag gaps accepted in congested conditions for each driver group is provided in Appendix H. S imilar to the overall average gap accepted, this yielded no conclusive trend. During congested conditions the number of smaller gaps is greater. It was hypothesized that when drivers change lanes under congested conditions the gap acceptance size is more r epresentative of the driver type This is because the smaller gaps during congested conditions are likely closer to t he critical gap of each individual driver This means that each driver would be forced to wait for a gap that they feel comfortable accepting before changing lanes. Therefore, it was hypothesized that the average size of gap accepted during congested condi tions may reflect a trend when compared across driver types. The mean lag gap accepted in congested conditions for each driver group is shown in Table 5 13. However, no trend was found as seen in Figure 5 9 The difficulty with this portion of the research was the collection of rejected gaps. Unlike at a stop controlled intersection, where rejected gaps are readily observable to the researcher, rejected gaps on a freeway are mostly latent. A few rejected gaps were observed when the driver verbalized a desir e to change lanes or the turning signal was used and available gaps were allowed to pass by. However, these events were very infrequent and did not occur for every participant. Without a sufficient distribution of rejected gaps the critical gap for each pa rticipant could not be determined. This was a major shortcoming of the research of this paper. Future research would involve going through the video data and attempting to determine a systematic way for extracting rejected gaps. Once a sufficient distribut ion of rejected gaps is obtained the critical gap

PAGE 90

90 of each driver can be determined. Then the critical gap of each driver group can be compared to determine if there is any significant difference among them. Summary This chapter outlined the process of clas sifying the research participants into groups based on their desired speed and number of lane changes completed per mile. A cluster analysis was performed for the combined Orlando and Jacksonville data and the separated data. It was decided to use the clus ter analysis results for the separated data due to possible effect of the difference in study sites on the grouping process. The final result was that four groups be used for both sites The purpose of creating groups was to relate different driver types t o their lane changing characteristics. A multiple means comparison test was performed on the lane change durations of each driver group. It was determined that the most conservative drivers take the longest time to complete lane changes on average. The gen eral trend was that the mean lane followed in the Orlando data; however, for the Jacksonville data the most aggressive drivers had the second fastest la ne change time, secon d to the somewhat aggressive drivers. A multiple means comparison test was also performed on the average accepted gap size in congested conditions of each driver group. There was no significant trend to report. Without a sufficient rejected gap distributio n the critical gaps were not determined for each research participant. It is hypothesized that a trend may exist between the driver groups and the average critical gap for each group. Future research would be to determine a method for identifying re jected gaps from the video data.

PAGE 91

91 The data analysis by driver type shows that drivers may be classified into groups with other drivers of similar driving characteristics successfully. It also shows that drivers d o not have uniform lane changing characteristics Th e results of this analysis support the concept of modeling lane change s from a driver type approach rather than disregarding the driv

PAGE 92

92 Figure 5 Plot

PAGE 93

93 Table 5 1. Summary of Orlando and Jacksonville Combined Data Cluster Analysis Centroids k, Number of Clusters Cluster Centroid (Speed Ratio, DLC/mile) Number of points within each cluster 1 Cluster 1 1.13, 0.47 46 2 Cluster 1 1.21, 0.79 18 C luster 2 1.08, 0.26 28 3 Cluster 1 1.15, 0.39 15 Cluster 2 1.22, 0.91 13 Cluster 3 1.05, 0.21 18 4 Cluster 1 1.08, 0.17 20 Cluster 2 1.22, 0.71 15 Cluster 3 1.10, 0.53 10 Cluster 4 1.28, 1.96 1 5 Cluster 1 1.09, 0.25 21 Cluster 2 1.22, 0.71 14 Cluster 3 1.11, 0.69 6 Cluster 4 1.00, 0.04 4 Cluster 5 1.28, 1.96 1 Table 5 2 Within Cluster Sum of Squares Results (Combined Data) k, number of clusters W(k), within cluster sum of squares 1 6.35 2 3.01 3 2.30 4 1.21 5 0.76

PAGE 94

94 Figure 5 2. Within Cluster Sum of Squares vs. Number of Clusters Plot (Combined Data) Table 5 3 Hartigan Index Results (Combined Data) k, number of clusters H(k), Hartigan Index 1 48.8 2 13.3 3 37.9 4 24.7 5

PAGE 95

95 Figure 5 3 Cluster Analysis Results f or the Combined Data

PAGE 96

96 Table 5 4 Summary of Orlando Cluster Analysis Centroids k, Number of Clusters Cluster Centroid (Desired Speed (mph), DLC/mile) Number of points within each cluster 1 Cluster 1 66.1, 0.80 15 2 Cluster 1 65.5, 0.68 13 Cluster 2 70, 1.51 2 3 Cluster 1 62.7, 0.60 7 Cluster 2 71, 1.96 1 Cluster 3 68.7, 0.83 7 4 Cluster 1 62.3, 0.42 4 Cluster 2 64.6, 0.81 5 Cluster 3 71, 1.96 1 Cluster 4 69.6, 0.85 5 5 Cluster 1 62.3, 0.42 4 Cluster 2 65.5, 0.75 4 Cluster 3 61, 1.07 1 Cluster 4 71, 1.96 1 Cluster 5 69.6, 0.85 5 Table 5 5 Summary of Jacksonville Cluster Analysis Centroids k, Number of Clusters Cluster Centroid (Desired Speed (mph), DLC/mile) Number of points within each cluster 1 Cluster 1 71.4, 0.31 31 2 Cluster 1 75.6, 0.542 10 Cluster 2 69.4, 0.19 21 3 Cluster 1 70.9, 0.24 19 Cluster 2 65, 0.04 4 Cluster 3 75.9, 0.60 8 4 Cluster 1 65, 0.04 4 Cluster 2 70.9, 0.23 18 Cluster 3 69.1, 0.59 3 Cluster 4 78, 0.57 6 5 Cluster 1 76.6, 0.43 6 Cluster 2 7 0.2, 0.20 18 Cluster 3 62.4, 0.04 2 Cluster 4 69.1, 0.59 3 Cluster 5 78.7, 0.77 2

PAGE 97

97 Table 5 6 .Within Cluster Sum of Squares Results (Separated Data) k (number of clusters) Orlando WCSS Jacksonville WCSS 1 189.15 582.76 2 152.25 319.94 3 35.43 218 .48 4 31.42 136.35 5 12.06 113.74

PAGE 98

98 Figure 5 4 Within Cluster Sum of Squares vs. Number of Clusters Plot (Orlando) Figure 5 5 Within Cluster Sum of Squares vs. Number of Clusters Plot (Jacksonville)

PAGE 99

99 Table 5 7 Hartigan Index Results (S eparated Data) k (number of clusters) Orlando Hartigan Index Jacksonville Hartigan Index 1 3.2 23.8 2 39.6 13.0 3 1.4 16.3 4 16.1 5.2 5

PAGE 100

100 Figure 5 6 Grouped Orlando Participants based on DLC/mile and Desired Speed Figure 5 7 Grouped Ja cksonville P articipants based on DLC/mile and Desired Speed

PAGE 101

101 Table 5 8 Naming Convention of the 4 Groups Group Number Group Name Group 1 Conservative Group 2 Somewhat Conservative Group 3 Somewhat Aggressive Group 4 Aggressive Table 5 9 Number of P articipants in each Group Group 1 Group 2 Group 3 Group 4 Orlando 4 5 5 1 Jacksonville 4 18 3 6 Combined 8 23 8 7

PAGE 102

102 Table 5 10 Comparison of Survey Responses, Field Observations, and Cluster Analysis Results for I 4 Orlando Data Survey Responses Field Observations Cluster Analysis What type of driver do you consider yourself? What type of driver do your fri ends/family consider you? If the speed limit is 70mph what speed are you likely to drive? How often do you change lanes if the vehicle in front is slower? Average Desired Speed (speed limit 55 mph) Discretionary lane changes per mile Driver Type Class ification based upon Cluster Analysis Consistency between Survey Response and Cluster Analysis? Subject 1 Somewhat conservative Somewhat conservative 65 to 70 mph Sometimes 63 mph 0.43 Group 1 No Subject 2 Somewhat conservative Somewhat conservative 65 to 70 mph Sometimes 66 mph 0.85 Group 2 Yes Subject 3 Somewhat conservative Somewhat aggressive 75 to 80 mph Very often 69 mph 1.07 Group 3 No Subject 4 Somewhat conservative Somewhat conservative 70 to 75 mph Sometimes 70 mph 0.85 Group 3 No Subjec t 5 Somewhat aggressive Somewhat aggressive 75 to 80 mph Very often 71 mph 0.71 Group 3 Yes Subject 6 Somewhat conservative Somewhat conservative 70 to 75 mph Very often 64 mph 0.71 Group 2 Yes Subject 7 Somewhat aggressive Somewhat aggressive 75 to 80 mph Very often 63 mph 0.54 Group 1 No Subject 8 Somewhat conservative Somewhat conservative 75 to 80 mph Very often 71 mph 1.96 Group 4 No Subject 9 Somewhat conservative Somewhat conservative 70 to 75 mph Sometimes 60 mph 0.54 Group 1 No Subject 10 Somewhat aggressive Somewhat aggressive 65 to 70 mph Sometimes 61 mph 1.07 Group 2 No Subject 11 Somewhat aggressive Somewhat aggressive 70 to 75 mph Very often 69 mph 0.89 Group 3 Yes Subject 12 Somewhat conservative Somewhat conservative 65 to 70 mp h Very often 63 mph 0.18 Group 1 Yes Subject 13 Somewhat aggressive Somewhat conservative 70 to 75 mph Very often 65 mph 0.71 Group 2 No Subject 14 Somewhat conservative Somewhat conservative 70 to 75 mph Very often 69 mph 0.71 Group 3 No Subject 15 S omewhat aggressive Somewhat aggressive 75 to 80 mph Very often 67 mph 0.71 Group 2 No Percent Consistency 33.3%

PAGE 103

103 Table 5 11 Comp arison of Survey Responses, Field Observations, and Cluster Analysis for I 95 Jacksonville Data Survey Responses Field Observations Cluster Analysis What type of driver do you consider yourself? What type of driver do your friends/family consider you? If the speed limit is 70mph what speed are you likely to drive? How often do you change lanes if the vehicle in front is slower? Aver age Desired Speed (speed limit 65 mph) Discretionary lane changes per mile Driver Type Classification based upon Clus ter Analysis Consistency between Survey Response and Cluster Analysis? 10 Somewhat aggressive Somewhat aggressive 75 to 80 mph Very Often 77 mph 0.51 Group 4 No 47 Somewhat aggressive Somewhat conservative 70 to 75 mph Very Often 71 mph 0.37 Group 2 No 49 Somewhat aggressive Somewhat aggressive 70 to 75 mph Very Often 72 mph 0.66 Group 3 Yes 52 Somewhat aggressive Somewhat aggressive 70 to 75 mph Very Often 68 mph (Rain) 0.44 Group 3 Yes 63 Somewhat conservative Very conservative 70 to 75 mph Sometime s 78 mph 0.88 Group 4 No 65 Somewhat aggressive Somewhat aggressive 75 to 80 mph Very Often 79 mph 0.66 Group 4 No 69 Somewhat conservative Very conservative 70 to 75 mph Sometimes 67 mph (Rain) 0.66 Group 3 No 71 Somewhat conservative Somewhat aggressi ve 70 to 75 mph Sometimes 75 mph 0.29 Group 2 Yes 72 Somewhat aggressive Somewhat aggressive 70 to 75 mph Very Often 78 mph 0.51 Group 4 No 73 Somewhat aggressive Very aggressive >80 mph Very Often 77 mph 0.44 Group 4 No 76 Somewhat aggressive Somewhat conservative 75 to 80 mph Very Often 79 mph 0.44 Group 4 No 23 Somewhat conservative Very conservative 70 to 75 mph Very Often 68 mph 0.29 Group 2 Yes 27 Somewhat aggressive Somewhat aggressive 75 to 80 mph Sometimes 68 mph 0.37 Group 2 No 32 Somewhat a ggressive Somewhat conservative 75 to 80 mph Sometimes 71 mph 0.16 Group 2 No 37 Somewhat aggressive Somewhat aggressive 70 to 75 mph Sometimes 71 mph 0.16 Group 2 No 51 Somewhat conservative Somewhat aggressive 75 to 80 mph Very Often 75 mph 0.16 Group 2 Yes

PAGE 104

104 Table 5 11 Con tinued Survey Responses Field Observations Cluster Analysis What type of driver do you consider yourself? What type of driver do your friends/family consider you? If the speed limit is 70mph what speed are you likely to drive? How often do you change lanes if the vehicle in front of you is slower? Average Desired Speed (speed limit 65 mph) Discretionary lane changes per mile Driver Type Classification based upon Cluster Analysis Consistency between Survey Response and Cluste r Analysis? 59 Somewhat aggressive Very aggressive 70 to 75 mph Sometimes 68 mph 0.37 Group 2 No 60 Somewhat aggressive Somewhat aggressive 70 to 75 mph Very Often 71 mph 0.29 Group 2 No 61 Somewhat aggressive Somewhat aggressive 75 to 80 mph Very Often 74 mph 0.37 Group 2 No 67 Somewhat conservative Somewhat aggressive 75 to 80 mph Sometimes 73 mph 0.16 Group 2 Yes 68 Somewhat conservative Somewhat aggressive 70 to 75 mph Very Often 70 mph 0.16 Group 2 Yes 74 Somewhat conservative Somewhat conservati ve 70 to 75 mph Sometimes 72 mph 0.29 Group 2 Yes 17 Very conservative Very conservative < 65 mph Sometimes 60 mph 0.00 Group 1 Yes 18 Somewhat conservative Somewhat conservative 70 to 75 mph Sometimes 70 mph 0.15 Group 2 Yes 19 Somewhat conservative So mewhat conservative 70 to 75 mph Sometimes 65 mph 0.08 Group 1 No 50 Very conservative Somewhat conservative 70 to 75 mph Very Often 71 mph 0.15 Group 2 No 56 Somewhat conservative Somewhat aggressive 70 to 75 mph Sometimes 67 mph 0.08 Group 1 No 58 Som ewhat conservative Somewhat conservative 70 to 75 mph Sometimes 71 mph 0.15 Group 2 Yes 66 Somewhat conservative Somewhat conservative 70 to 75 mph Sometimes 68 mph 0.00 Group 1 No 70 Somewhat aggressive Very aggressive 70 to 75 mph Very Often 69 mph 0.1 5 Group 2 No 75 Somewhat conservative Somewhat conservative 70 to 75 mph Sometimes 70 mph 0.07 Group 2 Yes Percent Consistency 38.7 %

PAGE 105

105 Table 5 12 Mean Lane Change Durations Group 1 Group 2 Group 3 Group 4 Orlando 6.20 5.42 5.21 4.58 Jackso nville 6.66 5.65 5.13 5.35 Combined 6.39 5.59 5.16 5.22 Figure 5 8 Mean Lane Change Duration by Driver Groups

PAGE 106

106 Table 5 1 3 Mean Lag Gap Accepted in Congested Conditions Group 1 Group 2 Group 3 Group 4 Orlando 44.9 65.1 43.6 58.5 Jackson ville 36.5 38.0 56.3 43.7 Combined 40.0 46.1 52.7 48.6 Figure 5 9 Mean Lag Gap Accepted in C ongested Conditions by Driver Groups

PAGE 107

107 CHAPTER 6 CONCLUSIONS Research Conclusions Lane changing models are an important component of microscopic traffic si mulation Therefore, lane changes have received much attention. Many previous studies have been aimed at understanding various details of the physical lane change maneuver ( Hetrick 1997; Brackstone et al. 1998; Lee et al. 2003; Tijerina et al. 2005; Toledo et al. 2007 ; Knoop et al. 2011 ). However, the majority of studies have disregarded the effect that the type of driver has on the lane changing process. In this thesis, 46 research participants drove an instrumented vehicle and performed a combined total o f 726 freeway lane changes A c luster analysis was performed to categorize each research participant into one of four groups ranging from conservative to aggressive. Drivers were grouped according to their desired speed and discretionary lane changes compl eted per mile. Once the drivers were classified into a specific driver type an analysis was done to determine any trends that exist ed between the different driver types and their lane change behavior. It wa s found that, in general, more conservative drive rs have greater lane change duration s than aggressive drivers. The mean lane change duration for the conservative grouped drivers was significantly different than the aggressive grouped drivers. These results provide support for basing lane changing models on different driver groups. Previous studies that disregard the driver performing the lane change maneuver can only provide results on the external observations based vehicle data. This thesis has shown that the physical lane change maneuver can be relat ed to the type of driver performing the maneuver. This connection

PAGE 108

108 between the lane change and the driver should be considered in the development of lane changing models. In addition to relating lane change duration to the driver groups, another aim was to determine if the critical gap of each driver group was significantly different from each other. Unfortunately, this part of the thesis could not be accomplished due to an insufficient distribution of rejected gaps. Average accepted gaps in congested condit critical gap. However, this did not result in any conclusive trend or difference between the four driver groups. Future research would be to develop a method for identifying rejected gaps in freeway lane change studies. H ypothesis tests were performed with the purpose of determining if significant differences occur for lane change durations and accepted gap sizes for different lane change types and congestion conditions. The hypothesis tests suggests that a mandatory lane change in congested conditions takes longer to complete than in uncongested conditions. This seems reasonable as the speed of the vehicle is lower during congestion. However, discretionary lane changes in congested and uncongested co nditions had no significant difference. This may be because in congested conditions, for discretionary lane changes, the driver waits for an opening and then begins the lane change. This would cause a shorter duration than mandatory lane changes; where the driver typically begins the maneuver prior to their being a sufficient opening in the target lane. The average lag gap accepted in congested conditions is less than the average lag gap in uncongested conditions. This is intuitive; a greater number of vehi cles on the freeway imply the frequency of smaller gaps increases. T he lane change duration to the

PAGE 109

109 right had no significant difference when compared to the lane change duration to the left. This result was expected. Finally, discretionary lane changes and mandatory lane changes under similar congestion conditions did not come back as significantly different. The same statistical software was also used to fit distributions to lane change duration and gap acceptance histograms. All 726 lane changes observed were fit to a Johnson Su distribution with a mean duration of 5.48 seconds. Th is distribution is similar in shape to the lognormal results reported by studies done by Hetrick (1997) and Toledo and Zohar (2007) Both of these studies and the results of this paper suggest that lane changes take between 1 to 14 seconds with a mean of approximately 5 or 6 seconds. The lag gap acceptance histogram in both congested and uncongested conditions was fitted to a Gamma distribution. These results agree with Blunden et al. (1962) and Bham (2008) who both obtained results that fitted a Gamma distribution to their lag gap acceptance histogram. Recommendations The following is a brief summary of recommendations and findings based upon the results of this paper, to be cons idered as part of lane change model development Drivers have different lane changing characteristics. Attempting to model lane changes by incorporating different driver types in troduces more variability which can closer replicate real world conditions. Si milar conclusions were made for lane changing on urban arterials by Sun (2009). A Gamma distribution for accepted lag gaps is recommended based upon this research. This is consistent with Blunden et al. (1962) and Bham (2008). A Johnson Su distr ibution (si milar to a lognormal distribution) is recommended for modeling the distribution of lane change durations The 726 completed lane changes had a range from 2.3 seconds to 13.8 seconds and a mean of 5.48 seconds. These are similar results to Hetrick (1997) an d Toledo and Zohar (2007).

PAGE 110

110 Characteristics of lane changing in congested conditions differ from those in uncongested conditions. Specifically, the duration of lane changes is greater and the average size of accepted lag gaps is smaller in congested conditi ons. All lane changes do not have a uniform duration. It is recommended that lane change durations be differentiated by driver types and according to the level of congestion. At times, drivers change two lanes in one continuous lateral movement rather than one at a time. This behavior was mostly observed shortly after the on ramp merge and in very light traffic conditions. Lee et al. (2003) also observed this behavior. Drivers are not restricted to choosing between the immediate adjacent lanes as the target lane but have been observed to choose from all lanes of the freeway. This implies that drivers may temporarily accept less than favorable conditions in an immediate adjacent lane in order to eventually occupy their true choice of a target lane. Ben Akiva et al. (200 6 ) also recommends that explicit target lane choice be used in lane change models. F reeway drivers were frequently observed to change lanes or adjust their speed as a respo nse to on ramp merging vehicles well before arriving at the merging secti on. Some d rivers are uncomfortable near heavy vehicles and will change lanes and/or adjust their speed accordingly. They also may use the extra space in front of a heavy vehicle to perform a lane change if the opportunity presents itself. Freeway drivers w occupy when conditions among the available lanes did not govern. explaining their lane changing behavior. The above recommendations are based on the results and observations reported in this paper. It is suggested that these recommendations be brought into consideration when developing lane change models. Overall, l ane change models based upon external observat ions of vehicular movemen ts are lacking compared to those which consider the type of driver performing the maneuver Future Research The main shortcoming of this research was that a sufficient distribution of rejected gaps was not obtained from the data. Therefore, the desired obj ective of determining if different driver groups have significantly different average critical

PAGE 111

111 gaps could not be attained. In addition, it was also not possible to see if the average critical gaps for the driver groups follow any conclusive trend without a sufficient distribution of rejected gaps. If future research developed a systematic way for recording rejected gaps during freeway lane changing then these objectives could be explored and further support may be given for modeling lane changing based upon different driver types. Another area of future research is to perform a similar study as the one presented in this paper but with a focus on heavy vehicles. Perform an analysis on distributions of durations, accepted gaps, and rejected gaps for heavy vehi cles in both congested and uncongested conditions. More general, a study could be done to explore the relationship between lane changing and vehicle type. This would benefit the development of future lane changing models. The impact of different sites on l ane changing characteristics would be an interesting study. During this research, it was seen that the driver type impacts lane changing characteristics. However, the comparison of the Orlando and Jacksonville research locations suggest that the site of a study may be significant. The same driver type may not have identical lane changing characteristics across all freeway locations. Drawing conclusions on this relationship, based on quantitative data is a suggested area of future research.

PAGE 112

112 APPENDIX A RESEA RCH PARTICIPANT FORMS

PAGE 113

113

PAGE 114

114

PAGE 115

115 APPENDIX B MEASURING FREEWAY TRAFFIC STREAM GAPS USING VIDEO DATA This section outlines the process used in this research for measuring the lead and lag gaps of a freeway lane change. The fugitive point of an image is the poin t at which every parallel line intersects. Every camera has a fugitive point; a point where all the lines in the picture intersect. Parallel lines were drawn, using AutoCAD, for the front and rear camera of the instrumented vehicle to determine the fugitiv e points. Once the fugitive point was identified, a horizontal line was drawn through it. Figure B 1 shows the lines drawn to create the horizontal line that intersects the fugitive point. A B Figure B 1. Determining the Fugitive Point for the A) front and B) rear camera. This process only needs to be done once for each camera used; the camera constant and height of the camera from the ground stay the same. The next step in determining the lag and lead gap is to find the distance from the bottom of the tires of the vehicle in the target lane, to the horizontal line created from the fugitive point. This was done using the DIST command in AutoCAD. Once the distance from the horizontal line to where the tires meet the road is known the following equa tion from Psarianos et al (2001) can be used to determine the real distance from the camera to the vehicle in the target lane:

PAGE 116

116 (B 1) Where: is the real distance from the camera to the object is the camera constant is the camera height above ground level is the distance measured in the image, from the horizontal line intersecting the fugitive point to the object The values used for the camera constants for the front and rear camera was obtained fro m the previous study done by Kondyli and Elefteriadou (2009). These values are in the default units of AutoCAD. This required that be measured in the same units. The front and rear camera heights are 3.96 ft and 6.65 ft respectively. The front and rear camera constants are 7.214 and 5.783 respectively. With all the variables accounted for, can be determined. This value represents the lead and lag gaps between the experimental vehicle and the leading and lagging vehicles in the tar get lane.

PAGE 117

117 APPENDIX C LANE CHANGE DURATION AND GAP ACCEPTANCE DESCRIPTIVE STATISTICS Table C 1 Discretionary Lane Changes Duration Lag gap Lead gap Orlando Sample size 130 113 101 Mean 5.20 sec 84.32 ft 184.99 ft Maximum 8. 70 sec 216.86 ft 1393.83 ft Minimum 3.20 sec 23.85 ft 20.30 ft Standard deviation 0.96 41.01 230.73 Jacksonville Sample size 191 121 131 Mean 5.33 sec 96.25 ft 98.17 ft Maximum 7.80 sec 323.29 ft 550.55 ft Minimum 2.30 sec 11.02 ft 18.82 ft St andard deviation 1.03 50.59 77.62 Combined Sample size 321 234 232 Mean 5.28 sec 90.49 ft 135.97 ft Maximum 8.70 sec 323.29 ft 1393.83 ft Minimum 2.30 sec 11.02 ft 18.82 ft Standard deviation 1.00 46.50 168.22 Table C 2. M andatory Lane Changes Duration Lag gap Lead gap Orlando Sample size 81 44 47 Mean 5.66 sec 73.22 ft 157.83 ft Maximum 12.50 sec 151.76 ft 1360.64 ft Minimum 3.20 sec 10.53 ft 22.26 ft Standard deviation 1.75 43.82 222.97 Jacksonville Sample size 324 230 224 Mean 5. 64 sec 72.91 ft 81.45 ft Maximum 13.80 sec 285.40 ft 309.24 ft Minimum 2.70 sec 9.59 ft 18.03 ft Standard deviation 1.51 43.72 54.23 Combined Sample size 405 274 271 Mean 5.65 sec 72.96 ft 94.70 ft Maximum 13.80 sec 285.40 ft 1360.64 ft Minimu m 2.70 sec 9.59 ft 18.03 ft Standard deviation 1.56 43.65 108.34

PAGE 118

118 Table C 3. Mandatory Lane Changes (merging maneuvers excluded) Duration Lag gap Lead gap Orlando Sample size 57 33 38 Mean 5.29 sec 81.44 ft 183.02 ft Maximum 10.20 sec 151.76 ft 1360.64 ft Minimum 3.20 sec 18.64 ft 22.92 ft Standard deviation 1.22 43.99 241.39 Jacksonville Sample size 5.58 68.46 91.71 Mean 12.50 sec 199.33 ft 309.24 ft Maximum 2.90 sec 9.59 ft 22.26 ft Minimum 1.30 sec 40.28 ft 56.08 ft Standard devia tion 5.58 68.46 91.71 Combined Sample size 193 141 125 Mean 5.50 sec 71.50 ft 119.47 ft Maximum 12.50 sec 199.33 ft 1360.64 ft Minimum 2.90 sec 9.59 ft 22.26 ft Standard deviation 1.28 41.39 146.10 Table C 4. Mandatory Lane Changes (merging mane uvers only) Duration Lag gap Lead gap Orlando Sample size 17 5 4 Mean 5.78 sec 71.44 ft 72.12 ft Maximum 12.00 sec 100.76 ft 91.29 ft Minimum 4.20 sec 17.19 ft 48.02 ft Standard deviation 1.76 34.12 18.28 Jacksonville Sample size 181 127 131 Mean 5.78 sec 74.87 ft 72.81 ft Maximum 13.80 sec 285.40 ft 268.04 ft Minimum 2.70 sec 11.81 ft 18.03 ft Standard deviation 1.80 46.65 52.97 Combined Sample size 198 132 135 Mean 5.78 sec 74.74 ft 72.79 ft Maximum 13.80 sec 285.40 ft 268.04 ft Minimum 2.70 sec 11.81 ft 18.03 ft Standard deviation 1.79 46.15 52.25

PAGE 119

119 Table C 5. Uncongested Discretionary and Mandatory Lane Changes Duration Lag gap Lead gap Orlando Sample size 183 132 124 Mean 5.20 sec 85.96 ft 194.71 ft Maximum 10.20 sec 216.86 ft 1393.83 ft Minimum 3.20 sec 18.64 ft 22.92 ft Standard deviation 0.99 41.59 242.15 Jacksonville Sample size 456 300 300 Mean 5.30 sec 87.83 ft 96.60 ft Maximum 11.30 sec 323.29 ft 550.55 ft Minimum 2.30 sec 11.02 ft 18.82 ft Standard deviation 1.05 46.70 65.53 Combined Sample size 639 432 424 Mean 5.27 sec 87.26 ft 125.30 ft Maximum 11.30 sec 323.29 ft 1393.83 ft Minimum 2.30 sec 11.02 ft 18.82 ft Standard deviation 1.04 45.16 148.60 Table C 6. Uncongested Discretionary Lan e Changes Duration Lag gap Lead gap Orlando Sample size 111 96 82 Mean 5.12 sec 87.64 ft 206.11 ft Maximum 8.70 sec 216.86 ft 1393.83 ft Minimum 3.20 sec 23.85 ft 26.39 ft Standard deviation 0.89 41.40 247.95 Jacksonville Sample size 184 116 12 4 Mean 5.34 sec 96.83 ft 100.33 ft Maximum 7.80 sec 323.29 ft 550.55 ft Minimum 2.30 sec 11.02 ft 18.82 ft Standard deviation 1.02 51.22 78.94 Combined Sample size 295 212 206 Mean 5.25 sec 92.66 ft 142.43 ft Maximum 8.70 sec 323.29 ft 1393.83 ft Minimum 2.30 sec 11.02 ft 18.82 ft Standard deviation 0.97 47.14 175.28

PAGE 120

120 Table C 7. U ncongested Mandatory Lane Changes Duration Lag gap Lead gap Orlando Sample size 72 36 42 Mean 5.32 sec 81.49 ft 172.46 ft Maximum 10.20 sec 151.76 ft 1360. 64 ft Minimum 3.20 sec 18.64 ft 22.92 ft Standard deviation 1.14 42.36 231.72 Jacksonville Sample size 271 184 175 Mean 5.28 sec 82.16 ft 94.34 ft Maximum 11.30 sec 285.40 ft 309.24 ft Minimum 2.70 sec 18.37 ft 19.90 ft Standard deviation 1.08 42.80 54.18 Combined Sample size 343 220 217 Mean 5.29 sec 82.05 ft 109.46 ft Maximum 11.30 sec 285.40 ft 1360.64 ft Minimum 2.70 sec 18.37 ft 19.90 ft Standard deviation 1.09 42.63 116.25 Table C 8. Uncongested Mandatory Lane Changes (merging maneuvers only) Duration Lag gap Lead gap Orlando Sample size 15 3 4 Mean 5.41 sec 81.97 ft 72.12 ft Maximum 6.40 sec 100.76 ft 91.29 ft Minimum 4.20 sec 59.58 ft 48.02 ft Standard deviation 0.77 20.83 18.28 Jacksonville Sample size 136 85 90 Mean 5.11 sec 93.78 ft 91.43 ft Maximum 11.30 sec 285.40 ft 268.04 ft Minimum 2.70 sec 33.72 ft 19.90 ft Standard deviation 1.12 43.56 53.62 Combined Sample size 151 88 94 Mean 5.14 sec 93.37 ft 90.61 ft Maximum 11.30 sec 285.40 ft 268.04 ft Minimum 2.70 sec 33.72 ft 19.90 ft Standard deviation 1.10 42.98 52.70

PAGE 121

121 Table C 9. Uncongested Mandatory Lane Changes (merging maneuvers excluded) Duration Lag gap Lead gap Orlando Sample size 57 33 38 Mean 5.29 sec 81.44 ft 183.02 ft Maximum 10 .20 sec 151.76 ft 1360.64 ft Minimum 3.20 sec 18.64 ft 22.92 ft Standard deviation 1.22 43.99 241.39 Jacksonville Sample size 125 98 78 Mean 5.44 sec 72.48 ft 97.33 ft Maximum 7.90 sec 199.33 ft 309.24 ft Minimum 2.90 sec 18.37 ft 24.97 ft Sta ndard deviation 0.98 39.81 56.12 Combined Sample size 182.00 131.00 116.00 Mean 5.39 sec 74.73 ft 125.40 ft Maximum 10.20 sec 199.33 ft 1360.64 ft Minimum 2.90 sec 18.37 ft 22.92 ft Standard deviation 1.06 40.92 149.96 Table C 10. Congested Disc retionary and Mandatory Lane Changes Duration Lag gap Lead gap Orlando Sample size 28 25 24 Mean 6.82 sec 56.13 ft 81.57 ft Maximum 17.20 sec 120.07 ft 318.90 ft Minimum 3.80 sec 10.53 ft 20.30 ft Standard deviation 3.03 35.07 84.55 Jacksonvill e Sample size 60 51 56 Mean 7.22 sec 40.53 ft 38.50 ft Maximum 13.80 sec 134.61 ft 116.53 ft Minimum 3.40 sec 9.59 ft 18.03 ft Standard deviation 2.06 27.67 19.94 Combined Sample size 88 76 80 Mean 7.09 sec 45.66 ft 51.42 ft Maximum 17.20 sec 134.61 ft 318.90 ft Minimum 3.40 sec 9.59 ft 18.03 ft Standard deviation 2.40 30.96 52.46

PAGE 122

122 Table C 11. Congested Discretionary Lane Changes Duration Lag gap Lead gap Orlando Sample size 19 17 19 Mean 5.69 sec 65.59 ft 93.84 ft Maximum 8.10 sec 120.07 ft 318.90 ft Minimum 3.80 sec 23.87 ft 20.30 ft Standard deviation 1.24 34.01 91.17 Jacksonville Sample size 7 5 7 Mean 5.27 sec 82.94 ft 60.06 ft Maximum 7.00 sec 134.61 ft 116.53 ft Minimum 3.70 sec 50.65 ft 27.10 ft Standard deviati on 1.47 33.77 31.53 Combined Sample size 26 22 26 Mean 5.58 sec 69.53 ft 84.75 ft Maximum 8.10 sec 134.61 ft 318.90 ft Minimum 3.70 sec 23.87 ft 20.30 ft Standard deviation 1.29 33.97 80.36 Table C 12. Congested Mandatory Lane Changes Duration Lag gap Lead gap Orlando Sample size 9 8 5 Mean 9.20 sec 36.02 ft 34.95 ft Maximum 17.20 sec 94.09 ft 62.37 ft Minimum 4.80 sec 10.53 ft 22.26 ft Standard deviation 4.25 29.88 16.20 Jacksonville Sample size 53 46 49 Mean 7.47 sec 35.92 ft 35.4 2 ft Maximum 13.80 sec 115.22 ft 92.44 ft Minimum 3.40 sec 9.59 ft 18.03 ft Standard deviation 2.00 22.97 15.93 Combined Sample size 62 54 54 Mean 7.72 sec 35.93 ft 35.37 ft Maximum 17.20 sec 115.22 ft 92.44 ft Minimum 3.40 sec 9.59 ft 18.03 f t Standard deviation 2.48 23.79 15.80

PAGE 123

123 Table C 13. Congested Mandatory Lane Changes (merging maneuvers only) Duration Lag gap Lead gap Orlando Sample size 2 2 Mean 8.55 55.64 Maximum 12.00 sec 94.09 ft Minimum 5.10 sec 17.19 ft Stan dard deviation 4.88 sec 54.37 ft Jacksonville Sample size 45 42 41 Mean 7.86 sec 36.62 ft 30.71 ft Maximum 13.80 sec 115.22 ft 55.57 ft Minimum 4.20 sec 11.81 ft 18.03 ft Standard deviation 1.88 23.69 9.10 Combined Sample size 47 44 41 Mean 7 .89 sec 37.48 ft 30.71 ft Maximum 13.80 sec 115.22 ft 55.57 ft Minimum 4.20 sec 11.81 ft 18.03 ft Standard deviation 1.98 24.90 9.10 Table C 14. Congested Mandatory Lane Changes (merging maneuvers excluded) Duration Lag gap Lead gap Orlando Samp le size 7 6 5 Mean 8.39 sec 29.50 ft 34.90 ft Maximum 12.50 sec 61.32 ft 62.37 ft Minimum 4.80 sec 10.53 ft 22.26 ft Standard deviation 2.95 21.29 16.20 Jacksonville Sample size 4 4 4 Mean 5.20 sec 28.55 ft 53.02 ft Maximum 5.90 sec 37.46 ft 9 2.44 ft Minimum 4.60 sec 9.59 ft 31.48 ft Standard deviation 0.65 13.01 27.56 Combined Sample size 11 10 9 Mean 7.23 sec 29.11 ft 42.98 ft Maximum 12.50 sec 61.32 ft 92.44 ft Minimum 4.60 sec 9.59 ft 22.26 ft Standard deviation 2.81 17.56 22.5 1

PAGE 124

124 Table C 15. Left and Right Uncongested Discretionary Lane Change Durations Comparison Left Right Orlando Sample size 52 59 Mean 5.28 sec 4.98 sec Maximum 8 7 0 sec 6.60 sec Minimum 3.50 sec 3.20 sec Standard deviation 0.91 0.85 Jacksonvi lle Sample size 145 4 0 Mean 5.24 sec 5.63 sec Maximum 7.70 sec 7.80 sec Minimum 2.3 0 sec 2.80 Standard deviation 1.00 1.06 Combined Sample size 197 99 Mean 5.25 sec 5.24 sec Maximum 8.70 sec 7.80 sec Minimum 2.30 sec 2.80 sec Standard devi ation 0.97 0.99

PAGE 125

125 APPENDIX D HYPOTHESIS TESTING Uncongested DLC Durations vs. Congested DLC Durations Hypothesis test: Are the average discretionary lane change durations during uncongested conditions significantly different than, greater than, or less tha n the average discretionary lane change durations during congested conditions? 1 = mean uncongested DLC duration 2 = mean congested DLC duration Table D 1. Uncongested DLC Durations vs. Congested DLC D urations Hypothesis Testing Hypothesis p value Conclusion of Test Orlando H 0 : 1 = 2 H a : 1 2 0.0678 Fail to reject H 0 At a 5% l evel of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested DLC duration is different than the mean congested DLC duration. H 0 : 1 < 2 H a : 1 2 0.9661 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested DLC duration is greater than the mean congested DLC duration. H 0 : 1 > 2 H a : 1 2 0.0339* Reject H 0 At a 5% level of significance, it is concluded that there is s ufficient evidence to suggest that the mean uncongested DLC duration is less than the mean congested DLC duration. Jacksonville H 0 : 1 = 2 H a : 1 2 0.7377 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evide nce to suggest that the mean uncongested DLC duration is different than the mean congested DLC duration. H 0 : 1 < 2 H a : 1 2 0.6311 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that th e mean uncongested DLC duration is greater than the mean congested DLC duration. H 0 : 1 > 2 H a : 1 2 0.3689 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested DLC d uration is less than the mean congested DLC duration.

PAGE 126

126 Table D 1 Con tinued Hypothesis p value Conclusion of Test Combined H 0 : 1 = 2 H a : 1 2 0.1290 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient ev idence to suggest that the mean uncongested DLC duration is different than the mean congested DLC duration. H 0 : 1 < 2 H a : 1 2 0.9355 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested DLC duration is greater than the mean congested DLC duration. H 0 : 1 > 2 H a : 1 2 0.0645 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested DL C duration is less than the mean congested DLC duration.

PAGE 127

127 Uncongested MLC (merge only) vs. Congested MLC (merge only) Hypothesis test: Are the average uncongested mandatory lane change durations (only merge maneuvers) significantly different than, great er than, or less than the average congested mandatory lane change durations (only merge maneuvers)? 1 = mean uncongested MLC (merge only) duration 2 = mean congested MLC (merge only) duration Table D 2. MLC (merge only) Durations vs Congested MLC (merge only) Durations Hypothesis Testing Hypothesis p value Conclusion of Test Orlando H 0 : 1 = 2 H a : 1 2 0.5300 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (merge only) duration is different than the mean congested MLC (merge only) duration. H 0 : 1 < 2 H a : 1 2 0.7350 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that mean uncongested MLC (merge only) duration is greater than the mean congested MLC (merge only) duration. H 0 : 1 > 2 H a : 1 2 0.2650 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that mean uncongested MLC (merge only) duration is less than the mean congested MLC (merge only) duration. Jacksonville H 0 : 1 = 2 H a : 1 2 <.0001* Reject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that the mean uncongested MLC (merge only) duration is different than the mean congested MLC (merge only) duration. H 0 : 1 < 2 H a : 1 2 1.0000 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (merge only) duration is greater than the mean congested MLC (merge only) duration. H 0 : 1 > 2 H a : 1 2 <.0001* Reject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that the mean uncongested MLC (merge only) duration is less than the mean congested MLC (merge only) duration.

PAGE 128

128 Table D 2 Con tinued Hypothesis p value Conclusion of Test Combined H 0 : 1 = 2 H a : 1 2 <.0001* Reject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that the mean uncongested MLC (merge only) duration is different than the mea n congested MLC (merge only) duration. H 0 : 1 < 2 H a : 1 2 1.0000 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (merge only) duration is greater than the m ean congested MLC (merge only) duration. H 0 : 1 > 2 H a : 1 2 <.0001* Reject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that the mean uncongested MLC (merge only) duration is less than the mean congeste d MLC (merge only) duration.

PAGE 129

129 Uncongested MLC (no merge) vs. Congested MLC ( no merge) Hypothesis test: Are the average uncongested mandatory lane change durations (no merge maneuvers) significantly different than, greater than, or less than the average congested mandatory lane change durations (no merge maneuvers)? 1 = mean uncongested MLC (no merge) duration 2 = mean congested MLC (no merge) duration Table D 3 Uncongested MLC (no merge) Durations vs Congested MLC (no merge) Durations Hypothesis Testing Hypothesis p value Conclusion of Test Orlando H 0 : 1 = 2 H a : 1 2 0.0267* Reject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is different than the mean congested MLC (no merge) duration. H 0 : 1 < 2 H a : 1 2 0.9 866 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that mean uncongested MLC (no merge) duration is greater than the mean congested MLC (no merge) duration. H 0 : 1 > 2 H a : 1 2 0.0134* R eject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that mean uncongested MLC (no merge) duration is less than the mean congested MLC (no merge) duration. Jacksonville H 0 : 1 = 2 H a : 1 2 0.4496 Fail to re ject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is different than the mean congested MLC (no merge) duration. H 0 : 1 < 2 H a : 1 2 0.2248 Fail to re ject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is greater than the mean congested MLC (no merge) duration. H 0 : 1 > 2 H a : 1 2 0.7752 Fail to reje ct H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is less than the mean congested MLC (no merge) duration.

PAGE 130

130 Table D 3 Con tinued Hypothesis p value Con clusion of Test Combined H 0 : 1 = 2 H a : 1 2 0.0562 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is different than the mean congested MLC (no merge) duration. H 0 : 1 < 2 H a : 1 2 0.9719 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is greater than the mean congested MLC (no merge) duration. H 0 : 1 > 2 H a : 1 2 0.0281* Reject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is less than the mean congested MLC (no merge) duration.

PAGE 131

131 Uncongested MLC (no m erge) vs. Uncongested DLC Hypothesis test: Are the average uncongested MLC (excluding merge maneuvers) durations significantly different than, greater than, or less than the average uncongested DLC durations? 1 = mean uncongested MLC (no merge) duration 2 = mean uncongested DLC duration Tabl e D 4 Uncongested MLC (no merge) Durations vs. Uncongested DLC Durations Hypothesis Testing Hypothesis p value Conclusion of Test Orlando H 0 : 1 = 2 H a : 1 2 0.3572 Fail to reject H 0 At a 5% level of significance it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is different than the mean uncongested DLC duration. H 0 : 1 < 2 H a : 1 2 0.8214 Fail to reject H 0 At a 5% level of significance, it is c oncluded that there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is greater than the mean uncongested DLC duration. H 0 : 1 > 2 H a : 1 2 0.1786 Fail to reject H 0 At a 5% level of significance, it is concluded th at there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is less than the mean uncongested DLC duration. Jacksonville H 0 : 1 = 2 H a : 1 2 0.3338 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is different than the mean uncongested DLC duration. H 0 : 1 < 2 H a : 1 2 0.8331 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is greater than the mean uncongested DLC duration. H 0 : 1 > 2 H a : 1 2 0.1669 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not suffic ient evidence to suggest that the mean uncongested MLC (no merge) duration is less than the mean uncongested DLC duration.

PAGE 132

132 Table D 4 Con tinued Hypothesis p value Conclusion of Test Combined H 0 : 1 = 2 H a : 1 2 0.0562 Fail to reject H 0 At a 5% level o f significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is different than the mean congested MLC (no merge) duration. H 0 : 1 < 2 H a : 1 2 0.9719 Fail to reject H 0 At a 5% level o f significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is greater than the mean congested MLC (no merge) duration. H 0 : 1 > 2 H a : 1 2 0.0281* Reject H 0 At a 5% level of signifi cance, it is concluded that there is sufficient evidence to suggest that the mean uncongested MLC (no merge) duration is less than the mean congested MLC (no merge) duration.

PAGE 133

133 Congested MLC (no merge) vs. Congested DLC Hypothesis test: Are the average c ongested MLC (excluding merge maneuvers) durations significantly different than, greater than, or less than the average congested DLC durations? 1 = mean congested MLC (no merge) duration 2 = mean congested DLC duration Table D 5 Congested MLC (no merge) Durations vs. Congested DLC Durations Hypothesis Testing Hypothesis p value Conclusion of Test Orlando H 0 : 1 = 2 H a : 1 2 0.0396* Reject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that the mean congested MLC (no merge) duration is different than the mean congested DLC duration. H 0 : 1 < 2 H a : 1 2 0.0198* Reject H 0 At a 5% level of signi ficance, it is concluded that there is sufficient evidence to suggest that the mean congested MLC (no merge) duration is greater than the mean congested DLC duration. H 0 : 1 > 2 H a : 1 2 0.9802 Fail to reject H 0 At a 5% level of significance, it is conc luded that there is not sufficient evidence to suggest that the mean congested MLC (no merge) duration is less than the mean congested DLC duration. Jacksonville H 0 : 1 = 2 H a : 1 2 0.6354 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean congested MLC (no merge) duration is different than the mean congested DLC duration. H 0 : 1 < 2 H a : 1 2 0.6823 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean congested MLC (no merge) duration is greater than the mean congested DLC duration. H 0 : 1 > 2 H a : 1 2 0.3177 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean congested MLC (no merge) duration is less than the mean congested DLC duration.

PAGE 134

134 Table D 5 Con tinued Hypothesis p value Conclusion of Test Combined H 0 : 1 = 2 H a : 1 2 0.1007 Fail to reject H 0 At a 5% level of signif icance, it is concluded that there is not sufficient evidence to suggest that the mean congested MLC (no merge) duration is different than the mean congested DLC duration. H 0 : 1 < 2 H a : 1 2 0.0503 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean congested MLC (no merge) duration is greater than the mean congested DLC duration. H 0 : 1 > 2 H a : 1 2 0.9497 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean congested MLC (no merge) duration is less than the mean congested DLC duration.

PAGE 135

135 Right DLC vs. Left DLC Hypothesis test: Are the average right DLC durations significantly different than, greater than, or less than the average left DLC durations? 1 = mean DLC duration to the right 2 = mean DLC duration to the left Table D 6 Right DLC Durations vs Left DLC Durations Hypothesis Testing Hypothesis p value Conclusion of Test Orlando H 0 : 1 = 2 H a : 1 2 0.0687 Fail to reject H 0 At a 5% level of s ignificance, it is concluded that there is not sufficient evidence to suggest that the mean DLC duration to the right is different than the mean DLC duration to the left. H 0 : 1 < 2 H a : 1 2 0.9657 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that mean DLC duration to the right is greater than the mean DLC duration to the left. H 0 : 1 > 2 H a : 1 2 0.0343* Reject H 0 At a 5% level of significance, it is concluded that there is sufficie nt evidence to suggest that mean DLC duration to the right is less than the mean DLC duration to the left. Jacksonville H 0 : 1 = 2 H a : 1 2 0.0431* R eject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that the mean DLC duration to the right is different than the mean DLC duration to the left. H 0 : 1 < 2 H a : 1 2 0.0215* R eject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that the mean DLC duration to the right is greater than the mean DLC duration to the left. H 0 : 1 > 2 H a : 1 2 0.9785 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean DLC duration to the right is less than th e mean DLC duration to the left.

PAGE 136

136 Table D 6 Con tinued Hypothesis p value Conclusion of Test Combined H 0 : 1 = 2 H a : 1 2 0.9088 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean DLC duration to the right is different than the mean DLC duration to the left. H 0 : 1 < 2 H a : 1 2 0.5456 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean DLC durat ion to the right is greater than the mean DLC duration to the left. H 0 : 1 > 2 H a : 1 2 0.4544 Fail to r eject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean DLC duration to the right is l ess than the mean DLC duration to the left.

PAGE 137

137 Uncongested Lag Gap vs. Congested Lag Gap Hypothesis test: Are the average right DLC durations significantly different than, greater than, or less than the average left DLC durations? 1 = mean uncongested la g gap 2 = mean congested lag gap Table D 7 Uncongested Lag Gap vs. Congested Lag Gap Hypothesis Testing Hypothesis p value Conclusion of Test Combined H 0 : 1 = 2 H a : 1 2 <.0001* Reject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that the mean uncongested lag gap is different than the mean congested lag gap. H 0 : 1 < 2 H a : 1 2 <.0001* Reject H 0 At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that the mea n uncongested lag gap is greater than the mean congested lag gap. H 0 : 1 > 2 H a : 1 2 1.0000 Fail to reject H 0 At a 5% level of significance, it is concluded that there is not sufficient evidence to suggest that the mean uncongested lag gap is less than the mean congested lag gap.

PAGE 138

138 Summary of Hypothesis Testing Findings At a 5% level of significance, it is concluded that there is sufficient evidence to suggest that: the mean uncongested MLC (merge only) duration is less than the mean congested MLC (me rge only) duration. that the mean uncongested MLC (no merge) duration is less than the mean congested MLC (no merge) duration. the mean uncongested lag gap is greater than the mean congested lag gap. At a 5% level of significance, it is concluded that the re is not sufficient evidence to suggest that: the mean uncongested DLC duration is different than greater than or less than the mean congested DLC duration. the mean uncongested MLC (no merge) duration is different than greater than or less than the mean uncongested DLC duration. the mean congested MLC (no merge) duration is different than greater than or less than the mean congested DLC duration. the mean DLC to the left is different than greater than or less than the mean DLC to the right.

PAGE 139

139 APPE NDIX E LANE CHANGE DURATION AND GAP ACCEPTANCE STATISTICAL SUMMARIES Uncongested Discretionary Lane Change Duration Statistical Summary Figure E 1. Uncongested DLC Durations Histogram Table E 1. Uncongested DLC Durations Descriptive Statistics Moments Mean 5.25 Std Dev 0.97 Std Err Mean 0.06 Upper 95% Mean 5.37 Lower 95% Mean 5.14 N 295 Table E 2.Uncongested DLC Durations Fitted Normal Parameter Estimates Type Parameter Estimate Lower 95% Upper 95% Location 5.25 5.14 5.37 Dispersion 0.97 0.9 1.06 2log(Likelihood) = 820.228201336884 Table E 3. Uncongested DLC Durations Goodness of Fit Test Shapiro Wilk W Test W 0 .992 Prob
PAGE 140

140 C ongested Discretionary Lane Change Duration Statistical Summary Figure E 2. Congested DLC Durations Histogram Table E 4. Uncongested DLC Durations Descriptive Statistics Moments Mean 5.58 Std Dev 1.29 Std Err Mean 0.25 Up per 95% Mean 6.1 Lower 95% Mean 5.06 N 26 Table E 5.Congested DLC Durations Fitted Normal Parameter Estimates Type Parameter Estimate Lower 95% Upper 95% Location 5.58 5.06 6.1 Dispersion 1.29 1.01 1.77 2log(Likelihood) = 85.8277391106903 Table E 6. Congested DLC Durations Goodness of Fit Test Shapiro Wilk W Test W 0 .93 2 Prob
PAGE 141

141 All Discretionary Lane Change Duration s Statistical Summary Figure E 3. Uncongested and Congested DLC Durations Histogram Table E 7. Uncongested and Congested DLC Durations Descriptive Statistics Moments Mean 5.28 Std Dev 1 S td Err Mean 0.06 Upper 95% Mean 5.39 Lower 95% Mean 5.17 N 321 Table E 8. Uncongested and Congested DLC Durations Fitted Normal Parameters Type Parameter Estimate Lower 95% Upper 95% Location 5.28 5.17 5.39 Dispersion 1 0.93 1.09 2log(Likelihood) = 912.251578696371 Table E 9. Uncongested and Congested DLC Durations Goodness of Fit Test Shapiro Wilk W Test W 0 .99 4 Prob
PAGE 142

142 All Lane Change Duration s Statistical Summary Figure E 3. Uncongested and Congested DLC and MLC Durations Histogram Table E 10 Uncongested and Congested DLC and MLC Durations Descriptive Statistics Moments Mean 5.48 Std Dev 1.35 Std Err Mean 0.05 Upper 95% Mean 5.58 Lower 95% Mean 5.3 9 N 726 Table E 11. Uncongested and Congested DLC and MLC Durations Fitted Johnson Su Parameters Type Parameter Estimate Shape 0.71 Shape 1.47 Location 4.63 Scale 1.33 Table E 12. Uncongested and Congested DLC and MLC Goodness of Fit Tes t Shapiro Wilk W Test W 0 997 Prob
PAGE 143

143 Discretionary Lane Change Lag Gap Statistical Summary Figure E 4. Uncongested and Congested DLC Lag Gap Histogram Table E 13 Uncongested and Congested DLC Lag Gap Descriptive Statistics Moments Mean 90.4 9 Std Dev 46.50 Std Err Mean 3. 0 4 Upper 95% Mean 96.4 8 Lower 95% Mean 84.50 N 234 Table E 14 Uncongested and Congested DLC Lag Gap Fitted Gamma Parameters Type Parameter Estimate Shape 3.8 1 Scale 23.78 Threshold 0.00 Table E 15. Uncongested and Congested DLC Lag Gap Goodness of Fit Tes t Shapiro Wilk W Test W 0 063 Prob
PAGE 144

144 Dis cretionary Lane Change Lead Gap Statistical Summary Figure E 5 Uncongested and Congested DLC Lead Gap Histogram Table E 16 Uncongested and Congested DLC Lead Gap Descriptive Statistics Moments Mean 135.97 Std Dev 168.22 Std Err Mean 11.04 Upper 95 % Mean 157.73 Lower 95% Mean 114.2 1 N 232 Table E 17 Uncongested and Congested DLC Lead Gap Fitted Gamma Parameters Type Parameter Estimate Shape 3.87 Shape 0.92 Location 16.6 3 Scale 1 Table E 18. Uncongested and Congested DLC Lead Gap Goodness of Fit Tes t Shapiro Wilk W Test W 0 .99 3 Prob
PAGE 145

145 Lag Gap Statistical Summary Figure E 6 Uncongested and Congested DLC and MLC Lag Gap Histogram Table E 19 Uncongested and Congested DLC and MLC Lag Gap Descriptive Statistics Moments Mean 81.0 4 Std Dev 45.7 9 Std Err Mean 2.03 Upper 9 5% Mean 85.0 3 Lower 95% Mean 77.0 5 N 508 Table E 20 Uncongested and Congested DLC and MLC Lag Gap Fitted Gamma Parameters Type Parameter Estimate Shape 3.02 Scale 26.82 Threshold 0.00 Table E 21. Uncongested and Congested DLC and MLC Lag Gap Goodness of Fit Tes t Shapiro Wilk W Test W 0 .062 Prob
PAGE 146

146 Lead Gap Statistical Summary Figure E 7 Uncongested and Congested DLC and MLC Lead Gap Histogram Table E 22 Uncongested and Congested DLC and MLC Lead Gap Descriptive Statistics Moments Mean 113.73 Std Dev 140.5 7 Std Err Mean 6.2 7 Upper 95% Mean 126.0 5 Lower 95% Mean 101.41 N 503 Table E 2 3 Uncongested and Congested DLC and MLC Lead Gap Fitted Johnson SI Parameters Type Parameter Estimate Shape 3.7 9 Shape 0.94 Location 15.8 8 Scale 1 Table E 2 4 Uncongested and Congested DLC and MLC Lead Gap Goodness of Fit Tes t Shapiro Wilk W Test W 0 .99 4 Prob
PAGE 147

147 Uncongested Lag Gap Statistical Summary Figure E 8 Uncongested DLC and MLC Lag Gap Histogram Table E 25 Uncongested DLC and MLC Lag Gap Descriptive Statistics Moments Mean 87.26 Std Dev 45.16 Std Err Mean 2.17 Upper 95% Mea n 91.53 Lower 95% Mean 82.9 9 N 432 Table E 26 Uncongested DLC and MLC Lag Gap Fitted Gamma Parameters Type Parameter Estimate Shape 3.71 Scale 23.5 2 Threshold 0.00 Table E 27 Uncongested DLC and MLC Lag Gap Goodness of Fit Tes t Shapiro Wilk W Test W 0 2 46 Prob
PAGE 148

148 C ongeste d Lag Gap Statistical Summary Figure E 9 Congested DLC and MLC Lead Gap Histogram Table E 2 8 Congested DLC and MLC Lead Gap Descriptive Statistics Moments Mean 45.6 6 Std Dev 30.9 6 Std Err Mean 3.55 Upper 95% Mean 52.73 Lower 95% Mean 38.58 N 76 Table E 2 9 Congested DLC and MLC Lead Gap Fitted Johnson SI Parameters Type Parameter Estimate Shape 2.50 Scale 18.28 Threshold 0.00 Table E 30 Congested DLC and MLC Lead Gap Goodness of Fit Tes t Shapiro Wilk W Test W 0 215 Prob
PAGE 149

149 APPENDIX F DRIVER GROUPS DESCRIPTIVE STATISTICS Table F 1. G roup 1 Very Conservative Drivers Summary of Statistics Duration Lag gap Lead gap Orlando Sample size 42 31 27 Mean 6.2 sec 81.8 ft 93.5 ft Maximum 12.5 sec 163.1 ft 275.8 ft Minimum 3.4 sec 10.5 ft 20.3 ft Standard deviation 2.04 45.04 73.75 Jacksonville Sample size 28 22 21 Mean 6.66 sec 72.79 ft 52.74 ft Maximum 10.3 sec 185.67 ft 123.53 ft Minimum 4.4 sec 12.92 ft 24.48 ft Standard deviation 1.61 46.05 28.59 Combined Sample size 70 (17 congested) 53 48 Mean 6.39 sec 78.06 ft 75.68 ft Maximum 12.5 sec 185.67 ft 275.81 ft Minimum 3.4 sec 10.53 ft 20.30 ft Standard deviation 1.88 45.24 61.44 Tabl e F 2. Group 2 Somewhat Conservative Drivers Summary of Statistics Duration L ag gap Lead gap Orlando Sample size 69 54 53 Mean 5.42 sec 86.38 ft 141.11 ft Maximum 8.7 sec 171.21 ft 695.22 ft Minimum 3.2 sec 18.64 ft 25.93 ft Standard deviation 1.07 40.50 130.52 Jacksonville Sample size 254 171 177 Mean 5.65 sec 84.23 ft 89.93 ft Maximum 13.8 sec 285.4 ft 387.17 ft Minimum 2.3 sec 9.59 ft 18.03 ft Standard deviation 1.38 45.51 66.13 Combined Sample size 323 (38 congested) 225 230 Mean 5.59 sec 84.74 ft 101.72 ft Maximum 13.8 sec 285.4 ft 695.22 ft Minimum 2.3 sec 9.59 ft 18.03 ft Standard deviation 1.33 44.28 87.72

PAGE 150

150 Table F 3. Group 3 Somewhat Aggress ive Drivers Summary of Statistics Duration Lag gap Lead gap Orlando Sample size 73 51 46 Mean 5.21 sec 79.42 ft 216.8 ft Maximum 10.2 sec 181.90 ft 13 60.64 ft Minimum 3.60 sec 19.64 ft 26.39 ft Standard deviation 0.92 40.86 277.30 Jacksonville Sample size 100 72 62 Mean 5.13 sec 70.61 ft 96.44 ft Maximum 7.7 sec 168.59 ft 550.55 ft Minimum 2.7 sec 11.02 ft 21.45 ft Standard deviation 1.04 4 3.67 80.71 Combined Sample size 1 73 (8 congested) 123 108 Mean 5.16 sec 74.26 ft 147.70 ft Maximum 10.2 sec 181.90 ft 1360.64 ft Minimum 2.7 sec 11.02 ft 21.45 ft Standard deviation 0.99 42.58 199.07 Table F 4. Group 4 Very Aggressive Drivers Summary of Statistics Duration Lag gap Lead gap Orlando Sample size 27 21 22 Mean 4.58 sec 71.39 ft 278.45 ft Maximum 7.2 sec 21.86 ft 1393.83 ft Minimum 3.2 sec 27.58 ft 22.92 ft Standard deviation 0.81 44.68 351.28 Jacksonville Sample size 13 2 85 94 Mean 5.35 sec 85.38 ft 85.86 ft Maximum 11.6 sec 323.29 ft 262.38 ft Minimum 2.3 sec 11.81 ft 23.93 ft Standard deviation 1.3 53.82 51.21 Combined Sample size 159 (19 congested) 106 116 Mean 5.22 sec 82.61 ft 122.39 ft Maximum 11.6 sec 323.29 ft 1393.38 ft Minimum 2.3 sec 11.81 ft 22.92 ft Standard deviation 1.27 52.24 174.36

PAGE 151

151 APPENDIX G MEAN LANE CHANGE DURATION COMPARISONS OF DRIVER GROUPS Figure G 1. Multiple Means Comparison of Lane Change Durations for Driver Groups

PAGE 152

152 Means Comparisons : Comparisons for all pairs using Tukey Kramer HSD q* Alpha 2.57508 0.05 Table G 1. Difference of Means for all Pairs of Driver Groups Dif=Mean[i] Mean[j] Group 1 Group 2 Group 4 Group 3 Group 1 0.0000 0.7956 1.1681 1.2233 Group 2 0 .7956 0.0000 0.3725 0.4277 Group 4 1.1681 0.3725 0.0000 0.0552 Group 3 1.2233 0.4277 0.0552 0.0000 Table G 2. Least Significant Difference Threshold Matrix Abs(Dif) LSD Group 1 Group 2 Group 4 Group 3 Group 1 0.56922 0.351641 0.685059 0.746253 Group 2 0.351641 0.26499 0.04624 0.110391 Group 4 0.685059 0.04624 0.37769 0.31479 Group 3 0.746253 0.110391 0.31479 0.36208 Positive values show pairs of means that are significantly different. Table G 3. Connecting Letters Report Level Mean Group 1 A 6.3857143 Group 2 B 5.5900929 Group 4 C 5.2176101 Group 3 C 5.1624277 Levels not connected by same letter are significantly different. Table G 4. Ordered Differences Report Level Level Difference Std Err Dif Lower CL Upper CL p Value Group 1 Group 3 1.223287 0.1852496 0.746253 1.700320 <.0001* Group 1 Group 4 1.168104 0.1875843 0.685059 1.651150 <.0001* Group 1 Group 2 0.795621 0.1724139 0.351641 1.239602 <.0001* Group 2 Group 3 0.427665 0.1232091 0.110391 0.744 939 0.0031* Group 2 Group 4 0.372483 0.1266922 0.046240 0.698726 0.0178* Group 4 Group 3 0.055182 0.1436725 0.314787 0.425151 0.9807

PAGE 153

153 APPENDIX H MEAN LAG GAP ACCEPTED IN CONGESTED CONDITIONS COMPARISONS OF DRIVER GROUPS Figure H 1. Multiple Means Co mparison of Lag Gaps Accepted in Congested Conditions for Driver Groups

PAGE 154

154 Means Comparisons : Comparisons for all pairs using Tukey Kramer HSD q* Alpha 2.63184 0.05 Table H 1. Difference of Means for all Pairs of Driver Gro ups for Lag Gaps Dif=Mean[i] Mean[j] Group 1 Group 2 Group 4 Group 3 Group 3 0.000 4.035 6.613 12.710 Group 4 4.035 0.000 2.578 8.675 Group 2 6.613 2.578 0.000 6.097 Group 1 12.710 8.675 6.097 0.000 Table H 2. Least Significant Difference Thres hold Matrix for Lag Gaps Abs(Dif) LSD Group 1 Group 2 Group 4 Group 3 Group 1 44.674 34.2216 27.835 25.5466 Group 2 34.2216 30.5181 23.0044 21.8432 Group 4 27.835 23.0044 19.4313 19.4857 Group 3 25.5466 21.8432 19.4857 30.5181 Positive values show pairs of means that are significantly different. Table H 3. Connecting Letters Report for Lag Gaps Level Mean Group 3 A 52.667198 Group 4 A 48.632258 Group 2 A 46.054123 Group 1 A 39.957316 Levels not connected by same letter are significantly different. Table H 4. Ordered Differences Report for Lag Gaps Level Level Difference Std Err Dif Lower CL Upper CL p Value Group 3 Group 1 12.70988 14.53605 25.5466 50.96639 0.8181 Group 4 Group 1 8.67494 11.59576 21.8432 39.19308 0.8772 Group 3 Group 2 6.61307 13.08899 27.8350 41.06116 0.9576 Group 2 Group 1 6.09681 9.72042 19.4857 31.67936 0.9230 Group 3 Group 4 4.03494 14.53605 34.2216 42.29145 0.9925 Group 4 Group 2 2.57813 9.72042 23.0044 28.16069 0.9934

PAGE 155

155 LIST OF REFE RENCES Ahmed K.I., Ben Akiva M., Koutsopoulos H.N. and Mishalani R.G. (1996), Models of freeway lane changing and gap acceptance behavior. Proceedings of the 13 th International Symposium on the Theory of Traffic Flow and Transportation, pp. 501 515. AIMSUN User Manual, Version 4.1. (2002), TSS Transport Simulation Systems Barcelona, Spain. Ben Akiva M., Choudhury C., and Toledo T. (2006) Lane changing models Proceedings of the International Symposium of Transport Simulation Lausanne, Switzerland. Bham G .H. (2008), Analyzing microscopic behavior: driver mandatory lane change behavior on a multilane f reeway National University Transportation Center Missouri University o f Science and Technology. Rolla, Missouri. Bl unden W.R., Clissold C.M., and Fisher R.B (1962), Distribution of the acceptance gaps for crossing and turning m aneuvers. Proceedings of the Australian Road Research Board 1 1, pp. 188 205. Brackstone M., McDonald M., and Wu J. (1998), Lane c hangi ng on the motorway: factors affecting its o ccurre nce, and their i mplications. Proceedings of 9th International Conference on Road Transportation Infor mation and Control 454, pp. 160 164. Coifman B., Krishnamurthy S. and Wang X. (2005), Lane changing maneuvers consuming freeway c apacity. Traffic and Gran pp. 3 14. CORSIM Reference Manual, Version 5.0. (2001), ITT Industries, Inc., Systems Division ATMS R&D and Systems Engineering Program Team. Colorado Springs, Colorado. Daganzo C. (1981), Estimation of gap acceptance parameters within and a cross the population from direct roadside o bservation. Transportation Research B 15 pp. 1 15. Finnegan P. and Green P (1990), The time to change lanes: a literature r eview University of Michigan Transportation Research Institute, Ann Arbor Michigan. Gi pps P. G. (1986), A model for the structure of l ane c hanging d ecisions. Transportation Research B 20 5, pp. 403 414. Hanowski R. J. (2000), The impact of l ocal/ short haul operations on driver f atigue. Ph.D. dissertation. Industrial and Systems Engineering, V irginia Polytechnic Institute and State University, Bl acksburg, Virginia.

PAGE 156

156 Herman R. and Rothery W.R (1961), C ar following and steady state f low. International Symposium on the Theory of Road Traffic Flow Elsevier, New York. Hetrick S. (1997), Examination of driver lane change behavior and the potential e ffectiven s ystems MS thesis. Industrial and Systems Engineering, Virginia Polytechnic Institute and State University, Blacksburg, Virgin ia. Hidas P. (2002), Modeling l ane c hanging and m erging in m icroscopic t raffic s imulation. Transportation Research C 10 5 6, pp. 351 371. Hidas P. (2005), Modeling vehicle i nteractions in microscopic traffic s imulation of merging and w eaving. Transportatio n Research C 13 1, pp. 37 62. Hidas P. and Behbahanizadeh K. (1998), SITRAS: a s imulation m odel for ITS a pplications. Proceedings of 5th World Congress Intelligent Transportation System Seoul, Korea. JMP, Version 7. (2007), SAS Institute Inc. Cary, N orth C arolina. Kesting A., Treiber M., and Helbing D (2007), General lane changing m odel MOBIL for car following m odels. Transportation Research Record 1999 pp. 86 94. Knoop V.L., Buisson C. Wilson E and Van Arem B. (2011), Number of lane changes determined by splashover effects in loop d etector c ounts. Transportation Research Board 90th Annual Meeting pp. 23 27. Kondyli A. an d Elefteriadou L. (2009), Driver behavior at freeway ramp merging a reas: focus group f indings. Transportation Research Record 2124, pp. 157 166. Kou C. and Machemehl R. (1997), Modeling driver behavior during merge m aneuvers. Southwest Region University Transportat ion Center, Austin, Texas. Laval J.A. and Daganzo C.F. (2006), Lane changing in traffic s treams. Transportation Research B 40 3, pp. 251 264. Lee S.E., Olsen E.C.B., and Wierwille W. (2003), A comprehensive e x amination of naturalistic lane c hanges. Report DTNH 22 00 C07007, Task Order 4. NHTSA, U.S. Department of Transportation. Lee G (2006), Modeling gap a c ceptance in freewa y m erges. MS thesis. Department of Civil and Environmental Engineering, Massac husetts Institute of Technology, Cambridge, Massachusetts. Miller A. J. (1972), Nine estimators of gap a cceptance p arameters. Traffic F low and Transportation. pp. 215 236.

PAGE 157

157 Moridp our S., Rose G., and Sarvi M (2010), Effect of s urrounding t raffic c haracteristics on l ane c hanging b ehavior. Journal of Transportation Engineering 136 11, p p. 973. Olsen E.C.B., Lee S.E. and Wierville W.W. (2002), Analysis of distribution, frequency, an d d uration of n aturalistic l ane c hanges, Human Factors 46 pp. 1789 1793 Psarianos B. Paradissis D. Nakos B. and Karras G.E. (2001), A cost effective road surveying m ethod for the assessment of road a lignments. Proceedings of the IV International Sympos ium Turkish German Joint Geodetic Days 1, pp. 235 244. Pursula M. (1999), Simulatio n of traffic systems an o verview. Journal of Geographic Information and Decision Analysis 3 1 pp. 1 8. Raff M.S. and Hart J.W. (1950), A volume warrant for urban stop s igns Eno Foundation for Highway Traffic Control, Saugatuck, Connecticut. Roess R.P. and Ulerio J.M. (1997), NCHRP r eport 385: c omparison of the 1994 h ighway c apacity m r amp a nalysis p rocedures and the FRESIM m odel. TRB, National Research Council, Washi ngton, D.C. Salvucci D.D. and Liu A (2002), The time course of a l ane change: d river control and eye movement b ehavior. Transportation Research F 5 pp. 123 132. Smith S. (1985), Freeway data collection for studying vehicle i nteractions. Technical Report FHWA/RD 85/108. Sun D. J. (2009), A lane c hanging model for u rb an arterial s treets. PhD thesis. Univers ity of Florida, Gainesville, Florida. Sun D. J. and Elefteriadou L. (2010), Research and i mplementation of l ane changing model based on driver b ehavior. Transportation Research Record 2161, pp. 1 10. Tibshirani R., Walther G., and Hastie T. (2001), Estimating the n umber of c lusters in a d ataset via the g ap s tatistic. Journal of the Royal Statistics Society 63 2 pp. 411 424. Tijerina L., Garrott W.R., Glec ker M., Stoltzfus D., and Parmer E. (1997), Van and p assenger c ar d river e ye g lance b ehavior d uring l ane c hange d ecision p hase, i nterim r eport. Transportation Research Center Report. NHTSA, U.S. Dep artment of Transportation. Toledo T., Choudhury C.F., and Ben Akiva M.E. (2005), Lane c hanging m odel with e xplicit t arget l ane c hoice. Transportation Research Record 1934 pp. 157 165. Toledo T., Koutsopoulos H.N., and Ben Akiva M.E. (2003), Modeling i ntegrated l ane c hanging b ehavior. Transportation Research Reco rd 1857, pp. 30 38.

PAGE 158

158 Toledo T. and Zohar D. (2007), Modeling d uration of l ane c hanges. Transportation Research Record 1999, pp. 71 78. Worrall R.D. and Bullen A.G.R. (1970), An e mpirical analysis of lane changing on multilane h ighways. High way Research Record 303, pp. 30 43.

PAGE 159

159 BIOGRAPHICAL SKETCH Corey Hill, a native Floridian, was accepted into the University of Florida as an undergraduate student in 2006. During this time he was exposed to fundamental transportation engineering courses and was quickly intrigued. He was selected for the Transportation Research Internship Program in 2010 and was impressed by the higher education provided by the Transportation Engineering Graduate School Program. He joined this graduate school program after gr aduating summa cum laude from the University of Florida with a B .S. degree in Civil Engineering in May of 2011. He also became a licensed Engineering Intern in the state of Florida shortly after receiving his B S degree. As a graduate student he worked as a research assistant under Dr. Lily Elefteriadou. His main assistance involved conducting research on the variable speed limit system along Interstate 4 in Orlando, FL via an instrumented vehicle driven by research participants. The data obtained from thi s research led to the formulation of his m thesis could be written on a seemingly simple, everyday freeway maneuver. However, he soon realized the dynamics involved in a maneu ver and the complexities of attempting to develop models to replicate lane changes in microscopic traffic simulation. He is also passionate about his faith in Jesus Christ and involvement with a local church in Gainesville, FL. Other interests include atte nding Florida Gator sport ing events and playing sports like basketball and golf.