PEAK PERTOD MODEL DEVELOPMENT BASED ON EXTSTING MICHIGAN ORIGTN - DESTINATTON SURVEY INFORMATTON Thesis for the Degree of M. U. P. MICHTGAN STATE UNIVERSITY LOUTS HUGHES LAMBERT 197.5 ‘____ L I B R A R Y Michigan State University - Iflfib‘lh .7 sinuous BY" “DAB 8 SONY ‘ 800K BINDERY INC. LIBRAF'Y' B! 'd DERS . _" five ABSTRACT PEAK PERIOD MODE L DEVE DOPMENT BASED ON EXISTING MH'HIGAN ORIGIN-DESTINATION SURVEY INTORMATION By Louis Hughes Lambert This study develops a process by which peak period travel can be mod— eled using data and programs available through the Michigan Department of State Highways and T ranSportation. Two processes are developed. One pro- cess involves the factoring of 24-hour travel matrix by purpose, percentages and directional travel splits. The second utilizes the typical planning model- ing process to develop calibrated gravity models and regression equations. The process assumes a calibrated network. This study concentrated on only the work purpose to demonstrate the process. Person trips as opposed to vehicle trips are utilized in the modeling effort. The process is carried out at an aggregated zonal level. A case study is conducted using the data from the Flint, Michigan ori- gin and destination study of 1966. Both AM and PM travel patterns for the work purpose are examined. Only a one hour period was used for ease of test- ing in each peak period. Equations are developed for the AM and PM period and tested using calibrated gravity models from the 24-hour study. The matrix Louis Hughes Lambert model is developed from the 24-hour travel matrix. After the testing and calibration phases are completed, a total model is tested to determine the overall reliability of the developed models. The models were developed based on a need to utilize existing data and programs, so that speed of delivery and process understandability would be maximized. The conclusion of the case study is that viable peak period processes can be developed using existing 24~hour calibrated networks, survey data bases, and with a minimal amount of gravity model recalibration. The standard in— stitutional (synthetic) model did a better job of predicting than. the matrix factor model. The matrix model is recommended primarily for use in any sketch planning activity. The work matrix model tended to underpredict more extremely at the higher volumes than the synthetic model. It is recommended that these procedures be utilized for all planning problems that center on peak period concepts . PEAK PERIOD MODEL DEVELOPMENT BASED ON EXISTING MICHIGAN ORIGIN-DESTINATION SURVEY INFORMATION By Louis Hughes Lambert A THESIS Submitted to Michigan State University In partial fulfillment of the requirements for the degree of MASTER OF URBAN PLANNING School of Urban Planning 1975 ACKNOWLEDGMENTS Appreciation is gratefully extended to my many associates in the Bureau of TranSportation Planning of the Michigan Department of State High- ways and TranSportation for the outstanding cooperation and assistance they have provided to me in the completion of this study. Special thanks go to Michael Eberlein, Robert Hull, and William Hartwig for sitting on my thesis committee and to Larry Britton, my supervisor, for the tolerance and encour- agement offered to me in this study. To Rachele I must add my graditute for assistance in typing and for basic support and understanding. Finally, I must thank my thesis advisor, Professor Donn L. Anderson, for lending his valuable efforts in helping me to complete this endeavor. Typist for the thesis was Margaret A. Stosik with editorial assistance shouldered by Arthur Trelstad. Louis Hughes Lambert ii TABLE OF CONTENTS INTRODUCTION 1. Background 2. Content and Purpose of Research PART I CHAPTER I WHAT IS THE PEAK HOUR ? 1. Defining the Peak Period 2. Peak Period Simulation Model Research A. Operational versus Research Models B. Purposed Peak Period Models CHAPTER II RESEARCH METHODOLOGY 1. Systems Approach 2. Peak Period Planning Process 3. Model Development A. ' Trip Generation (1) Regression and Rate Models (2) Factor Model Generation (3) Combination Models B. Trip Distribution C. Tests for Model Acceptability (1) Numerical (2) Graphical D. Process Development (1) Data Preparation (2) Gravity Model Testing and Calibration iii PAGE {Dr—l 15 32 32 36 46 49 5O 53 54 55 55 56 56 56 57 59 (3) Trip Generation Analysis (4) Matrix Factoring PART II CHAPTER III FLINT, MICHIGAN: A CASE STUDY Introduction . Flint-Genesee County Study Area AM Peak School Trips in Peak Period PM Peak ens-um;— CHAPTER IV FLINT-GENESEE COUNTY TRIP GENERA- TION AND FACTOR MODEL DEVELOPMENT 1. Introduction Developing equations A. Homebased Work Production and Attraction Equations PM Period (1) Homebased Work Production PM (2) Homebased Work Attraction PM B. Homebased Work Production and Attraction Equations AM Period (1) Homebased Work Production AM (2) Homebased Work Attraction AM 3. Factor Models A. Model Development B. PM Work Matrix C. AM Work Matrix CHAPTER V DISTRIBUTION OF PEAK PERIOD TRIPS COMPARED TO 24-HOUR TRIPS Introduction Actual Skim Trees Gravity Model Peak Period Distribution hum:— A. Work Comparison B. Non-Work Comparison iv PAGE 59 61 67 68 70 75 77 82 85 91 92 98 103 106 107 110 115 119 120 120 122 126 132 5. (1) Shopping (2) Social-Recreational (3) Other (4) Non-Homebas ed (NHB) (5) Truck (6) Cordon Conclusion CHAPTER VI EVALUATION OF TOTAL MODELS Auto;— Introduction Matrix Comparison Tools Evaluation of Assignment Total Model Applicability CHAPTER "VII CONCLUSION 1.. 3. 4. 5. Speed of Delivery Availability of Data and Programming Capabilities Trip Generation Trip Distribution Matrix Model versus Synthetic Model LIST OF REFERENCES APPENDICES A TRANSPORTATION PLANNING PROGRAM PACKAGE, UTILITY PROGRAMS AND B. A. S. I. So DATA FILE SUMMARIES NUMERICAL AND GRAPHICAL STATISTICAL STANDARDS FOR TRANSPORTATION MODEL BUILDING 24-HOUR TRIP GENERATION EQUATION FOR HOMEBASED WORK PRODUCTIONS AND ATTRACTIONS PAGE 132 133 133 134 135 135 136 138 139 140 150 153 153 153 154 155 161 165 167 172 TABLE 10 ll 12 13 LIST OF TABLES Vehicular Travel From O-D Surveys Summarized into Standardized Hour Periods Distribution of 24-Hour Travel Crossing the Screenline of Selected Michigan Study Areas Percentage of ADT in Peak Hour for One Direction and Both Directions by Peak Hour, 30th High Hour and 200th Hour, by Facility Type Heavy Recreational Traffic Characteristics on Michigan State T runklines Distribution of Work Trips Crossing the Screenline as a Percentage of Each of the Hour Period's Total Trips for Selected Michigan Study Areas Percent Travel During Peak Periods by Industry Groups— Flint O-D Data, 1966 Person Trips Percent Travel During Peak Periods by Income Groups— Flint O-D Data, 1966 Person Trips Basic Data: Flint-Genesee County, Michigan, 1966 AM Peak Person Trips by Purpose (Internal Trips) Flint 24-Hour Person Trips by Purpose by Half-Hour Increments by the Percentage that Each is of Each Time Period Percentages by School-Aged Categories Flint-Genesee County PM Peak Person Trips by Purpose (Internal Trips) Correlation Matrix vi PAGE 16 21 25 26 28 31 41 68 70 71 74 76 83 TABLE 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 Regression Summary - Homebased Work Productions PM (Population) Regression Summary — Homebased Work Productions PM (HBWP24) Base Year Employment by Zone (1966) Regression Summary - Homebased Work Attractions PM (Total Employment) Regression Summary - Homebased Work Attractions PM (HBWA24) Regression Summary - Homebased Work Productions AM (Autos Available) Regression Summary - Homebased Work Productions AM (HBWP24) Regression Summary - Homebased Work Attractions AM (Non-Manuf and Manuf) Percent by Purpose in AM and PM Peak Period for Flint-Genesee County Directional Percentages AM and PM Period for Flint—Genesee County Matrix Comparison of PM Work Actual Trip Inter— changes and PM Factored Matrix Model with Zero Values Removed for Analysis Trip End Summary Comparison of Work Matrix Model PM Period Matrix Comparison of AM Actual Trip Interchanges and AM F actored Model with Zero Values Removed for Analysis Trip End Summary Comparison of Work Matrix Model AM Period Comparison of 24-Hour, AM, PM, AM Gravity Model, and PM Gravity Model Trip Length Distributions vii PAGE 87 89 93 94 96 99 101 104 109 1’0 9 111 113 115 116 124 TABLE 29 30 31 32 33 34 35 Comparison of 24-hour, AM and PM Travel Percentages Assigned Link Comparisons by Volume Groups Link Class Comparison PM Peak for the Work Purpose VMT and VHT by Peak O-D, Factored Matrix, and Synthetic Assignment Jurisdiction Comparison PM Peak for the Work Purpose VMT and VHT by Peak O-D, Factored Matrix, and Synthetic Assignment Screenline Percentage Comparison Regression Summary: Homebased Work Productions (Resident Labor Force) Regression Summary: Homebased Work Attractions (Total Employment) viii PAGE 144 147 148 149 173 168 FIGURE 10 ll 12 13 14 15 16 17 LIST OF FIGURES Continuing Urban Transportation Planning Process Effect of Travel by Time of Day V/C Effect of Travel by Time of Day Freeway Speed Effect of Travel by Time of Day Travel Cost Directional Split of Work Trips in Flint, Michigan Classification of Modal Choice Models Urban Travel Forecasting Process Peak Period Development Process Peak Period Data Preparation Process Peak Period Gravity Model Calibration Process Peak Period Direct Demand Generation Calibration Process Peak Period Post Distribution Interchange Analysis Flint-Genesee County Study Area Map Residual Plot: Homebased Work Productions PM (Population) Residual Plot: Homebased Work Productions PM (HBWP24) Residual Plot: Homebased Work Attractions PM (Total Employment) Residual Plot: Homebased Work Attractions PM (HBWA24) PAGE 20 20 20 29 35 48 52 58 60 62 63 66 88 90 95 97 FIGURE 18 19 20 21 22 23 24 25 26 27 28 29 30 Residual Plot: Homebased Work Productions AM (Autos Available) Residual Plot: Homebased Work Productions AM (HBWP24) Residual Plot: Homebased Work Attractions AM (Manufacturing and Non-Manufacturing Employment) Residual Plot: Matrix Work PM Total Trips Residual Plot: Matrix Work AM Total Trips Flint F—Factor Curves 24-Hour O-D Work Distribution versus AM Peak Period Work 24-Hour O-D Work Distribution versus PM Peak Period Work Distribution. Flint Study Area 1966 AM Peak Period Work Distribution versus Gravity Model AM Distribution. Flint Study Area 19 66 PM Peak Period Work Distribution versus Gravity Model PM Distribution. Flint Study Area 1966 Residual Plot: Total Link Volumes for 0-D PM versus Matrix Model Residual Plot: Total Link Volumes for 0-D PM versus Synthetic Model Examples of Residual Plots PAGE 100 102 105 114 117 127 128 129 130 131 141 142 171 INTRODUCTION 1. Background Almost 70 percent of the population of the U.S. and the majority of our industrial capacity is centered in our urban metrOpolitan areas.1 Rapid growth and change in our urbanized areas is causing dramatic effects on tranSporta- tion needs of the country. As planners, our greatest challenge is to create ur— ban environments that satisfy basic human needs and that are economically healthy. A sound tranSportation delivery system is basic to the satisfaction one can achieve from his environment due to the increased mobility it provides. Much of the dissonance that is present in our environment can be traced to this very same tranSportation system as it generates pollution and driver frustra- tion. The need to deveIOp a workable transportation system is the challenge of the tranSportation planning process. The passage of the Federal—Aid Highway Act of 1944 first provided regular Federal-Aid funds for use in urban areas, and the Federal Highway Administration has actively promoted urban tranSportation planning since that time. This was followed by Section 9 of the Federal-Aid Highway Act of 1United States Department of Transportation, Highway Needs fiport of 1972, Part II (Washington, D.C., 1972), p. II-64. 2 1962, which amended Chapter I of Title 23, U. S. Code, by Adding Section 134. 2 This section requires that programs for Federal-aid highway projects approved after July 1, 1965 in urban areas of more than 50, 000 pOpulation must be based on a "continuing, coordinated, comprehensive (30)" tranSportation planning process that will be carried on cooperatively by state and local communities. The Highway Act of 19703 changed the priorities assigned from the longstand- ing overemphasis in highways to the study of modal types. This act required that all types of movement systems be evaluated. Urban tranSportation planning, as practiced since 1944, has been de- signed to develop and continuously evaluate long—range highway system plans. Only recently, have multi-modal approaches been developed that are reflec- tive of the short and long-range goals and objectives of urban communities. This development is a response to a national reevaluation of goals which pro- duced a shift from goals emphasizing highway development (movement of cars and and trucks), to goals emphasizing the movement of people by the most en- vironmentally efficient methods. The urban decision maker faces an extensive range of pressing prob- lems. The questions of short-range adjustment in the system necessary to meet and maintain air quality standards and conserve energy or to provide access to the system for those with mobility handicaps (i. e. , young, poor, elderly, etc.) are replacing much of the past interest in projections of 24-hour tmffic volumes at 20-year intervals. The old concerns for more and more highway development are being reexamined in light of the need to expand and 2Federal-Aid Highway Act of 1962, Section 9 (1962). 3Federal-Aid Highway Act of 1970, 82 Stat. 836 (1970). eventually improve less personal modes of operation, and to minimize their operating difficulties. While these readjustments of priorities are happening, it is important to remember that highway development has its rightful place in the new panthe- on of priority consideration. America is still the home of the auto and because of our Sprawling urban—scape must, in the foreseeable future, remain that way. The auto will be the target of much critical comment and new attention will be rightfully directed to its environmental and social costs to the commu- nity. But, as long as the Federal Highway Trust Fund is not completely "busted" and the American consuming public demands his individual mobility in the form of the auto, the planner cannot become a one dimensional advocate of any tranSportation mode but must strive to adequately present the facts con- cerning the costs associated with any set of land use and tranSportation solu- tions. The tranSportation planning process is tied intimately to the develop- ment of simulation modeling. Simulation models are developed as a result of the collection of data from individual metropolitan areas. Using this in- formation, the three basic elements of the travel milleu are modeled: Trip characteristics, trip maker characteristics, and tranSportation system char- acteristics. Four models are normally developed: (1) The tranSportation network, (2) trip distribution, (3) trip generation, and (4) traffic assignment. Before discussion of these models, a brief consideration of the assumptions behind the process is presented. 4 A major assumption is that human travel behavior has order and this order can be described. If human behavior has order, then it is also possible to establish and forecast relationships between a person and his acti- vities. The process is basically a four part activity involving the inventory of existing socio-economic characteristics of a trip maker and his basic trav- el patterns, the development of relationships that explain observed conditions and activities, the use of these developed models to forecast what might be the situation in the future (establish future problems), and the use of this prob- lem identification to implement short and long—range policies and activities to alleviate these problems. The use of the computer has facilitated the process because of its ability to quickly assimilate large amounts of data. It has also caused, to a great extent, the over—concentration of effort in the areas of data collection and model building, to the detriment of future problem identification and im- plementation of corrective measures. This is due, in large measure, to the "ivory tower" kinds of models proposed which have often required the prac- ticing professional to be a researcher more than a "doer". The direction of this thesis will be to emphasize strongly the utilization of existing procedures and computer program technology with real emphasis given to Speed of pro- duct delivery. An overview of the elements of the tranSportation planning process, as practiced in the United States, is sketched in Figure 1.4 (FOP a more 4United States Department of Transportation, "3lst. Planning Course", Office of Highway Planning, Federal Highway Administration (Washington, D.C., 1972), p. 1. mmOOORA mcficcmHm :oprPhogmcmse MEASCHpcoo one H oseu-s ItIIIIIIIIIIIIIIIIIIJ a? 72.: 20:2 _ _ “2.822. sodas 5.25.5.2. as” , - “ £522 525:: p E: 3:23.. :2: Egg] _ $353 $2”: 252;: _ gang: 22.2% e: s: _ 2%: age as: _ was: sass A is: :2: _ w 322 as: zséfimfifi _ 225:: 2: II :2: eggs 11.... :25 eggs _ 22:34: .25: .2: mg a: 2:522 . “sigma“: _ :52“: 2:5: 53% 555% 22:22:: 5:5 _ Se 22;. E25: 2: 3:2 _ 3.52%; a: L , EEEEEEEE _ 22:3st .3: 11 , A 3525:: _ team .2002 p .z .. z 2: spasms a... a; 525.5% _ ”:22 E: 292528 2E . 3222:: 2:2 as 92.. 11 _ 2555 “2:252 3802 mEmmsg _ sea a 32: 3:52 sass .- n a: «.23 in-depth look at the system relationships, refer to Figure 9, Chapter II.) The process originates with the gathering of data in the home interview survey. This survey inventories all persons over five years of age and records all the trips taken in the previous 24-hour period. The survey includes socio—economic data such as age, sex, race, in- come, education, and characteristics of the trip such as purpose, mode of travel and number of persons in the car. During the data gathering, inter- views are also taken at a predetermined series of stations which cordon off the study area. Commercial vehicles and taxis are also surveyed. In addi- tion to interviewing, extensive traffic volume counts are taken on local sys- tem streets and classification of the types of vehicles using the streets are recorded. This information, along with the interview data, is used to cali- brate simulation models for the area. As stated earlier, four basic models are used to develop the total tranSportation system model. Trip generation models are developed to simulate trips produced by a given set of socio-economic factors. Models are developed for several purposes of trips, since trip characteristics by purpose reSpond to different types of socio-economic factors. Trip distribution relies on the use of the gravity model. This model is calibrated to a base year trip table built from the expanded survey trip data acquired in the home interviews. The model itself distributes trips based on the attractiveness between two areal zones of activity as a function of the physical proximity of those areal zones. Models are developed for different purposes during this process to correspond to the generation models developed. The network model is built to reflect calibrated impedances on the system. This model utilizes ground counts taken during the study and com- pares them to an all or nothing assignment of the base year trip table (based on expanded interview data). Calibration consists of attempts to match, with- in statistical limits, the interview counts to ground counts by adjusting the speeds used on the network. The assignment model itself is based on the assumption that a trip maker has perfect rationality and therefore knows and used the quickest way of reaching his destination. Other models are available which are based on the probability of the trip maker using different routes. The "all or nothing" minimum path routes method is the one used in this study. A recent book entitled Bureaucrats in Collision details many of the weaknesses of the process. The authors identified several areas where the transportation planning process (i.e. , the models) has basically shown a lack of an ability to respond to the real need of the decision maker. "One of the more serious problems in the area of tranSpor- tation studies involves their time schedules. Although the large studies were generally launched in a crisis atmos- phere of urgent needs, the actual design called for research to proceed at a measured pace, with the result that years passed without the studies producing any hard recommenda- tions. . . . In brief, because the studies and the persons re- sponsible for their designs and conduct were not geared to the realities of legislative and administrative timetables, they were unable to exert much influence on the problems they were asked to solve. "5 5Norman A. Abend and Marvin R. Levin, Bureaucrats in Colli- Lion: Case Studies in Area T raDSportation Planning (Cambridge, Mass. : M.I.T. Press, 1971), p. 13. One of the real contributing factors of this propensity for non— responsiveness is summarized as an undue emphasis on research in opposi- tion to basic technical movement. "The development of an array of old and new tools, such as computers, organization charts, "PERTing" and the study of land use alternatives opened up a number of fashionable temptations to indulge in expensive experi- mentation. . . . Another set of major problems arose be— cause technicians stress the pioneering research aSpects of project objectives. . . . They sought to make personal contributions to the solution of a variety of knotty research problems in metropolitan areas. "6 It is not hard to see the hesitation of an administrator to attempt to stifle innovation in his staff, but just such action is undoubtedly called for on "criti— cal path" kinds of operations. If the tranSportation planning process is to be acknowledged as a force for positive action in the future, the planner must solidify his technical expertise in such a manner that the decision-maker can both understand the assumptions of the process and appreciate the value these results can have for his community. We as a profession, have not been able to give local policy persons the kinds of data with which they can reSpond to local political pressures. Nor does the 24-hour, 20—year assignment process really help the local politician to answer his daily concerns. It is hoped that a peak period poten- tial, while applicable at a 20—year level, will be able to reSpond to the short- range needs of the local area and aid in establishment of short—range deci- sions concerning needed traffic and safety, geometric, and choice of mode problems. 6Ibid., p. 14. 2. Content and Purpose of the Research In reSponse to the many problems voiced concerning the 24-hour trans— portation planning process, this author proposes to develop a modeling capa- bility that will allow the tranSportation planner the opportunity to examine the most critical times of the day in the urban area in terms of travel activity and its effect on the urban tranSportation infrastructure. The peak periods of travel have not been adequately examined by the planning professional (es- pecially in the State of Michigan) because of emphasis on 24-hour capacity which dealt with very gross highway design problems only. The examination of the peak hour allows intensive study of the changing priorities of the nation, by allowing not only analysis of highway capacity problems, but better esti- mates of environmental and social costs which are at their peak during these times. The kinds of multi-modal questions that local decision makers are formulating concerning diversion of persons from one mode of travel to another are better reSponded to with a peak period modeling capability. The kinds of uses that such a model could be put to are briefly examined below. This list is certainly not complete, but will serve to identify areas where peak period information is becoming more and more essential. A large percentage of the travel activity occurs during the peak period in an urban area. Effective analysis of a person's movements throughout an area during these peaks is critical to decision making concerning transit deveIOpment activities. The directional volumes could be used to improve Scheduling of routes, and examination of travel interchange could allow deci- Sions in the areas of service expansion and the location of bus shelters. 10 A critical activity of the transportation planner is prediction of the design hour volumes in the area (the 30th highest hour of the year). These predictions are the basis for develOpment of design standards for new travel facilities. In urban areas, the design hour normally occurs during the peak periods. The peak model would give the planner a new tool with which to make these vital predictions. This would give structure to the now, very indivi- dualistic approaches used, by giving them basic common assumptions. A similar situation exists in the development of environmental impact statements. These documents require forecasted traffic during AM and PM peak periods. The estimates are again based on hand techniques. These pre- dictions are used to predict air and noise pollution on preposed new construc- tion so that the overall impact to the environment can be evaluated. TranSportation planning is moving into the development of congestion and micro-assignment modeling capabilities. Such models allow testing of congestion control devices such as signals and signs based on placement and timing parameters and provide information on turning movement at intersec- tions. These models require finely coded highway network models with peak hour volumes assigned (often by lane). The use of capacity restraint to interrogate tranSportation problems is presently based on 24-hour capacity concepts. Capacity restraint is an assignment process used to divert traffic based on a reduction of link speeds as the capacity of a roadway is approached. Through the use of peak period volumes and Speeds, much better analysis can be accomplished. 11 A critical concern, in many smaller urban area studies, is that over a 24-hour period, very few real capacity problems exist. This does not, however, mean that these smaller areas have no problems; only that the 24-hour analysis tools based on 20-year concerns are inadequate for identi- fication and resolution of these problems. In these areas, only the peak peri— ods really pose a threat to community safety, the environment, and local mobility. The peak period model can Speak to many of the kinds of small area problems more effectively than the 24-hour procedure. Analysis of the peak period can lead the planner to a better under- standing of the kinds of people who are in motion during this period by exam- ining the socio-economic profiles of these actors. Special knowledge of the employment categories or income groups are vital to enlisting citizen support for any effective planning effort for the community. Other important areas where effective peak period efforts could be useful are in the use of work trips and work trip interviewing to update major 24—hour studies. This would be done by establishment of relationships with- in the area between peak conditions and 24-hour conditions, the use of census first work trip information to update or review changes in the area, sub-area models, and parking models. All of these ideas are just scratching the sur- face for uses of the peak model procedure. As the procedure becomes imple- mented and refined, many, many more will surface. The purpose of this research effort, then, is to build and evaluate methods of producing peak period traffic assignments. These models will be based on existing 24-hour procedures as much as possible for three reasons: 12 1, Available Data 2. Available Facilities 3. ReSponsiveness As has been indicated earlier, information is gathered at the household level about the trip making habits of residents. The question of beginning and ending time is responded to, thus making it possible to determine when the peak hour is, its characteristics, and its movement through the area. Secondly, the Michigan Department of State Highways and Transporta— tion has a computer program package which is used to develop 24—hour projec— tions. The various procedures discussed in Chapter I, Section 2A, on "other concepts" are not, for the most part, Operating systems nor are they well understood by the practicing professional. The institutionalized 24-hour origin-destination procedures are well understood. Third, the transportation planner understands the basic process and because the local planning agen- cies are, likewise, familiar with the same process, no lengthy'period of acceptance will be encountered. This will make the process immediately usable and understandable because it requires a minimum of new programming and systems work to be done within the State of Michigan. The scope of this study entails the development of two types of peak models. Comparisons of the results are made to determine which process is most appropriate in terms of the manpower efforts required based on the vari- ous data and policy needs of the planner. One of the models will deal with factoring present 24-hour trip tables based on purpose and directional infor- mation to get a peak period model (post distribution trip interchange model). The second approach involves the development of all the classic models l3 discussed briefly in this section and in depth later on. This procedure, using standard trip generation, distribution, and assignment procedures provides a powerful tool but can be considered a bit bulky in terms of the creation of another complete set of models (direct trip end generation model). To Speed the process, the traditional model will use an already calibrated 24-hour net- work. Only the work purpose, because of its importance to the period, will be tested completely. This purpose provides an excellent vehicle to test the applicability of the procedure. This study will also use the person trip as the unit for modeling. The person trip is used instead of the more commonly used vehicle trip because of the growing need to emphasize moving people and not just vehicles. While much of Chapter I's discussion of peaking characteris- tics deals with the vehicle as it moves about the area, it is still appropriate to the understanding and the definition of the peak period, since, one of the person trips is that of the driver and the majority of the other person trips are auto passengers. Some of the concepts presented and defined (ADT, DHV, etc.) are tied to vehicle operation and if a person model were developed, it would re- quire application of a vehicle occupancy rate to reduce person trips to vehicle trips. This would be used in project analysis. The person travel peak (most peeple in motion during a given period of time) might very well be a different time period completely than the vehicle travel peaking period. This must be taken into consideration beforehand in selecting which approach should be used (i. e. , vehicle or person). Chapter I really emphasizes the lack of work 14 that has been done with person movement information, but is the best data available. It should be noted that the procedures developed to predict person movement are totally applicable to the development of vehicle models. After the models have been created and described, comparison be- tween the models and the base year data are made to determine the confidence level for use of the models and comparisons between the different type of models. With these tools in hand, it is possible to Speak to some of the cri- ticism voiced toward the process and Open new doors by examining peaking problems for which most physical tranSportation development is designed. Part I will deal with the description of the peak period and discussion of the methodology involved in the development of a peak period potential. Part II will look at the results of a case study conducted for the F lint-Genesee County area based on calibrating a work model based on the two developmental procedures. All data used in this thesis for the case study is drawn from the Origin and Destination Study conducted in the Flint-Genesee County area dur- ing the Spring and summer of 1966 by the Michigan Department of State High- ways and Transportation. PART 1 CHAPTER I WHAT IS THE PEAK HOUR ? 1. Defining the Peak Period The tranSportation planner has deveIOped many models to forecast 24-hour travel problems, and a vast amount of literature exists in the field; however, in the area of modeling travel peaking, less work is evident. The congestion of the rush hour, the over-crowding of facilities, the long delays and the increased travel time create many problems for the community from frustration and excess worry for the trip maker, to the aggrevation of an area's social problems because of the economic inefficiencies that develop. Table 1 reveals the percent of vehicle travel occurring during peak hours in major study areas throughout the nation. 7 It can be seen from the table that a majority of the daily trip totals occurs during the two peak periods; 5-9 a. m. and 2-8 p. m. These are the times of the day when work and school trips are very prevalent purposes. Both of these purposes make the most repetitive 7Lawrence H. Tittemore, et a1. , An Analysis of Urban Area Ravel by Time of Dy, Peat, Marwick, Mitchell and Co. , United States Department of Transportation, Federal Highway Administration, Office of Planning (Washington, D.C., 1972), p. 28. 15 16 00.00H 00.0 00.0 0N.0 0N.o 00.0 0:.H H¢.N 00.0 00.: 00.0 No.0 00.0 00.0 00.0 00.0 00.0 H0.0 HN.0 00.0 00.: NN.0 00.0 N0.0 00.H u0>0m Hawk 00.00 00.0 N0.0 00.0 00.0 00.0 00.0 00.H 00.N 0s.N 00.N 00.: 00.0 00.0 00.0 0N.0 00.0 N0.0 00.0 00.0 0N.0 00.0 N0.0 00.0 0H.m COP 1xoo»m 00.00 00.0 0N.0 0N.0 00.0 00.0 00.H 00.N 00.N 00.0 00.: 00.0 N0.0 0N.oa 0N.0 0H.0 00.0 0N.0 00.0 00.0 00.: 00.: H0.0 00.: 00.H .0000 .0Hoo 00.00 ad.o HN.o 0N.o 0m.o N0.o H0.o 0:.H Nn.N 0o.m 00.0 00.0 00.0 m0.oa 00.0 00.0 00.0 00.0 H0.0 0H.0 00.0 0N.0 00.0 H0.0 00.H 0p00 .naxo 00.00H 00.0 H¢.o 00.0 00.0 00.0 00.H om.H 00.N 00.0 NN.0 a0.: :0.0 00.0 N0.0 :H.0 ma.0 00.: m0.d 00.: N0.: 00.: Ho.0 no.0 m0.N 0HH0> 1mason 00.00 N:.o om.o 0N.o H0.o OH.H N©.H MH.N 00. 0N.m (VUWOWNOWDOMfi O O O c-Ixocodddmsxn \OMCDCDMWCO m0 l\ oavvdmm $0.00H 0H.H m wommmmm.m0mw urn-1:1: 0mm memdédddmhowmanNNr-t OOH :3 U) 0 1..] O—‘INCDBMHMNWGOOOQOM-imm .p 000900 0509 scam :0 mcauusooo Ho>aue zflaao mo Psooumm mucunmm use: ownwoumvcmpm ovcH coudhnsssm m>m>usm Q10 Scum Ho>0ue uwasoano> H m4mdm Adam 0H.0 00.00 m0.0~ om.H~ No.0 :0» 100000 00.0 00.H¢ 0H.0N H0.0H 0m.N .muam .oaoo 00.0 am.H# N0.0N 0N.NN 00.H 0000 .naxo N0.0 00.0: 00.:N m0.NN 00.~ 0AAA> umfisog 00.0 Na.aa 00.:N 00.HN N0.N mappnmm 0.0.92000 a manna 00.0H 00.na 0d.NN :0.NN mo.m 0050A 00.0 0m.H: 0N.dm 00.N~ m0.N Covmom mNH 1 0 wCHC0>m m0 1 N coocpmvm< 0N 1 0 000 002 00 1 0 wasnnos <0 1 NH npsox 003 .nanpoppam sosnom 18 trips (those least likely to involve personal adjustments) and therefore suffer the most because of the problem associated with travel during these periods. What, then, do we need to know that we presently do not know about the peak hour that is unobtainable from the 24-hour process ? Martin Wohl reSponds to this question with: "We want to know how usage and performance of the system and its parts will be affected. How much extra travel will there be’ ? ‘.Which and how many pe0ple will shift from one mode to another/ How many pe0ple will shift out of carpools either to driving alone or another mode ? . . .Will peak period volumes increase ? Will peOple shift from other hours of the day to the peak period as capacity is added ? Will the peak period shorten and by how much ? Will total daily travel increase ?"8 These kinds of questions are the ones that cannot easily be answered by any process in use, at the moment, in transportation planning in Michigan, and as such, are usually met with "best guess" approaches. It is hoped that a modeling process based on well known assumptions will give confidence to the planner, the local agency, and politicians to consider these important ques- tions. Congestion on the American tranSportation system is the most easily recognized result of the peak hour phenomenon. The levels of congestion are directly related to the simple fact that total vehicular travel is not uniform throughout the 24-hour period while the supply of highway facilities remains 8Martin Wohl, "A Methodology for Forecasting Peak and Off- Peak Travel Volumes", Highway Research Record Number 322, Travel Analysis, Washington, D.C. (1970), p. 183. l9 uniform. Three graphs (Figures 2, 3, and 4) from An Analysis of Urban Area Travel by Time of Day by Peat, Marwick, Mitchell and Company9 reflect the problems caused by congestion for the urban area. As can be seen during the peak periods, the volume to capacity levels almost reach saturation, travel speeds plummet dramatically and the cost of motor vehicle operation radically increases. The most noticeable growth occurs in the 4—7 p. m. time period. During this period, 40 to 42 percent of all daily traffic occurs. The morning peak is shorter in duration and not as critical even though the capacity of the system is often approached. The morning peak is, however, a potentially more stable predictor because of its shorter duration and work purpose orientation. Since it is obvious that for most of the day the system is not under pressure, it could be assumed that peaking characteristics are directly related to extreme cost of highway develOpment. This brief analysis explains why it is imperative that peak, as well as 24-hour volumes, be con- sidered in the transportation planning process. For Michigan cities, results similar to national figures can be seen by reviewing volumes that cross the screenline of the study areas. (The screenline is an imaginery line cutting the study area into two sections.) Table 2 shows that the 6—9 a.m. and 3-6 p.m. periods are the most active periods for travel. Often the size of the peak becomes more noticeable as the size of the city and total trips decrease. This is apparent from examina- tion of Midland's (the smallest area examined) peak being higher than any of 9Lawrence H. Tittemore, op. cit., p. 7. Ratio Volume/Capacity Speed (MPH) Time Cost 5 Running Cost (¢/VMT) 20 004' 0.; 1 v j' I j 3AM 7AM Noon 4PM 8PM Time of Day Figure 2 Effect of Travel by Time of Day on Volume/Capacity Ratio 60 40« 20 y 1 v I T 3AM 7AM Noon 4PM 8PM Time of Day Figure 3 Effect of Travel by Time of Day on Freeway Speed J 19 Time cost based on $3.25 per vehicle hour value of travel time. d 17' d 15 3AM 7AM Noon ATM ébM Figure A Effect of Travel by Time of Day on Travel Cost 21 Table 2' Distribution of Zh-Hour Travel Crossing the Screenline in Selected Michigan Study Areas Grand Ann Time Flint Rapids Arbor Midland Muskegon AM 12 - l 1.2 1.1 .6 .9 1.1 1 " 2 105 05 03 03 07 2 "' 3 07 .11 .2 02 on 3 " h 05 o .2 .1 .1 a - 5 03 01 01 01 .1 5 "' 6 103 on 107 02 o6 6 " 7 “08 “‘07 6.0 103 5.1 7 - 8 “.8 6.6 6.3 8.1 6.1 8 - 9 “.6 5.8 “.2 5.2 5.1 9 - 10 3.8 3:6 “.8 “.1 3.9 10 - 11 “'05 306 5.1 5.3 “on 11 - 12 u.6 n.3 6.5 5.6 4.6 PM 12 - 1 5:0 5.6 “.“ 8.9 6.“ 1 "’ 2 506 “09 307 60a “'09 2 "’ 3 5.9 “'09 605 500 5.0 3 - ’4‘ 902 709 808 508 9.0 “ ¢ 5 9.5 8.6 9.2 8.2 8.6 5 - 6 8.5 9.“ 7.4 10.0 8.“ 6 - 7 6.2 7.1 7.“ 5.6 6.2 7 - 8 5.“ 6.“ 5.9 6.“ 5.6 8 " 9 “'03 501 “on 5.1 “.2 9 - 10 305 309 208 302 309 10 - 11 2.5 2.7 2.2 2.3 3.1 11 - 12 1.6 2.2 1.3 1.7 2.6 Total 100.0 100.0 100.0 100.0 100.0 O-D Study Areas City Study Year Population Total Trips Flint 1966 ““6.903 1,“80.395 Grand Rapids 1965 363,088 1,200,578 Ann Arbor/Ypsi. 1960 12a,563 359.830 Midland 1969 51.876 261.927 Muskegon 196“ 117,500 387,165 “Information taken from Factual Data Studies published by the Michigan Department of State Highways and Transportation for each study area. 22 the others in both AM and PM periods. Peaks in the larger areas are more spread while the smaller areas show much sharper effects. This spreading of the peak is usually due to the more serious congestion in larger areas. In addition, the varied activities in the larger areas result in an increased 0p- portunity for non-work purpose trips causing an extension of the peak. Final- ly, the increased physical size of urban areas requires trips of longer dura- tion which maintains the peak period. Although, in most out—state Michigan areas, transit does not play a major role in the movement of people, in Detroit and nationwide transit usage is at its height during the peak period. The study by Peat, Marwick, Mitchell and Company10 mentioned earlier, indicates that during the most highly con- gested times a significant diversion of trips to public tranSportation takes place. The report concludes that "modal Split may occur as a consequence of high congestion, as represented by volume/capacity. Quite likely, there is no situation where volume/capacity is low and modal split is high. "11 Two other key considerations in the development of peak period data assumptions are the direction of travel on the route and the percentage that the peak period comprises of the 24-hour average daily traffic (ADT). The ADT is the average count for a segment of road for a 24—hour period over a 12-month Span of time. Thus, fluxuations due to time of the year or special 10Ibid. , p. 37 . llIbid. , p. 39. 23 events are leveled. For this reason, one cannot design a highway based on ADT volumes, rather a design hour factor is applied. This concept is typically referred to as the Design Hour Volume (DHV). This DHV is actually the 30th high hour of the year on a segment of road and not the peak. The 30th highest hour is used because it allows for those un- expected or unreasonably high travel periods to be excluded without allowing the building of a facility that cannot handle typical peak conditions. The DHV is always expressed as a percent of ADT. However, for all practical purposes, in most urban areas, the afternoon work peak is the DHV hour. Therefore, any peak hour model will be invaluable in projecting the kinds of data needed for estimating design hour traffic in metropolitan areas. The Procedure Manual for Use in Preparation of Traffic Analysis Re- BLUE.” presents tables from the Bureau of Public Roads capacity manual which lends some evidence of the close relationship between urban peak and the design hour (Tables 3 and 4). In Table 3, the rural percentages are wide- ly different from the urban activities. This is often due to recreation varia- tion that does not really occur in the urban sector. Table 4 shows the reader just how much impact recreation traffic can have on a road. Notice the wide range of DHV percentages. Congestion often causes people to change arrival and departure times which often lower the DHV. In most urban areas, the major vehicle travel purposes in the peak are work and showing. "If the 12 Michigan Department of State Highways and Transportation, Bureau of T ranSportation Planning, Procedural Manual For Use In Prepara- tion of Traffic Analysis Report (1969), p. 329. 24 types of trips using a roadway is quite regular and continuous through the year (work trips or shopping trips), they would add greatly to the daily aver- age (ADT), but would keep the percent DHV of ADT quite low (around 10 to 12 percent), as there would be few radical variations from the average day. "13 School trips fit very well in this category of phenomenon which keep urban DHV percentages lower. The functional classification of a facility will also produce variations. The higher the classification of urban roads, the less one can expect a high DHV, and vice-versa. Thus, travel characteris- tics of an urban route are vital to determining the design hour and tied closely to peak period activity. Any motorist is aware of the directional movements one encounters in the simple task of travel to and from work. Travel does exhibit heavy direc- tional movement. The morning peak shows a more pronounced directional movement because of the higher work purpose percentage. This directional movement is also apparent in Table 3. The rural roads showing greater variation due to an overall lack of traffic mix on the roads to balance the flows. Urban areas are more balanced giving lower DHV percents of the ADT. It is important to emphasize again that DHV is not a "real number" but a percent of ADT. Table 5 also shows this relationship revealing work purpose trips crossing the screenline in several Michigan urban areas by percent of each hourly total. The afternoon peak is less pronounced because of the greater purpose mix. Other determinants of the directional mix are 13Ibid. , p. 331. 25 Table 3 Percentage of ADT in Peak Hour for One Direction and Both Directions by Peak Hour. 30th Highest Hour and 200th High- est Hour. by Type of Facility One Direction 30th 200th Peak Highest Highest Type of Facility Hour Hour Hour Rural: Freeway 23.6 15.“ 11.“ Expressway 21.5 1“.1 10.6 Highway with more than 2 lanes 21.2 13.7 10.3 Z-lane 2-way highway -- -- -- Urban: Freeway 15:0 12.? 10.7 Expressway l“.6 11.“ 8.9 Street with more than 2 lanes 13.8 11.1 9.6 Both Directions 30th 200th Peak Highest Highest Type of Facility Hour Hour Hour Rural: Freeway 18.3 13.5 10.9 Expressway 19.2 12.7 9.7 Highway with more than 2 lanes 16.“ 12.7 9.9 Z-lane 2-way highway 19.7 13.6 11.2 Urban: Freeway 13.6 11.0 9.6 Expressway 11.6 9.5 8.3 Street with more than 2 lanes 12.0 10.0 8.7 2-1ane z-way highway 13.“ 10.6 9.0 26 Table “ Heavy Recreational Traffic Characteristics on Michigan State Trunklines Location in Michigan Station # Total DHV % of Recorder Locations “10 3“;l Houghton (Co. Rd.) “08 31.0 Rose City “ll-“12 29.1 Houghton Lake “06 28.9 Sterling 60“ 27.8 Port Sanilac 202 26.2 Brevort 302 2“.7 Farwell 206 23.6 Raco 203-20“ 23.5 St. Ignace 303 23.3 Sears 310 22.6 Baldwin 27 the distance that a road is from major employment or shopping centers and the physical geography of the area. Graphing of work trips in Flint shows the directional movement of these trips in the area (Figure 5). During the morning period, trips are directed heavily toward employment centers (purpose to work) and purpose from work is the dominant direction in the afternoon. Figure 5 reflects these movements. This kind of analysis is critical to the planner if any re- commendations concerning changing travel patterns by revision of working hours, etc. , are to be made. This kind of graphic presentation should be made for other purposes in the area to aid in this analysis. A study by Loewenstein of the relationship between work places and the journey to work indicates that there is reason to believe that locational and travel problems may be related on an inter-city level. The study con- tends that the type of employment and not the particular urban area is the best means of determining the directions of these trips. While this direc- tion will not be explored within the realm of this thesis, the conclusion that the "industry of employment rather than the city of employment is a better means to analyze the direction of these trips"14 lends credence to an attempt to model the activities of the peak hour because of the strong relationship between peak hour travel and the amount of work trips. It suggests that the peak hour distribution of trips may be similar in all areas; therefore, l4Louis K. Loewenstein, The Location of Residences and Work Places in Urban Areas (New York: The Scarecrow Press, Inc. , 1965), p. 194. 28 Table 5* Distribution of Work Trips Crossing the Screenline as a Percentage of Each of the Hour Period's Total Trips for Selected Michigan Study Areas Grand Ann Time Flint Rapids Arbor Midland Muskegon AM 12 - 1 51.5 33.1 10.7 11.7 39.8 1 - 2 6681 29.8 11.“ 8.1 36.1 2 - 3 “706 3307 10.8 “.1 36.3 3 - “ 71.9 “2.9 26.9 27.3 73.1 “ - 5 68.8 27.5 58.8 0.0 10.7 5 - 6 79.0 33.7 “5.2 20.9 62.8 6 — 7 83;? 63.9 36.1 16.0 7“.9 7 - 8 62.7 56.8 27.“ ““.1 65.8 8 - 9 55.1 50;1 8.“ “0.5 71.3 9 - 10 2705 2701 307 12.0 2709 10 - 11 12.9 11:8 5;“ 8.7 16.7 11 - 12 9.0 11.8 11.5 8.8 30.6 PM 12 - 1 15.“ 15.7 12.0 15.3 27.2 1 - 2 - 18(1 22.3 7.3 25.7 37.1 2 - 3 23:0 18.“ 11.2 8.6 35.9 3 - “ “0.0 28.3 12.7 10.“ “9.1 “ - 5 36:8 27:6 18.8 1“.8 “7.6 5 - 6 3807 3“.0 8.? 3003 5307 6 - 7 20.3 17.6 5.6 10.6 38.6 7 - 8 806 709 503 “03 230a 8 ’ 9 709 5‘2 803 307 31.1 9 - 10 10.6 11.6 6.9 7.2 5“.8 10 - 11 22.1 11;1 11.5 7.3 6“.3 11.- 12 37.3 22.5 10.7 17.6 70.? “Information taken from Factual Data Studies published by the Michigan Department of State Highways and Transportation for each study area. 29 mhsmflm camseofls .pcmam as agape sacs co emsam sacchpompsa m moo—mum can: s m n mm d c. d 7 d M d H M M. M W W C m W W M W w M...r H m w 0 W 0 0 ll” rill IIIIIlI-.-.-.-.- :.-...-. .D..~.9-- K N.— O I ..... 5 3...... 10m“. Mmommam croo— 3:... . ......... .0. 6 s. \\\\ oooo’....... ssss \ ..0 00.. \s s 0 833.0 0 es to .0 .0. sale see—- 0.... nooon v3.03 0... memmnm onoum IA 0 1 3.3.3:... V M :2. it: cos: m m m _saoom 800w .oosck 30 strengthening the applicability of similar procedures of distribution in each area. Also, attention given to different types of travel patterns for different industrial categories suggests that work models for the peak can be developed to reflect these differences. Table 6 shows that manufacturing employment, which is a very vigor- ously scheduled industry, has the highest percentage of trips during the AM and PM periods in Flint—Genesee County. As we move away from the person peaks (AM and PM), other types of industries begin to increase in importance. This information comes from Flint, Michigan, which is highly oriented to auto manufacturing. Other cities would surely reflect different types of industrial emphasis. This analysis would lead to the conclusion that Flint most likely could solve many peaking problems by revision of work scheduling. This section has presented many of the factors that are involved in explaining the peak period in an urban environment. It has detailed the im- portance of having a firm analytical approach based on this type of data. The uses for peak period data are endless as we tackle the problems of the move.- ment of persons and goods by auto, truck and transit, and the reduction of noise and air pollution by both capital intensive and non-capital intensive policy decisions. Decisions are aided greatly by adequate knowledge of the transportation system for any time frame but especially during the peak peri- od. With these concepts in mind, the study turns to a discussion of what others have pr0posed as possible approaches to the problem. It is signifi— cant to reemphasize, in reviewing these contributions, the strong desire of 31 oa.m oa.m NN.0H mm. on.: mm.mm 50.: 55.0n mm.© 0H. dd. oo.o1oo.m 2a mo.s mH.m as.m ms. mfl.m ma.sa ma.m sm.mm Hm.s mo. oa. oo.m100«: Em mm.m 00.: mm.m mo.~ ms.a nH.H om.m no.6 os.m as. ma. em. Ho.m ma.m we.m He.m mm.m sm.o an.~ oo.~ mo.~ :~.mm mm.mn Hm.wn ma.~ eo.a am. no. oo. oo. oo. NH. 50. om.s1om.m oo.s1oo.m oo.m1om.~ 2m 21 2a mo.m ow.m sa.o~ mm. 0:.u em.om sm.s mm.ma em.e ma. no. 00.0100.m 24 wasps compoa woos .mpmo Q10 passe museum zupmsvcH an uncapmm xmom mcfiusa Hm>mue accused e manna mn.m mu.a um.ma mm. mm.m os.o~ Hm.m :H.wm He.m no. mo. ooum1oouw E< honpo v:o&:ho>oo adsowmmmmohm nowpmmsoom moa>umm Hanomsmm mammmaonz cowvmvhonmcmse mcHhSPommscmz cowposupmcoo messes ousvHsowum< OHNM#V\\01\Q30\>< hhvmsucH 32 the author to continue to work in the area of making use of available tools to produce the best possible model with minimum data requirements. The models presented in Subsection B will reflect, in part, the theoretical construction discussed in Subsection A. 2. Peak Period Simulation Model Research A. Operational Versus Research Models An appropriate discussion to set the stage for a realistic assessment of the various models deveIOped in the field should center on classifying the different philosophies of model development available to the researcher. Two general categories of model types are in evidence: The operational model and the research model. Obviously, these two categories are opposite ends of the continuum with the pure form of either, non-existent. The discussion which follows on various peak period models will attempt to offer an oppor- tunity to place these models on that continuum. It might also be added that the author favors the operational approach, not from a lack of desire to ex- plore fully the intricacies of the peak period, but because of the need to effi— ciently utilize existing programs and data bases. The comparison between the two approaches concerns the operational method's emphasis on stability and predictability compared to the research methodology's concern for extreme sensitivity to very minute environmental changes. The research model looks to short-range, present predictions and descriptions based on a disaggregate data base. This approach stands in basic opposition to the planner's need for adequate projections based on 33 existing aggregate kinds of data. The operational model will emphasize sim- plicity and understandability compared to the research model's complexity and conceptual nature. The operational model reflects the need for a few highly aggregated models utilizing a few control variables. The variables in tranSportation planning are generally quite standarized and readily available. The basic advantage of these variables lies in their predictability and acceptance by local planning agencies. As a rule, operating agencies for political, as well as technical, problems are not willing to generate the kinds of disaggregate household level data that is required of attitudinal and econometric models. (i. e. , Not only is it difficult to project income levels for a 20-year period, but it is also not politically sound to predict that a certain area will be a slum.) In that sense, it is better to project surrogate variables such as automobile availability instead of income. Income is a strong determinant in the number of automobiles a person will own, and thus the number of trips he will make. Further discussion in this area can be found in a thesis published by Phillip Wheeler on Modal Choice Models. In effect, the thesis is probably misnamed since his emphasis, while directed to modeling transit generation, is really a discussion of all generation and diversion models and the categor- ization of these approaches. (See Figure 6. )15 Mr. Wheeler's theme is well represented in the following statements: 15Phillip Hampton Wheeler, An Evaluation of Models of Modal Choice, Thesis for the Degree of M.U. P. Michigan State University (1973), p. 61. 34 "The conventional system of travel demand modeling lacks a considerable degree of validity and accuracy, and should be improved to the point of accounting for the various feed- back 100ps now neglected, and of reflecting a consistent set of causal variables in every step of the process. This means that feedback processes such as the relationship of levels of congestion to the demand for travel, and the effect of land use to travel behavior in general, must be accounted for. This can be done either through the use of explicit demand models, or through an iterative procedure with sequential implicit models. In addition, regardless of the overall model system that is used, requirements of consistency must be met. Thus, for example, variables that are used to predict modal choice must relate concep- tually to the variables used in distribution, generation, and assignment. If time and cost are important in ex- plaining mode choice, they should apply as well to other travel decisions predicted. "16 His ideal model is the perfect research model. While the author, for one, will agree with a research or "ivory tower" justification of this approach, such a modeling process does not operationally exist for the transportation planner. This is the problem I am addressing. What then does exist; how can we best and most efficiently use what does exist, and within what guide- lines of predictability should the results be accepted ? The Operational ap- proach is the only viable method to develop a much needed modeling proce- dure. Where we can respond to the criticism of the operational approach, discussed by Wheeler and many others (Conference on Demand Estimation), 17 it will be incorporated. The chart (Figure 6) from Mr. Wheeler's thesis 16Ibid. , p. 120. 17Highway Research Board, Urban Travel Demand Forecasting, Special Report 143 (Washington, D. C. , 1973). 35 .3352 030:0 .2302 m0 sou—domhuumfio v mailman“ 03030 vsmfiov _ 1 92300.30 03000350 _ _ omsmnouousw1mwh use: A?» _ 1 so «uduosom. £3 EucYanus _ _ _ 0.3082800 330.933.» Hmcoflsokaou 3.30 _ _ _ _ A Hmfimoosoo 303350 F 1. 33008 03030 $005 36 would place this author's modeling approach under ”early" (trip generation) and "conventional" (trip end and trip interchange). B. Pr0posed Peak Period Models As has been stated, the work trip plays a large part in the composi- tion of peak hour trips. This analysis has led to the attempt to develop a peak hour model that utilizes the work trip as the basic estimator of not only peak hour trips, but total trips. Carl Shelbume Armbrister18 has proposed such an approach. The interest in this approach has been Spurred by the recent publication of the Bureau of Census relative to a 20 percent sample of households who were asked transportation questions during the 1970 census. This includes a "first work" trip question. The Census Bureau will aggre- gate this data to the zonal level for metrOpolitan areas. This will provide an excellent means of monitoring activities in the area. "The prospect of a national data base of primary home- to-work trips, coded to small areas, should provide a strong incentive for the development of peak hour vol- ume forecasting procedures from work oriented trip information. "1 It, therefore, is necessary to develop a relationship between total volume on the system and work trips. 18Carl Shelburne Armbrister, Primary Work Trips as Estimates of Urban Travel Patterns, Thesis for Masters of Science in Engineering, Uni- versity of Texas at Austin (1970). 19Ibid., p. 7. 37 Armbrister's study develOps four different models: PM non-directional peak hour, PM directional peak hour, AM non-directional peak hour, and 24-hour total. The study used assigned work trip link volumes and assigned total link volume at 221 locations throughout the Abeliene, Texas area. In the analysis, assigned work trips were used as the independent variable and total assigned volumes were used as the dependent variable. The basic idea was, if we already have an area model, we can predict area—wide traffic at various locations based on the census work trips' relationship to counted volumes at these locations. Peak hour as well as total models can be built. The major advantage of this approach is simplicity in development. The link volumes can easily be provided. However, using one equation to pre- dict all link volumes for all facility types and different purposes suggests real difficulties for the planner with the reliability of the results, especially as the number of links increases. There has also been technical problems in the coding of census files with non-responders and the "other" categories causing some skewing in the data sampling. This approach may be most use- ful as a short—cut monitoring technique. The model by Armbrister is of an early trip generation nature. A second study conducted under the auSpices of the Highway Research Program, in cooperation with the North Carolina State Highway Commission and the School of Engineering of North Carolina State University, 20 was 20John W. Horn, et al. , The Examination and Comparison of Peak Hour Gravity Models, North Carolina State University School of En- gineering (Raleigh, 1965). 38 directed to the comparison of the distribution activity of peak hour models. It was hoped that by calibrating peak hour gravity models significant improve- ment in the total model could be attained. This study was to be used as one of the determinants of possibly synthesizing origin-destination survey results. The study was conducted in Fayetteville, North Carolina. The scope of analysis included calibration of two gravity models: (1) Actual O-D gravity model and (2) reduce 24-hour matrix by purpose. "The purposes used were homebased work, homebased other, and non-homebased. Comparison of the results of the calibration indicated that the peak hour grav- ity model results, with both actual and predicted trip ends, more Closely ap- proximated the peak hour O-D trips table values than did the reduced ADT trip table values. "21 This means that we can expect the gravity model to do a relatively good job in distributing factored trip ends, if factored matrices do not deliver adequate modeling results. It was the hope of the North Carolina study that overall trip end pre- diction would be improved by calibrating gravity models for the individual peak period purposes. This result did not occur. The accuracy of predic- tion was not as good as expected because work trips did not account for the amount of peak hour travel as had been hypothesized. A conclusive decision on matrix comparison was not possible because of small interchange values. A final thought was that possibly a whole new means of predicting peak hour trip ends was needed. 21Ibid. , p. 22. 39 This study is important to the author in that it indicates that calibra— tion of peak period purpose gravity models may not be absolutely necessary as the 24-hour gravity models are probably going to give somewhat similar results. Also, some success was had with simply flat factoring 24-hour trip matrices and utilizing the trip ends as input to a gravity model. This model- ing effort falls under the conventional trip end category of our model charac- teristics chart. William Ockert, Richard Easler and Franklin L. Spielbergz2 have developed a peak period model based on the factoring of 24-hour traffic pat- terns. This model was developed for the Baltimore, Maryland, region. The model attempted to answer some of the data stability problems and cost prob- lems inherent in developing a second set of models based on the institutional procedure. It also attempts to respond to the problem of temporal as well as spacial trip distribution. "It is obvious that peak travel occurs at different times in various sections of the region; that, over the peak time Span, the "demand" is limited by capacity; that factors, such as working hours that are beyond the scope of trans- portation planners will influence peaking conditions; and that, for certain modes (transit in particular), the peak travel time is set by scheduling practices. "2 Because of these problems, a two hour period would be sufficient for demand analysis and evaluation. 22William Ockert, Richard Easler, and Franklin L. Spielberg, An Analysis of Travel Peaking, Baltimore, Maryland Regional Planning Coun- cil and Alan M. Voorhees and Associates, Inc. (Mc Lean, Va. , 1971). 23mm. , p. 160. 40 The results of this analysis confirmed that the type of industry in which the trip maker worked has the most to do with his peak. This concept was discussed earlier in the Chapter. Thus, the heavier industries with fixed hours tended to have the greatest peaking potential. The more service oriented industries in which more flexible hours of operation exist (serving those work- ing in the basic occupations) showed lessened peaking activities in the Baltimore study. Table 6, showing Flint's industrial trip mix during the peak, lends credence to this concept. A surrogate of this is then posed by the authors. It is assumed that as income levels rise, peaking characteristics should go down and vice-versa. This is due to the type of work a traveler engages in com- pared to compensation. The conclusion that is drawn from this analysis is that stability of the peak for forecasting is based on the occupational mix of the area. Table 7 shows that for the Flint area income was not as dramatic an indicator of peaking as this study would indicate. The Baltimore model was aggregated to a zonal level for application. The equation in effect indicates that the percentage of work trips between two zones can be determined by developing a factor for the median income of a zone to each type of industry in a given employment area. For non-work trips, it is considered inappropriate to use work variables. It is, however, stated that the relation to income does hold. School trips appear as a simple linear function. The model suggests that the higher the income is, the greater the percentage of homebased school trips will be in the peak period. "Other" purpose trips indicate a slight de- cline as income increases up to $6, 500 (in Baltimore) and then increases as 41 m:. an. an. on. an. mm. on. czacxca 00.03 00.03 30.33 00.33 00.33 00.03 00.03 000.03 00>0 00.00 00.00 00.00 00.00 00.00 00.30 03.00 000.03-000.03 00.03 00.03 00.03 03.03 00.03 00.03 00.03 000.03-000.03 00.0 00.0 00.33 00.33 00.33 00.0 30.0 000.33-000.03 00.0 00.0 33.0 00.0 30.0 30.0 00.0 000.0 -000.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 000.0 -000.0 00.0 03.0 00.3 00.3 30.3 00.0 00.0 000.0 -000.0 00.3 00.3 00.3 03.3 03.3 00.3 00.3 000.0 i000.0 00. 00. 00. 00. 00. 30.3 30. 000.0 .0 00.0-00.0 00.0-00.0, 00.0-00.0 00.0-00.0 00.0-00.0 00.0-00.0 00.0-00.0 0003300 03 000003 20 :0 20 20 20 :0 :0 00300 000000 0003 .0000 0-0 00330 00000 050003 00 0003000 x00m w03usa 30>009 0:0000m n candy 42 income rises further. This is logical due to economic constraints on a per- son's income which will make it difficult to make all the trips a person would want, with work trips being most important and declining as the purpose be- comes less necessary. After a certain level, however, income strata of the household allows for leisure activities and more trips will be made. Also, the higher one's income, the more likely one is to possess more than one car. The study indicated that non-homebased travel is difficult to quantify and that the peak period is similar throughout the study area. Little change is expected over time for this purpose. Model stability, over time, is a constant problem in all models. The stability of this model, because of its reliance on employment and income characteristics, appears to be good. Therefore, as socio—economic or land use conditions change, the interchange concept used can be very effective. The model leans toward the research end of the Spectrum with its problems of availability of disaggregated data and the length of time needed to develop its capability. It does, however, apply the model at zonal or aggregate levels by establishing zonal mean income figures. While this model is more conceptual than any of the others this far, it still basically is in the conven- tional trip interchange category. Our fourth model takes us over the barrier from the empirical to a conceptual approach. This model developed by Martin Wohl24 reflects sim- ultaneous interaction of all sectors influencing traffic generation and 24Martin Wohl, 100. cit. 43 distribution. The procedure reflects closely the conceptual econometric ap- proach. The typical forecasting procedure treats the segments of the process sequentially. Various activities, i.c., trip generation, distribution, modal split, route assignment, are treated separately at an aggregate level and brought together only after each phase has been completed. "This paper is to formulate a model structure such that these phases can be treated simultaneously, that the total amount of trip making (as well as the destination, modal, and route choices) can be varied with the trans- portation system and its performance and price charac- teristics (among other factors), that shifts from car- pooling to driving alone can be represented, that shifts from one hour of travel to another can be characterized, and that the amount of travel during peak hours (i. e. , the absolute build—up or decrease in peak and off—peak flow) can be determined. "25 Wohl proposes to examine the demand side of trip making which deals with the prOpensity for travel with "reSpect to service, price, and socio- economic conditions". 26 The supply side is a performance measure with re- spect to the amount of system available and the travel mix on that tranSpor— tation system. It is Wohl's hope that by reSponding to these demand and supply factors dynamically that there will be no need to differentiate between short-range or long-range models. The differentiation is implicit in our static model. Rather one model will serve to handle all periods of time. Examination of some major difference conceptually between Wohl's models and the empirical models might be useful. The real differences are 25Martin Wohl, op..cit., p. 183. 26Martin Wohl, op. cit., p. 184. 44 in the area of defining supply and demand. The empirical approach sees de- mand as the number of trips on the system at a given time. Wohl sees demand as a function or a prOpensity of a person's trip making. "Thus, a demand function represents the dependence of demanded quantity of trip making on the price of or ser- vice afforded by trip making. Implicit, of course, is that more trips will he demanded if either the price is reduced or the service is increased. "27 What this really means is, if congestion goes up (i. e. , service is lowered) trip making will go down. Supply is categorized as the "dependent relationship between travel service and the usage of travel facilities". 28 Basically, the supply function is viewed similarly to the capacity restraint concept which states that as capacity is approached, Speed on a link will be lowered and travel diverted to different routes. The author then proceeded to develop his model by de- fining: (l) The behavior of the trip maker in both behavioral and econometric terms; (2) the service or performance of the system giving a price based on capacity, controls, usage, technology, etc.; (3) the interrelationship of the supply and demand so that they are in an equalibrium and travel demand and price of the system can be determined. It is obvious, from the quick survey of literature, that much effort is necessary to adequately model the peaking environment. The hope of the author is that by the reasonable research of existing data sources, much 27Martin Wohl, op. cit., p. 186. 28mm. 45 that is implicit in the peak period can be ferreted out, and that logical assump- tions can be constructed. An obvious point of the discussion is that it is not reasonable for the practicing planner to devote the kinds of time that would be needed to accomplish a complicated peak model construction. It is, how- ever, also obvious that the subject of the peak period should be examined and decisions made about it and its effect on the total transportation system. Our next discussion will delve into how this kind of need can be fulfilled. Empha- sis on a strong operational model approach, based on the structure of a good "systems" planning context, will be explored. CHAPTER II RESEARCH METHODOLOGY 1. Systems Approach "Systems analysis is fundamentally an attempt to define issues and alternatives for the decision maker and then provide him with information relevant to his choice. "29 This definition sounds very much like what the function of the tranSportation planner should be. The popularity of the trans- portation planning process is based on its strong systematic and iterative nature. This ability to understand, duplicate, and calibrate the models serves the rigors of reproducibility well and puts the planner on a sound footing to make difficult decisions. Any attempt to model the character of the peak period must follow a similar process. Often the planner finds himself faced with the need to explain to other planners, citizen groups, local politicians, etc. , the genealogy of a proposed plan or project. With a strong process orientation, the planner can adequately indicate how decisions have been made, and that they were made in an estab- lished, systematic manner. This protects decisions from the questioning of 29R. DeNeufuille and J. Stafford, systems Analysis for Eggi— neers and Managers, (New York: McGraw-Hill Book Co. ,1971), p. 12. 46 47 the "motives" of the actors, and thus, limits the discussion to only the key assumptions of the process. Meaningful dialogue is then possible since pre- conceptions of secret discussions and "smoke-filled rooms" are dismissed. The systems approach forces the participants to discuss the validity of the assumption and not the process itself. Because the process is on-going (has a discernable history), it lays the groundwork for an eXplanation of why a project is now necessary or not necessary. Projects neatly lend themselves to priority development simply because the decision making is logically laid out, and subject to the feedback mechanisms of the process. Further, a systems approach adds the concept of total awareness to a project or plan. Changes made in one plan or program area impact upon the entire community and can be tested to determine this total impact. Additional controls can then be instituted and the process recycled. Thus, total impacts as well as limited project impacts are subjected to analysis. The systematic tranSportation planning process is illustrated in Figure 7. This figure shows the basic flow of activities from data collection to project implementation and reveals the feedback processes that make it such a viable process. 30 System analysis, then, as it is understood by the transportation planner, is a problem solving activity. The process provides an organiza- tional framework within which the planner is allowed to think in broader and 30United States Department of Transportation, Guidelines for Trip Generation Analysis, Federal Highway Administration, Office of Flaming, (Washington, D.C., 1967), p. 12. INVENTORIES ANALYSIS OF EXISTING CONDITIONS AND CALIBRATION OF FORECASTING TECHNIQUES FORECAST SYSTEMS ANALYms‘ Figure 7 < J t r. 48 . TRAVEL TRANSPORTATION 5c°“°""c.‘c“v'7" LAND USE CHARACTERISTICS menu-r. as AND POPULATION (Highway-Trans“) (Highway-Tuna!) I >- r\ I I I ECONOMIC ACTIVITY LAND USE Tn", ACCURACY CHECKS SELECTION OF AND POPULATION FORECASTING GENERATION ~— AND ..— “Tff OK”) “No PROJECTION 1‘ ECImes TRIP TABLES 00005 TLCRNIQUIzs 1 I INITLAI. ASSIGNMENT AND NETWORK ADJUSTMENT CALIBRATION or TRIP DISTRIBUTION a ,——— MODEL. COMMUNITY GOALS AND POLICIES II II II J\ FUTURE FUTURE FUTURE ECONOMIC ACTIVITY r LAND USP. /\ NETWORK AND POPULATION I ' I I V rUTURn 3 rUTURI: Tmr GENERATION TRIP DISTRIBUTION w ASSIGNMENT # nsonAcx TRANSPORTATION ”—§\ ’ \ \ I IMPLEMENTATION \\ I ‘0..." SYSTE MS ANA LYSS RCCOMM ENDED SYSTEM (Highway-Trans“) 'V Urban Travel Forecasting Process 49 more comprehensive terms about a particular problem. Therefore, in all planning, one must define the objectives and formulate evaluation before it is possible to develop and coordinate alternative solutions and make reason- able decisions. Normally, decisions are based on many non-technical as well as technical considerations. The systems approach helps the planner to at least remove many technical uncertainties from the process. The feedback mechanism of the process Shows that there is continual flow of new informa- tion and analysis which can be used to reorganize decisions based on the latest developments . 2. Peak Period Flaming Process The flow diagrams in Section 3, Model Development, illustrate the relationship both paralleling and diverging from the present traffic projec- tion process. All efforts have been made to utilize existing programs, data and procedure as they exist within the Michigan Department of State Highways and T ranSportation. The process itself is a modified tool reflecting work done by the Federal Highway Administration, Department of T raRSportation, 31 and the author. The State of Michigan does not, at this time, possess an operational peak period modeling capability. The procedure developed in this thesis will be adopted to do the work now presently accomplished through a "potpourri" 31United States Department of Transportation, Federal Highway Administration, Traffic Assignment Manual (Washington, D. C. , 1972), p. 128-142. sim h0p reh' of E fior Voc higl arr; gilt doc: keel 351 50 of hand techniques. The present procedures of personalized techniques are often called into question because of the lack of documentation of results or simply the failure to properly communicate complex assumptions. It is hoped that the strong process factor inherent in the peak period model will relieve much of this basic conflict. The flow charts to be presented are based on Michigan Department of State Highways and TranSportation program batteries developed in coOpera- tion With the State of Pennsylvania Highway Department, DOT, and Alan M. Voorhees and Associates. 32 The Transportation Flaming Battery (T P) is a highly flexible planning and analysis tool that affords the professional a full array of descriptive, utility and analysis programs. It should be noted that, given minor revision, the analysis of the peak period as presented in this document would be usable on other computer planning systems. However, in keeping with the author's desire to create a usable and understandable sys- tem for Michigan studies, such alternatives are not discussed. A brief dis- cussion of the TranSportation Planning package can be found in Appendix A. 3- Model Development It is important that the transportation planner have available to him as many positive tools as possible. With this in mind, it becomes critical 32Michigan Department of State Highways and TranSportation, seMSylvania Department of State Highways, Burroughs Corp. , and Alan M. oO'I‘hees and Associates, Inc. , A System of Transportation Planning Pro- wfor the Burroughs B-5500 (1969). 51 that usage of available tools be intensive. Development of new tools is both costly and often takes longer than the planner has to wait. The peak period model attempts to fit into this kind of intensive use mold. Available technol- ogy and data is utilized in new forms to provide the answers desired. The typical planning process (Figure 7) is well documented and exten- sively used. The attampt of the thesis is to develop new uses for the old pro- grams and realign the system requirements. The modeling system for each activity can be divided into three phases: Data interrogation phase, model calibration phase and an assignment phase (evaluation). These same phases are used in the 24-hour modeling effort. Statistical testing will be extensive and feedback opportunities are present at all levels. It is the provision of short cuts in trip end development, new iteration procedures, and the reflow- ing of the 24-hour process that is of real importance to successful usage of the existing tools. Figure 8 gives a brief overview of the two prOposed processes available to a planner to develop the peak period model. The parallel processes reflect two different philosophies within the cOIISizraint of building delivery systems that are responsive to the analytical nBeds of the planner. The matrix approach is directed to providing more "broad brush" systems kinds of answers while the more typical process based on the development of a new set of models should be used to achieve more p0liCBy sensitive kinds of decisions. Realistically, the matrix model should be ‘18 ed to make gross statements about longer range kinds of problems, Since it is strongly based on the 24-hour prediction. The synthetic approach (dirBCt generation, et. a1) can be used to develop means of setting priority 52 Q» (J 4 “‘0‘ $68 ( a [W .\3° Q0“ 9’6 q. A o\ 0" .3 F‘ \\°“ '° "‘0 0““ We “04 p h,- "‘ Vt" "I” ' I? (3' °°’ ’ «0 a K I" r} 90’; (6‘ \° Q‘°\° ‘o ’0’ \ \l. 0\ 5V9 9 '1' \ ‘o Project Planning ' Sketch Planning O-D Peak Period Trip Tablas Y v Purpose Direct Trip End Demand Model Calibration PostoDistribution Trip Generation Calibration Trip Distribution Model Calibration (Gravity Model) Evaluation § 0‘ ‘v Total Model Within Who} Parameter II the Model Usable? V Application at Peak Period Model ' Also can be used for sketch planning eflort depending on time-home and data need of study Figure 8 Peak Period Deve10pment Pracess 53 improvements based on decision both of capital intensive as well as non- eapital intensive approaches. It is up to the planner to evaluate the levels of uses he might have for the data and to utilize it as well as possible. It occurs to the author, however, that if quickness is essential, that the matrix mani- pulation be used initially, and after reSponsiveness has been proven, that the synthetic approach be introduced to fine tune the results. A. Trip Generation The generation of trips is perhaps the key element in the development of the peak period model. It is for this reason that many methods were ex— amined and many "yes" or "no" Situations were developed. The ability to either independently generate trips, rely on factored 24-hour interchange matrices, factored 24-hour trip ends, or a combination of all is the real strength of the peak period model development. (1) Regression and Rate Models (Direct Demand Estimation Process) A standard procedure for generating future trips is to relate socio- economic information gathered during the O-D study to trip information. Thus, regression or rate models by trip purpose are developed. This me- thod was also attempted for peak period development. Trip tables and socio-economic information for the afternoon and morning peak periods were developed. Correlation matrices were run to determine relationships between variables, and regression analyses were made to develop the models. 54 Problems not encountered when doing the 24-hour process were cri- tical. As stated earlier, the peak period projection requires that a greater number of variables be input in order to produce an effective model. There- fore, usage of the same independent variables employed in the 24-hour model presents problems. New variables that must be considered are time of day, direction of travel and trip purpose imbalance. The 24-hour model is based on the assumption that a trip when made, will also return, thus creating smooth directional assignments. The peak hour must deal with one-way trips, and the fact of uneven loadings at the zonal level. This fact causes problems in the development of the models and leads to the exploration of alternative generative approaches. (2) Factor Model Generation (Post Distribution Trip Interchange Process) The data problems encountered in development of trip generation equations and the need for a grosser level of analysis led to an approach that bypassed the possible need to gather further information. It was de- cided that factoring purpose trip tables by percentage of trips during the peak period and then applying a directional adjustment to the production and at- tractions would be a viable approach. As stated earlier, after the peak period was determined, it was broken down by the percentage that each trip purpose was of that peak. Fur- ther attempts were made to discover the directional flow percentage on the road at that period. These two figures were then applied to a 24-hour base year trip table. 55 A production and attraction purpose trip table was factored by correct percentage. Each of the resulting factored trip purpose tables were then ad- justed to reflect directional percentages. The results of this activity were then tested by comparing the O-D peak period trip table by purpose and the factored peak trip table by purpose. Statistical comparison of matrix interchange values and trip end val- ues and residual plot were attempted. In addition, assignments to the street network of the factored tables were made and evaluated based on a link by link basis, volume group basis, and residual plot in the total model evalua- tion. (3) Combination Models It is possible to develop models in combination of the above approaches. If a good regression model could be generated, then it would be retained, the future peak trip ends created, and then run through the gravity model. If good comparisons were obtained by the factor model method, it vas also re- tained. Furthermore, if an attraction or production comparison of a factor model proved adequately independent, then its factored trip ends could be re- tained and plugged in with a good regression production equation to the gravi- ty model and a model generated. All of the methods used singularly or in combination can be used to develop peak period generation models. B. Trip Distribution Since it first appeared, more than 10 years ago, the gravity model has been designated as the most prominent distribution tool for tranSportation 56 planning. All studies undertaken since 1964 by the Michigan Department of State Highways and Transportation have used the gravity model described in Chapter V to model area-wide travel distributions. The existing procedures and the calibrated models were tested as part of the peak period development process. It should be noted that comparison of the Gravity Model Should be performed whether a factor model or a peak generator model is to be used because it is important to know how good a job of matching distributions is being done. The matrix manipulator is basically a post—distribution diver- sion of trips. If the distributions are Similar, more confidence can be given to the results . C. Tests for Model Acceptability The basic goal in the verification of the models is to determine how well the simulation recreates the actual Situation in the field. That is, the aim of all simulation modeling is an actual versus predicted condition, and that the descriptive statistical tests used in this thesis can be applied in all comparisons. (1) Numerical (a) Mean (b) Standard Deviation (0) Standard Error of Estimate (d) Correlation Coefficient and Coefficient of Multiple Determinations (e) T -Value (2) Graphical (a) Actual versus Predicted Plot 57 (b) Actual (Independent or Dependent Variable) versus Residual Plot (0) Actual (Independent or Dependent Variable) versus Normal Deviate Plot These basic tests are drawn from a manual on the evaluation proce- dure in simulation modeling developed by Wayne Meyerowitz33 for the Michi- gan Department of State Highways and TranSportation. This manual attempts to systematically approach the verification procedures necessary to adequately analyze a tranSportation model. Numerical tests describe the validity of the calibrated models, while graphical tests attempt to explain the predictability of the model. Appendix C defines each of the aforementioned statistical and graphical tests and the procedure for their proper implementation. D. Process Development (1) Data Preparation The source of the basic information needed to construct a peak trip table is portrayed in Figure 9. To begin analysis, it is necessary to have a base year 200 character master origin-destination data file and, if possible, a 200 character merged person trip file (see Appendix B for data file sum- mary). From these files, the following basic peak period information about the area can be gained: (a) When is the peak hour, (b) the directional Split by hour, and (c) the percent purpose distribution. The peak period AM and 33Wayne Meyerowitz, Documentation Manual: Statistical Anal- ysis In Transportation Planning, Michigan Department of State Highways and Transportation (1974). 58 e033.“ 3 n 500 7:00.925 00‘hl. 0 .0153& 9 eBeU encezd .3 3.0 u ._. egos—i 0.. an... Eeuooet Pete...” 00m 0:» 00V 30 fine .I.h- I .‘0‘.,t 018:8 sneOl :< 5:33:30 suceavei £0.34 nth 3200 0.35.0... Awe—oven: mic; .0; en eeeezi 3 .53». 330 3:00 :3 A .4 d .t 2380 H Q :< sowpmnmmmnm mama aowmmm Noam a onswfim 03:0 JO... 05.50 35 3 5an 2:3. :( .ue_em 02:0 2.02.3.0... .5! o I .26.... 120nm ”ODO-P meet eemetepeo e» 20:23 a Vocmieehoo 00.0 ruoudw Odo“ ~0e> nan 59 PM trip tables, based on the most trips (person or vehicle) on the road for a given period of time, are developed as an input to trip generation (trip ends, by zone), and trip distribution analysis (trip length frequency distribution by purpose). Other necessary basic data elements are a 24-hour O-D trip table, 24-hour calibrated gravity model "F" factors, topographical or socio-economic factors, a base year 24—hour calibrated network, and skim trees (updated), 24-hour production and attraction (P & A) trip tables and base year socio- economic zonal data. The existence of a 24-hour process is assumed. (2) Gravity Model Testing and Calibration The testing process required to calibrate an AM or PM peak period gravity model is illustrated in Figure 10. Analysis in Chapter V shows exam— ples of the results of this testing. The flow chart lays out the ground-work to be followed by the practitioner. Necessary inputs at this juncture include a calibrated 24-hour gravity model, 24-hour network skim trees, and peak hour trip ends from the base year. Discussion of the gravity model formula- tion is included in Chapter V, as well as basic calibration techniques. If recalibration is necessary, it is best to begin the analysis with the final 24‘- hour "F" factors. (3) Trip Generation Analysis This analysis requires that the planner has peak period trip end data at the zonal level, and that he also has area—wide socio-economic zonal data to run regession or rate analysis againSt. The analyst should run a mmooopm soapmnpflamo Hacoz hpfl>wnu cofiamm 000m 0H madman .000303& ’02 Id I ‘3...” ATP JOIQ .ee» econ 5..» .0.” 013E ..o> ‘- ..eea 00 .02 no _w m u. m . . w _ «eve: + ................. I s... s. '0‘ 0.000 0:» 00.0. 29.»..auea 60 o. noez e 33.0000. a .ewoi ensuewuemmo> I!“ .7) e ..!0>m.Wr.. he ...Uanew( T :54 :2 c. II. . “Hr—000 O euwafi “NV—00 000N210 c. ._a>m a». a 02 r ‘7 .aec._3U.. 00 o. .01 0.0-m .e e303¢ so accurédEoU ..e0o.uom.u.. oEceeeum 00.0... Vu 3E 0330.... u tel 30030. .52000£k~.10x: 005°.U’mnuuo >0 .doaeam 3.7.. 0 r a). . x—na( p V< 2.50.00l L0. 3.2 0155 $2233 .ae» euum 61 correlation matrix initially to determine which variables are most Significant and also to determine the desirability of attempting some form of matrix mani- pulation. This decision would be based on lack of good independent variables in predicting peak period travel. The analyst may also wish to combine pur- poses other than work to eliminate much of the variance present due to the small number of trip observations. The diagram shows the iterative nature of the process (Figure 11). (4) Matrix Factoring The basic requirements for the development of a factored 24-hour model are: (1) 24-hour base year production and attraction purpose trip tables, (2) the percent that the peak hour purpose is of the 24-hour purpose, and (3) the percent directional Split during the peak hour. With this informa- tion, it is possible, based on the calibration procedures shown in Figure 12, to develop a suitable model. The basic calibration of the model involves the adjustment of the zonal production and attraction trip tables based on the directional Split. Productions and attractions (PS + A5) are discussed in Chapter IV. Development of the actual models is also discussed in Chapter IV. Having examined the criteria necessary to develop a peak period model based on programs and the data base available from a typical Michi— gan Department of State Highways and TranSportation origin and destination study, the following chapters will document a case study of these procedures to develop an AM and PM peak period work assignment model for the 62 0000000 00000000300 000000 000000 000000 0000 33 000000 .:0 00 A; i 5...: 42.6 D nae—l 03$ ‘1 .2000 .0. ..0< 0.0.0 .3200”. 0:02:53 aceteaeu Sen. 5 08.0. I.) CeiEam e301e> D e_e»_ec< .3333) Joe.— ee ecu at... 0‘2 00......2.< e535 .3230. .2. .030( 001.1 .3 23.30 . . A1 ee.c¢ nth .eatmoeu fierce... $250.23.» fiat—£0 £00.22 sci-2.3. Clem :2.an e30 .COUM I .00” 0.0 emm30e0( .30 0 0:0 eezeh ee.e¢ OE. .( d .L 00 L 04:22:05 leiei 3: T ce.eee.ue¢ Re.» 02m :20 3.2.0.00» :0;ng saleavu ‘1 “Beam“ 25¢ . O... .< d .a 00005.; .5 01cm nth aoem 2‘324 39.00:!” 33 eaten-0m 0 £00.23 00.0. .eawneem ee.ae_0a> I02 3.03:4 5° is.._..do a“00_ac( ‘5 .CeEEam 01_Ioe.< Ea.oo.& vatea Joel 10:32.3 .230 5:02 cote—QBU 5.32 A $0.31 400.. Cote—OCOU 63 mamzamc< mmcmnonmch 0: 03:00.8; z...» 7 3.32025 .0 :02 SE22:— ten .20 0:0 60.0. .0) Jo< mEo‘U mk co-_0anoU .P.h 31°C 8.0. .25. .213 .0002 .200 4 Onasak )3 .k .._. 00.0 00 00.3—am 00m 0:... :28 L9 7:23.05 * JOOl 3003...- $0830& .0. .._. .0... vn 001m». _ 03:10! 3th st... 2200 003000300030 0000 003000 0000 N3 000030 30> :on 64 Flint-Genesee County, Michigan, Metropolitan Area. Mention has been made periodically through this thesis concerning some of the findings of the study. Chapter III will detail the socio-economic characteristics of the study area and profile more closely the peak hour in the Flint area by examining basic travel data. Chapters IV, V and VI will detail the calibration of the work models and evaluate the model's effectiveness. PART II CHAPTER III FLINT, MICHIGAN: A CASE STUDY 1. Introduction Having concluded the discussion for the need of a peak period model- ing effort, and having detailed the process by which such an effort can be accomplished, the study will turn its attention to the testing of the modeling system. The area in which the process will be tested is Flint-Genesee County, Michigan. Figure 13 shows the study area. This area was chosen for a peak period study for several reasons: 1. An Origin-Destination study was conducted during the spring and summer of 1966 making additional data requirements minimal. A completed set of 24—hour models have been de- ve10ped and the author has been intimately involved in that development process. The area to be studied (Genesee County) is large enough to offer great variability in travel charac- teristics, i. e. , purpose and trip length. The area is compact enough not to require redefi- nition of the areal zone configuration to account for homogeneity. This is very important when using aggregate models, otherwise the procedures may not be applicable to small urban areas in the state. 65 66 Figure 13 Flint-Genesee County Study Area Map L W .5 .r J .Jl "‘ 67 As indicated, Part II will document a case study of model development, beginning in Chapter III, with the analysis of the socio—economic characteris— tics of the study area, problems inherent in the use of aggregate survey data, and a profiling of the characteristics of its particular peak periods. Analysis of the model development will be limited to examination of only the AM and PM work purpose. Chapter IV will look at the generation of trip ends by use of regression and rate analysis techniques as well as matrix manipulation, and Chapter V will examine the distribution of work trips in the area (other tra- vel purposes will be briefly discussed). Chapter VI will provide procedures for testing the total models that are deveIOped by the process (in this case, only the models for work trips in the PM for matrix and synthetic will be tested). 2. Flint-Genesee County Study Area The data utilized in this study were obtained from the 1966 Genesee County Origin-Destination Transportation Study. The study area (Flint- Genesee County) had a 1966 base year population of approximately 430, 000 persons. As in many Michigan urban areas, the automobile and automobile related manufacturing is the major industry in the study area. Manufactur- ing employment accounted for about 60 percent of the approximately 147, 000 employees in Genesee County. Almost all of the manufacturing plants are located in the City of Flint pr0per. Flint, located in the center of the county, contained about 50 percent of the total population and about 48 percent of the 68 total dwelling units of the whole study area. Table 8 provides a summary of the 1966 profiles of data necessary to understand the area. 34 3. AM Peak To ascertain the one-hour period during the morning and afternoon when the most person trips throughout the area are on the road, the 200 character base year data file (description in Appendix B) was interrogated by the program "Tape Select". The results of this selection process were then used to develOp AM and PM trip tables. (This process is discussed in Figure 9, Chapter II). The resulting base year peak period trip ends are discussed in the next section. A thorough understanding of the character of the modeling period is essential to understanding and using the model for projections . Table 8 Basic Data: Flint-Genesee County, Michigan 1966 Socio-Economic 1966 Residential - Dwelling Units 124, 765 Population 432, 946 Resident Labor Force 141,127 Autos Available 163,104 34Michigan Department of State Highways and Transportation, Bureau of TranSportation Planning, Summary Report: Flint-Genesee County TranSportation Study (1972). 69 Table 8 (cont'd.) Employment - Manufacturing 87, 894 Wholesale-Retail 22, 047 Service 7,171 Other 12, 387 Professional & Government 18, 201 Total 147, 685 Travel - 24-Hour Total Purpose Trips Work 233, 560 Sh0p 309,280 Social-Recreational 17 3, 358 Other 430, 941 Non-Homebased 285, 357 Truck 107, 095 Cordon 35, 211 Total Internal Trips l, 541, 633 Total Trips l, 576, 844 The morning period in the Flint-Genesee County area during which the most persons are traveling in the area falls from 7:30 a. m. to 8:30 a. m. based on the 1966 origin and destination interview data. Table 9 shows the purpose breakdown of trips for this period. The hour period accounts for 11 percent of the daily person trips in the area. Several facts concerning the AM peak were in evidence from an exam- ination of the data. The most obvious being the strong directional flow of traffic in all purposes from the residential end of the trip to the attraction end. Apparent also is the existence of a great number of school trips during the period. To account for this seemingly serious error, because of the lack of expected work trips (see Chapter I) in the AM peak, a test was run to de- termine purpose distribution around this hour period. Analysis of Table 10 shows that school trips are rising during this period. This partially eXplains 70 the reduced number of trips to work for that period. Further investigation was needed. Table 9 AM Peak Person Trips by Purpose (Internal Trips) To Purpose From Home Work 28, 266 Business 2, 713 Shopping 1, 597 School 101, 143 Social-Recreational 3, 477 Change Mode 531 Eat Meal 208 Medical-Dental 273 Serve Passenger 8, 302 Total Homebas ed Total Internal 4. School Trips in Peak Periods Analysis of both morning and afternoon peak hour travel, by purpose, From Purpose TIoiHrnne ffoual 1,983 30,249 431 3,144 581 2,178 702 101,845 387 3,864 75 606 33 241 273 1,629 8,661 151, 061 154,366 92. 0 100. 0 resulted in the discovery of an extremely large number of "other" purpose trips. The "other" purpose during these two hours constituted almost 50 percent of the total "other " purpose trips for the 24-hour period. It was ex- pected that the work purpose, not the school (other) purpose would be the major contributor to peak periods. It is commonly accepted that work trips will account for a relatively large percentage of peak hour traffic. In Flint, this was surprisingly absent. What could cause such a massive infusion of ”other" traffic ? 71 0.0 N.N N.0 0. n.0N 0.0 0.0N H.0 a.ma 00.NH-00.HH N.NH 0.0 N.m N. 0.0a 0.HH 0.00 0.0 0.0a 00.HH-00.HH N.N 0.0 0.H s. m.ma H.0 0.0m N.0H N.NH 00.HH-00.0H m.0 N.0 0. H. H.0N m.N a.mm m.0a a.HH 00.0H-00.0H N.0 H.m N. N. 0.NN N.N 0.00 0.0a n.0N 00.0H-0m.0 m.HH 0.0 0.H N. 0.NN 0.N a.mN m.nH N.NH 0m.0 -00.0 n.0H m.N s. N. m.ma 0.5a n.0a s.m H.0N 00.0 -0m.0 m.m~ 0. N. N. 0.m n.0m N.N m.N 0.0a 0m.0 -00.m m.mH N. m. N. 0.0 0.00 N.N 0.H N.Nm 00.0 -0m.N 0.0H H. m. N. H.m 0.N: 0.. N.N 0.00 00.0 -00.N m.HH H. a. N.N H.m N.0N 0.N 0. 0.00 00.N -0m.0 0.HH 0. m. 0. n.N 0.H 0.H N. m.N0 00.0 -00.0 m.N 0. H. m. 0.H 0. n. 0. N.00 00.0 -0m.m 0.0 N. s. m. N.H N. 0. n. 0.00 om.m u00.m H.0 0. N.N N.N :.H 0. N. 0. n.00 00.m -00.: n.0N 0. 0.0 0. 0.N 0. 0. 0. N.00 00.0 -00.: H.0m 0. 0. 0.0 0. 0. N.N 0. H.0m 00.0 -0m.n 0.NH 0. N.N 0. 0.0 H.H 0.0 0. n.00 00.0 -00.m 0.0 0. 0.0 0. H.0 N. 0.H 0.H 0.0m 00.0 uom.N a.ma 0. n.0H 0. 0.HN 0. H.m H.n 0.00 0m.N -00.N 0.0a 0. 0.0 0. 0.NN 0. 0.N 0.N 0.00 00.N -0m.a a.ma 0. N.0 0. N.NH 0. N.N m. 0.00 00.N -00.H 0.0H 0. 0.0 0. a.mH 0. 0.H 0. n.H0 00.N uom.NH 0.0a m. 0.0 0. 0.NN 0. 0.N m. 0.0m 0m.NN400.NH z< uomcwmmum Hopcon H002 woos Hmcowpmohomm Hoozom mcannonm mmocfimsm xpoz coahmmuuso: o>hom HNONomS pom mmcmno Hmfioom vOMmcmua . weapon mews scam nu ma omonusm zoom page omwvcoouom on» an PaosouocH usom Hadm an omonusm an mnwua.comuom.aum_dm vcwam OH manna Table 10 (cont'd.) Social Work Business Shopping School Recreational Transact Serve Passenger Medical Meal Dental Change Eat Mode Hour-Period 72 MM‘OQNWHO\N3 HO‘QNMWOOHMdOmO . . O . O C O O . NQBNNQNNOWMNN(\BNBCDO\O\OHN\OH H HHHHHH HHHHN HOCHNOm-ttBMM42BNmOdNMHr-lm0m NMMWMHNNNNNN H H H wmnmamaamddeémooxoooer-nxma WMNHH HHH HOéMdJWWdOWMN ##ét—ld’dMBHMWBHHdHOr-IHMVNMOH H (\BH WOOD mmoo (\amnxnomo O‘\f\~+ (I: mb~\O\O H O . . . . . . . . . . . ° m$§6mmm$dfim§fiaom HHHHMdssssssssmm \OM\O\O MWN 0 [\MN C (““053 OM44} “202% O gdwiogmimmmdmmmmaamma \OCDCDMdNNmMWNWHr-IWQCDCD(EWc-ld'd‘N coo coco-0.000... coco. “NMHMMNQJNNQOQNOHWQJOV‘N: NNMMNHHHNNHHmNMMMNH HF. CO\OO\\OO\V\H-=TOO\H3NM®®V\\OL\O < o h u o . n o m a 000.0 000.0 «an.o doa.o “00.00 owu.o 00”.: oa~.o omu.o oeu.o ~u~.o ooo.a JIL sarcoma: sno.o uu~.o Nmn.0 ud~.o nuo.o onN.o hem.o N.N.o an".o nu~.o nmfl.0 000.0 onnoo 000.— ul ‘JWUIIIJF- wakes: Coapmaouaoo ma manna noo.o ano.o nnn.o enu.al Ono.o un~.as o~«.ao anu.oo 550.00 000.00 ano.o on0.00 )HOQOI nno.OI 300.“ “1.. 1‘ZIL nun.o N4N.0 0nn.0 0n0.0. 00N.0 owo.00 oncooo Nso.oo fluOoOI 000.0- omn.o nou.o «mn.o ooq.o oo».o coo.— uxa. OHIU¢ ocu.on o0000 ou«.o «00.0 “do.o «no.0. cfio.o noo.o ~no.0 swo.o ono.o omo.ou 00 L .o nom.o un~.o "a”.oo oeo.on 000.» N1) 131.31. uc~.o 000.0 000.00 «no.0 noo.oo moo-o. ”00.0. n~o.uo ova.00 uoo.o Noo.o no”.o onu.o «no.0 nnn.o 000.00 000.” an O NV znx< no“ nu“ an” no" and as no an a“ a“. ‘J ~u¢uwoxuzuz “ZdAI7UZ 14 «rm; x< acmx It uuoo¢m mam awn-«mx .10 aux.mm.. mxu 5141 axu mmxku cm 6.0: 0N «1m; 84 summarized in the development of the work models depicted in the following pages. After a review of the matrix, the best variables were entered into a stepwise regression analysis. Analysis based on statistical and graphical criteria, established earlier, are used to determine acceptability. This analysis included consideration of potential Special generators. These are zones with residual values (actual — predicted = residual) outside of 3 or 4 Standard Errors of Estimate. If valid socio-economic characteristics can be identified to explain the over or under prediction of trips by the equation, they are removed from the data base and Special trip rates established. (Example: A major ShOpping center with many trips but relatively few em- ployees). With these "Special" generators removed, the equation is rerun. The process continues until a best model is developed. Figure 11 in Chapter 11 details the calibration procedure utilized. A. Homebased Work Production and Attraction Equations - PM Period The afternoon peak period for person trips in the Flint Metropolitan Area reflects almost 19 percent of total trips for that period. Approximately 45, 763 trips have been identified by the modeling process as being homebased work. This is approximately 20 percent of the daily total of person work trips of 233, 560. This is not the dominant percentage that might be antici— pated for the work purpose. This, as mentioned earlier, is based on several reasons. A major one being the use of a person rather than vehicle trip. 85 The smaller percentage also reflects the greater mix of travel during the period. These occur as individuals stop on the way home from work for a variety of personal activities. During the AM peak period, only 8, 695 non- homebased trips were identified, while the PM peak Shows 40,449. Thus, many potential homebased work trips are lost to this non-homebased category. The non-homebased purpose is usually discussed in total terms since they have similar travel characteristics. (1) Homebased Work Productions PM Analysis of the correlation matrix (Table 13) shows four variables exhibiting a strong relationship to homebased PM work production: These variables are p0pu1ation, resident labor force, autos available, and home- based work production for a 24-hour period. All four variables are strongly intercorrelated. The variables of population and 24-hour homebased produc- tions were chosen to develOp the model. These two models were developed because it was felt that since 24-hour projections were already present for these variables, that very expediently, peak period projections could be developed and analysis begun. The high correlation between 24-hour work trips and peak period work trips also opens the door to research into using peak period interviewing as a means of generating total area prediction. This would cut the time and cost of new origin-destination surveys and add impetus to the peak period research effort. Obviously, other purposes and other time frames should be examined for each area before actual short-cutting of the process can begin. 86 The average number of work production trips, per zone, is 145 with a standard deviation of 130. One independent variable was included in the final equation for both models. Population was chosen for one model because of the high relationship of the variables to the dependent variable. In the PM peak period, each person in the area generates . 09 work trips. No special generations were identified for the purpose using population as the indepen- dent variable. Appendix D contains the 24—hour models calibrated for the Flint-Genesee County study and presently being utilized in the development of future travel facilities planning. The equation using 24-hour homebased work production as the depen- dent variable resulted in an excellent prediction model. Almost .19 peak work trips are generated for each 24-hour work trip. This reflects closely the 19 percent that work trips are of the 24-hour work trips. The 24-hour model was deveIOped based on resident labor force data which was the better prediction for the entire day. Resident labor force and homebased PM work production correlate very highly, but homebased work production for a 24- hour period produced a better prediction model. Again, no Special genera- tors were encountered. The statistical data for both equations are presented in Tables 14 and 15. Analysis of this data reveals that the models are well within acceptable limits of reliability. Figure 14 and 15 show the excellent predictability of both models. It is recommended that either model be used for any future attempt to model homebased work trips for the PM period. In testing the PM 87 Table 14 Regression Summary - Homebased Work Production PM (Population) Dependent Variable (Y): Mean Value (Y) Standard Deviation (Y) Multiple Correlation Coefficient (R) Coefficient of Multiple Determination (R2) Standard Error of Estimate Constant Term Number of Observations Special Generators (None) Unit of Analysis The Equation: Y = 8096 (X1) + 12.3205 Independent Regression Variable Coefficient (x1) Population .096 Homebased Work Production PM 145.27 130.83 .87 .77 273.92 12.32 315 0 Zone Simple Correlation With Y .87' T Mean value Value 32 1196.98 88 ACONpmasmomv Em m:0flposooam x903 Ummwnmsom «voam Hmsuflmmm 0H mpzmflm It 030... 00.000 00.0«0 00.0ms 00.0no 00.ocn 00.000 00.00" 00.0NN 00.00“ 00.00 00.0 mOIIIIIIIIIOIIIIOICIIOIIIICOII.OIIIIIII8898IIIOIIIIOIII'IIIIIOUIIIIIIOIOUIOOIIUIIOIIIIIIOIIOIIIUIIIOIOn u C HI 000¢mm0 N N N N N N N N N N N N H N H o H N N N N: 00.0an N N N N N o H u a H m c c H a o o n H a o no N M C C C i C Iw H 0 I Cl C I a. D N J H 0.. a o c a o o “a 00.3... 0 H 0 ace 0 c e out. a on a u ace out. on not 0 5H 0 N O c a i can... ecu-.00 a an M n a o o a s a. so. 0.00.00... on m H c a coat I at c. at. 9 H w a O can 0 see c a... a as H i H e on c 9 cc. 0 c as. o co m N c .0 so a c c c c 0 a. use... w H .0 c c c c c 0 o H H a a a c a m. 00.0» H a c c a q u u C C CD I H u I o a c n N o c N N c . N H a N N o N N N N N N o N- 00.0Nm N o N N N N N N N N N N o N N N N N N N N N- 00.00n uOOOOOIIIII0-0-8-0...OIIIIIOIOIO'IIIIOIIIOIOIIIIIIIOOSIOOIIIIOIIUOIIIIIOUIIIIIISO§IOIOIOIII+IIII8|IOIOH 89 Table 15 Regression Summary - Homebased Work Production PM (HBWP24) Dependent Variable (Y): Homebased Work Production PM Mean Value (Y) 145.27 Standard Deviation (Y) 130.83 Multiple Correlation Coefficient (R) .91 Coefficient of Multiple Determination (R2) .83 Standard Error of Estimate 53.30 Constant Term 5.61 Number of Observations 315 Special Generators (None) 0 Unit of Analysis Zone The Equation: Y = .1884 (X1) + 5.6115 Simple Independent Regression T Mean Correlation Variable Coefficient Value Value With Y (X1) Homebased Work Productions 24 Hr. .1880 39 741.46 .91 9C) Admmzmmv 2m mcowvozcohm xhoz UmmMflmEom «poam stcfimom ma mhzwfim 1m mxmx 00.000 00.0N0 00.0NN 00.000 00.000 00.000 00.000 00.0NN 00.00N 00.00 00.0 HOIOIOIIOIIOIOIIICOIIOOIIIIIIIIOIOIIOOOIOQIIIIIIIIIOOIIOOISIOQOOOUIIIUIOIIIIIIDII+I|IIOIID|OIII'OIIIOO~ N 7 00.000- N N N N N N N N N N N N N a N N N N N N . N- 00.9,:- N . N N . N N u N N c o a c e H N N N 0 o c u N o a c e c N H I C IN H O C C C C I m J N O c a to: e a a o NO QUILOO 4 H C C O C O I. I a. CO. C Cm J N o cc .0. u o a cue-no 0 0H m u a sue a he or :00 c on m H C c O c O a t a cache... .0 a an m N a .0 u 0 0 out. 0.: 0.0.6:... N w N c 0 so out. O a one It. 90 0.0 u .-. u c a a a c o. o to. I v a. a N o c a c o a the... N a C 0 CC I CC C u N c a 0 cc 0 o a N: 00.00 N e c o o co. m H .0 e .0. a 0 c a e e a N H C C H N .0 o a N N o e N N N N N N N N . N. 00.0NN N N N N N N M I H N N N N N N N N N N N N- 00.000 NOOIIIIICIIOil...0.0-6OIOIIIIIOOIIUII.IIIOIOICOIOIIO'IIUOIOCIOIIIIIIUIIQIIIIIIIIIOOIIOIIIUIOI|I|OIIII+~ 91 total model in Chapter VI of this thesis, the equation based on population will be used to generate the synthetic trips. (2) Homebased Work Attractions PM Examination of the correlation matrix (Table 13) presented several possible independent variables that could produce adequate models. Many of these would have required the combination of more than one variable. After testing the many possibilities, two variables were determined to best satisfy the need of the calibration process. Total employment has a simple correla- tion of . 902 and was used for one model. As with the homebased work pro- duction equation, the variable for the 24-hour period also provided good predictability. Two models were deveIOped giving the planner a choice as to which model would provide the best results for the needs of the study. Possi- bly the use of the model based on the 24-hour projections for long-range requirements and the total employment variable for short-range or sub-area decisions would be appropriate. The average number of work attractions, during the afternoon peak, was 145 per zone with a standard deviation of 632. The high standard devia— tion reflects the large number of workers employed in Flint in very few activity zones. Table 16 shows the zones, activity in the zone, percentage and 92 cumulative percentage in each employment zone.40 Over fifty percent of the persons employed in the area are in 12 zones. This can cause problems for predictability but in the homebased work attraction PM model, such was not the case. Each employed person in the attraction zone generated . 36 trips. The homebased work attraction 211-hour variable generated . 25 trips. The 24-hour model for homebased work attrac- tions was also based on total employment per zone. The statistical summaries of the models are presented in Tables 17 and 18. The data indicates that each model is predicting within acceptable limits. Figures 16 and 17 reveal basically good predictability in the models. No Special generators were identified for the area. Either of the equations is recommended for use in projections of future year volumes in the area. In testing PM total synthetic models in Chapter VI, the attraction equation based on total employment will be utilized. B. Homebased Work Production and Attraction Equations - AM Period The AM peak for person work trips, in the F lint-Genesee County area, includes 18. 5 percent of the total trips during this hour and 13.1 percent of the 24-hour work trips. This is the second largest purpose movement during the period (second only to school trips in importance). The two purposes 40Michael Eberlein, et a1. , Travel in Genesee Comm in 1966: The Impact of Land Usefloon the T ranSportation System (unpublished Report, Michigan Department of State Highways and TranSportation, 1975]. p. 23. 93 Table 16 Base Year Employment by Zone (1966) Rank Zone # Employment Percent Cum.% 1 #2 18136 12.28 2 112 11977 8.11 20.39 3 103 7915 5.36 25.75 h 102 7355 5.32 31.07 5 3“ 7686 5.20 36.27 6 186 6321 0.28 “0.55 7 55 5362 3.97 44.52 8 287 4051 2.7h #7. 26 9 100 3840 2.60 09.86 10 237 3566 2.01 52. 27 11 3 3038 2.33 54.60 12 h 3314 2.20 56. 8h 13 57 2686 1.82 58. 66 1h 69 199h 1.35 60.01 15 56 1976 1.34 61.35 16 ' 1 181k 1.23 62. 58 17 18 1635 1.11 63. 69 18 2 1h63 .99 6h. 68 19 106 1385 .9“ 65. 62 20 20 119a .81 66. #3 21 52 1027 .70 67.13 22 lhl 995 .69 67. 77 23 51 903 .60 68. #1 24 236 943 .6“ 69. 05 25 6 918 .62 69. 6? 26 138 876 .59 70.26 27 27 871 .59 70.85 28 105 829 .56 71. 41 29 16 819 .55 71.96 31 5 7&1 .50 72. 97 94 Table 17 Regression Summary - Homebased Work Attractions PM , (Total Employment) Dependent Variable (Y): Homebased Work Attractions PM Mean Value (Y) lb5.27 Standard Deviation (Y) 632.09' Multiple Correlation Coefficient (R) .90' Coefficient of Multiple Determination (R2) .81 Standard Error of Estimate 273.92 ConStant Term - 24.70 Number of Observations 315 Special Generators (None) 0 Unit of Analysis Zone The Equation: Y = .3626 (X1) - 2h.7064 Simple Independent Regression T mean Correlation Variable Coefficient Value Value With Y (X1) Total Employment .3626 36 1571.75h1 .90 95 Npcmezoamsm Hmpoev Em mQONpomaPP< xaoz UmmemEom «poam Hmswfimom ma mnzmwm In 42ml 00.0000 00.0Nmo 00.0mmn 00.0000 00.00N. 00.000" 00.00NN 00.0N0N 00.00nN 00.000 00.0 NOOIIIIIIIIOIIIIOIIIOOII.IIOOOUOIOIIOICIIOIIIIIIIIIOIIUIIIIIIOOIOIIOIIIOUIIIIIIIIQIODIOCIDOOCIOIIIIOIQw N. oo.ooomn 0N . oc.oooNu C O J . . 00.000. 4 . 0 a 9 .. I O... H 5 PI. W O C a... w a c a n a c C C o . 00.000 . 00.00NN Hl-‘hOV-OHHhat-.HHO'QHHHU-‘HHO-tNHHO-‘HFOHO-ob-ot-OIOO-‘HHFiD-OFQHDJHHHO-OHHHD‘INF‘CH'W - oo.ooon “HHHHHNNHFINHHHHHHFCMHHHHMl—OHP4§OHHDQO4HF£HHNHHHHHHDQPCFCHHHHHH OOIIIIOIOIOIIIII.OIIOCIIIOIUUIOIOIIIIIIIOIIIIIIIIOOIIIOIIIOOQIIIIOOIIIOOOIIIIDIIOIOIIOIIIICIIIIOOIII9w 96 Table 18 Regression Summary - Homebased Work Attraction PM (HBWAZh) Dependent Variable (Y): Homebased Work Attractions PM Mean Value (Y) 105.27 Standard Deviation (Y) 632.09 Multiple Correlation Coefficient (R) .91 Coefficient of Multiple Determination (R2) .83 Standard Error of Estimate 260.31 Constant Term - 02.01 Number of Observations 315 Special Generators (None) 0 Unit of Analysis Zone The Equation: Y = .2530 (XI) - 02.4167 Simple Independent Regression T Mean Correlation Variable Coefficient Value Value With Y (X1) Homebased Work Attractions 20 Hr. .2530 39 741.79 .91 97' Adm<3mmv Em mCONPomHPp< x903 UmmemEom Npoam Hmscwmmm NH musmwm at 1101 00.0000 oo.ono oo.o~nn 00.0noc oo.och oo.om¢n oo.oo~n 00.050“ oo.onnN 00.000 00.0 NOIIIOIIIOOOIIIIIIIIIO0IIIIIIIIOIIIIIIOOIOIIIIIIIIIOIIIOIIIIIOIOIOIIIIIQIIIIIOIII§IIIIIIIIIOIIIIIIIIION N N. 00.000N. N CH N N N N N N N N N N N N N N N N N N. 00.000N. N N o N N N N N N N N N N N . N N . N N N a N N. 00.000! 4 N N a N a on. N n. N a I ou-N N N c cuocN w a C IQCDIM m N c c o c N u. N u N N. N N N N N. 00.00. N N N c N N N N N N c N N N N N N N N N N N. 00.0NwN N N N N N N N N N N N N H C N N N N N N N: oo.ocom N0.00.8008.6808-.....OCIIIIOOCUOUIUIIOIOCO|IIOI|UOI9.8.8.0...QICOIIIOIOQUOIIIIOOIOIIOCOO...0.0.8.8...9N 98 comprise well over 90 percent of the total hourly trips for the period. This overall lower number of total work trips (30,000 for AM to 45, 000 for PM) caused some calibration problems based on less data points to compare. Better results would have been achieved by use of more than one hour period for projection purposes . (1) Homebased Work Productions AM Once again, analysis began with a review of the correlation matrix to determine the most important variables to test the model. Examination showed much lower correlation coefficients than were available to the PM calibration. Two models were attempted with one based on 24-hour homebased work pro- ductions and the other using the variable of autos available. It would have been preferable to the researcher to use the population variable because relatively more confidence can be assigned to the projection, but the compari- son of the two models was too different to allow its use. This model is sub- ject to the problems inherent in the use of automobile ownership figures as auto ownership rises faster than population, plus the additional concerns re- lated to energy conservation. The average number of work trips generated at the residential end is -19 per every auto owned in the zone. About . 06 trips for work are generated for each daily work trip taken in the area. Both of the coefficients reflect the reduction of the importance of the work trips during this person peak (See Chapter III, Section 3). Tables 19 and 20 and Figures 18 and 19 review the statistical reliability and predictability of the models. Both are within 99 Table 19 Regression Summary - Homebased Work Production AM . (Auto Available) Dependent Variable (Y): Homebased Work Production AM Mean Value (Y) 97.27 Standard Deviation (Y) 98.19 Multiple Correlation Coefficient (R) .81 Coefficient of Multiple Determination (R2) .67 Standard Error of Estimate 56.00 Constant Term - 2.35 Number of Observations 315 Special Generators (None) 0 Unit of Analysis Zone The Equation: Y I .1935 (X1) - 2.3580 Simple Independent Regression T Mean Correlation Variable Coefficient Value Value With Y (X1) Auto Available .1935 25 510.99 .81 100 N.NQMNNm>< mopsMmmmo Auummo A4909 oo.ooo oo.o~o oo.o~s oo.o~c oo.cqm oo.omq ocjcom 00.05“ oo.oou 00.90 03.0 H9.0.0....I9I'll.IIII’CUICII--.OIUIICI.I'.‘II--.I|IOIIUUIU'CIQIIUIIIII-QUIIIUIIIIOUUOIIIIIIO'CII'III'OH "I .UCoOUmAI p I ou.ococ ».Hw.4r-rot—b—HH-—o H'- “WV-OD».- 4 no ac.cnmu 4 H 4 n u p M w m £OCCCCIch U taoccoa H u accccctc m «case-ac H cocoa. w a to. I ogoovm CO I uC.Cuo ‘HHHD-OFOHWHHHHHHHHHO‘Fin-pr‘wwhwp.”._....~fi,_....HN.‘hH~—H._~HN u unoccmn b-oHu-QO-oHO-Ib-D—O-‘HHH01bib-bahHD—cb-Ir-IO- OI.|.IIIII§III-III|‘§||IIIIUI|+.IIIIUIUUOCI'IUIIIUOOI0|I|O--OI.IIOIIOI+IIIIUIIIOOIOIIIIIIUQIIIIUIIIIf ~HHH. 118 This concludes the development of the generation phase of the model- ing process. The next step will lead to a discussion of the distribution func- tion within the process. Several generation models have been developed for the AM and PM work trip. These models are felt to be adequate for use by the planner to develOp a total modeling capability (with reservations). In the development of the total model to test overall model sensitivity Homebased Work Production and Attraction and Matrix models for the PM period will be used. CHAPTER V DISTRIBUTION OF PEAK PERIOD TRIPS COMPARED TO 24-HOUR TRIPS Peak Hour Distribution Comparison 24-Hour, 24-Hour Gravity Model, AM and PM Peak 1. Introduction This chapter is intended to determine whether it is necessary to cali- brate a new set of gravity model "F" factors (indicating relative attractive- ness) for each peak period and analyze the purpose travel distribution characteristics for trips for the peak hour period. The thrust of the anal- ysis is a comparison of AM, PM, 24-Hour, and calibrated 24-hour gravity model mean trip length frequency distributions, means, standard deviations, and the relative location of trip length peaks. This type of descriptive ex- amination should reveal any major differences that are attributable to differ- ent time periods. If significant problems are discovered, they can be traced to two possible sources: (1) Lack of a peak period calibrated network (i. e. , peak period travel Speeds), or (2) actual differences in driving habits. All purposes are discussed, but this study will concentrate on the work purpose to demonstrate the calibration process (Figure 10, Chapter II). 119 120 2. Actual Skim Trees A study was conducted to develop travel time from the actual O-D survey data; i. e. , the times that persons said they Spend in traveling. A comparison of trip length distributions from this study to the calibrated 24- hour minimum path trees (skim trees) reveal very little difference. The "skim trees" are the cumulative travel time between zones and for this the- sis are based on 24—hour calibrated network speeds. Obviously, Speeds vary throughout the day and this causes problems in any attempt to utilize 24-hour network impedances. The positive findings of this small study were only for a 24-hour comparison. It is recommended that the practitioner not pursue this type of anal- ysis because of the positive results of this 24-hour comparison study and due to the amount of time and programming involved. It is most likely that at some future date, the data may well demand a method of calibrating a peak period highway network using peak period Speeds and volumes. This study has not attempted to calibrate a peak network but rather has chosen to devel- 0p as good a model as the available data will permit. 3. Gravity Model The trip distribution model is a major component of the total plan- ning process. The most widely accepted distribution formula is the "gravity model". The model is based on Newton's law of gravity which states that the relative attraction between two objects is directly proportional to the 121 relative size or mass of each object and inversely proportional to some func- tion (square of distance) of the physical separation between the objects. The model's mathematical form as used in trip distribution is as follows :41 Tij = Pi Aj Fij Kij IgAj Fij Kij Tij = Trips produced in zone i and attracted to zone j Pi = Trips produced by zone i Aj = Trips attracted by zone j Fij = Emperically derived travel time factor (Friction Factor) which expresses the average area-wide effect of Spatial separation on trip interchanges between zones which are T ij apart. Kij = A Specific zone to zone adjustment factor to allow for incorporation of the effect on travel patterns of definite social or economic linkage not otherwise accounted for in the gravity model formulation. This formula reflects awareness that distance does not manifest a uniform effect upon attractiveness, but varies by purpose. In this study, "over the road" as opposed to "airline" distance is used to determine travel time (i. e. , distance). Basic conclusions about the gravity model formulation are listed below: 1. Spatial separation between zones appears to be best measured by "over the road" driving time between zones, plus some measure of terminal times in the zones at each end of the trip. 2. The numerical value of the exponent of travel time (F-Factor) is not constant for all intervals 4'1United States Department of TranSportation, Calibrating and Testinga GravithModel for Any Size Urban Area, Federal Highway Admin- istration, Bureau of Public Roads (Washington, D.C. 1968), p. I-2. 122 of time within each trip purpose. For most trip purposes, the exponent generally in- creases as the time interval decreases. 3. The exponent of travel time (F-Factor) alone does not, when considered in relationship to the use of land, completely explain the pro- pensity for travel between two zones. Travel patterns can also be affected by various social and economic characteristics of zones which, to date, have not been completely identified or quantified. Hence, factors other than travel time must be taken into account. These secondary adjustments are called "K" factors. 42 4. Peak Period Distribution Having examined the gravity model theory, it is now appropriate to examine the results of comparing 24-hour calibrated gravity model distribu- tions to peak period distributions. Table 28 shows comparisons of all trip length frequency distributions based on the standard deviations and mean travel times. A quick glance reveals the similarity of values of these mea- sures between the gravity model and peak hour distribution for the purpose tested. The peak period analysis is restricted by the purposes chosen for the 24—hour mode. A particular purpose, as has been discussed earlier, is not always a basic ingredient of the particular peak period under analysis, 1. e. , school trips. Often it may be necessary, if the gravity model is used as a part of the direct demand estimation process, to calibrate at a more gen- eral purpose category level. It is apparent from review of Table 29 that 42mm. , p. 114. 123 many of the purposes in the peak hour do not contain adequate observations by themselves to allow a reasonable comparison. The analysis, which follows, examines each purpose utilized for the Flint 24-hour modeling effort as it relates to peak AM and PM hour activi- ties. (Special emphasis will be attached to analysis of work trips.) Bear in mind that it is not imperative that all of the models be acceptable for use in the peak period model. Those models that are useful can be retained and a different approach can be taken for the rest. While factoring of 24-hour trip tables, based on a really poor distribution comparison could produce equally poor results, it should be understood that, relatively Speaking, in the Flint- Genesee County study (Table 28), the comparison (criticisms notwithstanding), are not unacceptable. In testing the compatibility of the calibrated gravity model of 24-hour models with peak period distributions, two analyses must be performed: 1. Comparison of the 24-hour O-D distribution to the actual peak period distribution. 2. Comparison of the gravity model peak period synthesis to the actual peak period. The first comparison allows a decision as to whether the calibrated 24-hour distribution is even relevant for a particular peak period. The se- cond determines whether the calibrated "F" factors, when applied to peak period trip ends, can reproduce the actual peak period distributions. If the first test is successful, the second should also be significant. Tables 28 and 29 examine the percentage that peak period travel com- prises of 24-hour travel by purpose; and relates the comparison of 24-hour 124 oaa.a cud.” 0H:.H om~.H omo.a oHH.H ooa. 0mm. cam. 0mm. oao.a 0mm. oao.H com. xwom 20 £4 gamm 2< k o~5.m~ omm.m -N.0H mmm.m omn.~a mom.“ ~mm.oa omm.¢ amm.HH omm.o mmH.HH 000.0 nam.ma 003.0 mem 20 2< ova. omo.a muo.H omo.H omo.a 0:0.H mam. mom. mmo.a NNH.H omo.a mam. moo.H omo. moa.m~ Hum.oa don.:a mwm.w Nom.NH mmw.m mda.oa mdm.d :om.HH 0:0.0 Nom.m omm.m oom.ma mm¢.w xdom 20 2m mem 20 xMom 2m m Em cam. owo.a omm. 0:0.H mmo. 0N0. mmo.a OHN.H ONH.H 0mm. mm. 0mm. mm. 00m. mmm.mN wno.aa H0¢.¢H mow.w wam.ma mmm.m oom.a mmm.¢ :Nm.oa dNN.© mom.HH mo~.o www.ma mom.w mem E< xmmm E< .p: 3N a mom. mmo.H 30m. owm. mmm. Hm. owo.H m:a.a owm. cum. 0mm. cum. 0mm. moo.a swam 2m .nfium E mcofipsnfiupman spmcma gape Hmnoz Spfi>muc am can deco: Spfi>auu :< .:m .z< .usom Sm co consumaeoo mm oHnma «no.0N Hom.HH Hmm.ma oma.m 0H0.NH oam.m oao.m om:.# onm.~a omo.n oom.oa omm.m mmm.ma omn.m xMom 2m moo.mm Smw.aa Ham.ea 000.0 oem.HH oun.m oo~.oa oma.m omo.~H oma.w ooo.oa oom.m omN.mH onm.m .pm Sm cams .n.m coeuoo .Q.m cams .a.m xosna coma .a.m mxz cams .o.m amgpo Gama .oomu.oom cmmz .a.m nonm coma .m.m xhoz mowed omaw ads: mmamn couuoo nmwum nanwa Nanaa noHdHH xosue moon condos mooa momomw Reed amused Reed mmNNmSH H6908 Hmcuoch N.NH adama 0.5H mason ~.m maww a.ma mmmnmw mmz a.mm mnoswu «.4: massed n.~u waamna o.on Hemoma Hosvo aN canoe 0.: omnwa o.w wanda d.~ Nnoa o.Nd mmnmna coapmeuoom Hmaoom S.m mmmnn ~.nH Osman m.” mNNN w.H~ owmmon mcanqonw owmpcaouom Ho>uua .zm .£< .usom aw uo conduSQEoo 0.0H woaou n.0a nwnma m.ma namon n.wH ownmmN xuoz dance no R xmom Sm R S< Haven Haves co m xmom Em oozouoomo Haves co m xmum a< onmo-onao Haves co m Hmpoa unox SN 126 O-D to peak period O-D to peak period gravity model. These tables, plus the figures reflecting the present frequency of trips by work purpose, portray the results of the study of travel times and distribution of trips for the Flint area during the peak periods. The study next examines the work purpose comparison and comments on its acceptability as a distribution instrument. Comment will also be offered for non-work purposes. A. Work Comparisons Table 28 reveals a good comparison of standard deviation and mean trip length between 24-hour, AM and PM peaks and especially the artificial gravity model Situation. (All are within 1. 3 percent.) The actual number of trips is significant in supporting the results. Analysis of graphic compari- sons (Figures 24 and 25) show that the basic test of similar peaking and visual fit are well achieved for both AM and PM peak origin-destination trip length distribution to the 24-hour distribution. These results are as ex- pected. Work trips are the least flexible of all trips, being literally a re- quired trip for the person. Calibration of 24-hour curves for work usually reveals a relatively flat curve (F-Factor) in comparison to other purposes. Figure 23 Shows the final calibrated F-Factor curves for internal homebase models for the 24-hour Flint study.43 As can be seen, the curve for work 43Michigan Department of State Highways and Transportation, Bureau of T ranSportation Planning, Trip Distribution Report, Flint-Genesee County Transportation Study (1972), pp. 51-57. 127 100000 10000 FRICTION FACTORS i000 Figure 23 Flint F-Factor Curves MINUTES ' FoFACTOR CURVES FOR VII/Ia WORK - - - - SHOPPING uummm SOCIAL RECREATIONAL —— OTHER SEMI-LOG FUNCTION 50 _ 60 128 ,_II “I -‘—.W-—u—Q.- = ‘ E 0. . E “p := . 3 $9 :4 a: < 3 § u: <3 4. IL I g " d” 35 en ‘3‘. t“ atttttIItItost “ II‘ 0“ . H .{ it. I I!!“"‘ III!"" GIIIII"" ' \S“‘ ""'~ “ §‘ - 1 '0' 0'" ‘3 I I". It!!! i call. ; llllllll ’ ~ I It: . H .. cansannncauauallql . - I «a «a " a” c‘ '- 39V1N3333d Figure 24 24 Hour O-D Work Distribution versus AM Peak Period Work Distribution Flint Study Area 1966 PM PEAK quuu HII‘IIIIIII‘IIIII V’ll...l' 24 HOUR III-— we “‘ v 39V1N3383d 40 35 fi ”.75 -,_..__-.4-._2_§.-_._-._-_ MINUTES 10d Work Distribution PM Peak Per 1011 versus tribut is Flint Study Area 1966 25 24 Hour O-D Work D Figure 130 coma won< Seapm ecaam coapsnfiupmfln s< Hmeoz Apn>mue mamum> soapsnanpmnn sue: coanmm xmmm z< om shaman 3.52.: E E m oooooo u . [In-i ¥mnu mamnm> consannnvmna xnoz eofinmm seem 2m SN magmas mmhazi o— in ooo ’ 0: C ' C .l O C C C C C O 0: at C C u: ...- lulu-3032.. . ‘O BOVINZDUBJ =3:=== 20 In. . Illllllll “I an: 132 is considerably flatter than the other purposes, indicating that the effect of time on travel is more constant. This assists in the conclusion that work "F" factors for the Flint 241-hour study can be used both in the AM and PM to dis- tribute trips and/or that a factored 24-hour matrix will reflect well the peak period distribution characteristics. Figures 28 and 29 show the reSpective AM and PM peak origin- destination trip length distributions to AM and PM gravity model distribution. The great similarity in both figures adds further to the conclusion that the calibrated gravity model for work in Flint is appropriate for the peak peri- ods. All of these comparisons were attempted for each purpose to verify the acceptability of those distribution models. General discussion will be presented for the non-work purposes in the remainder of this chapter based on the same tests as were used for the work purpose case study, however, since work will be tested in the total model evaluation, no graphic compari- sons for the non-work ptn'pose are included. B. Non-Work Comparison (1) Shopping Examination of Table 28 reveals that 24-hour and PM peak shopping trips are very similar in their average trip length; however, AM peak trips appear to be almost 1/2 a minute longer. This Shifting of the curve is re- flected in the poor comparison between standard deviations and means of 83 percent and of 88 percent, respectively. This discrepancy is partially due 133 to the small number of AM data points (not much showing in the morning, (7:30-8:30). Shopping trips account for 10.1 percent of total shapping trips in the PM, but only 1. 3 percent in the AM and in real numerical terms, the PM period has almost 30, 000 more trips (Table 29). Review of peaking charac- teristics Shows a good fit between 24-hour distribution and the PM peak dis- tribution, but the AM period requires recalibration. (Curve shifted to right.) The recommendation for the model, if direct demand procedures were to be used for the AM period, is to combine AM peak with other purposes and the recalibration of a combination purpose model. (2) Social-Recreational Analysis of Table 28 Shows results similar to shopping for this pur- pose. The PM period mean comparison is acceptable, but slightly higher. The AM peak shows the reverse. This indicates longer trips for the purpose in the AM and less peaked travel in the PM. Both the AM and PM periods together account for only a little more than 10 percent of all social- recreational trips in the area (Table 29). The small size of the sample suggests that this purpose should be combined with other non-work purposes and recalibrated for the Flint area peak periods. (3) Other The "other" purpose combines the purposes of personal business, eat meal, school, medical-dental, change mode, and serve passenger. Pre- vious discussions concerning the preponderance of school trips in the peak 134 hour "other" category (over 50 percent of daily total) have been put forward. A comparison of trip length indicates that both AM and PM periods have con- siderably shorter trip means than the 24-hour period. This probably is due to the extreme concentration of school trips during these periods. School travel is of a fixed distance and normally would contribute to Shorter trip lengths, and less variability in friction factors. This purpose is different enough from the 24-hour mix to require complete recalibration. Based on previous discussions, the AM period shopping and social-recreation purposes should be added to the purpose. For the PM period, only social-recreation should be merged. (4) Non-Homebased (NHB) Non-homebased trips are all trips that have neither end of the trip terminating at the residence of the tripmaker. In this category, all pur- poses are included. During the AM peak, many of these trips will have one end at the work place, while during the PM, a much stronger mix of linked trips can be anticipated. Linked trips are, for example, those which re- flect moving from work to Shopping, to social recreation, to home. Approx- imately 25 percent of all NHB trips occur during the two peak hour periods. The afternoon peak reflecting a far greater percentage (over 15 percent) be- cause of the better mix of purposes on the road based on the availability of services (Table 29). This is reasonable since these are linked trips and most trips are assumed to eventually return to the location of residence in a 24-hour period. 135 The mix of trips is fairly constant, while the trip length is almost one minute higher for 24-hours. When the 24-hour "F's" were applied to PM peak hour trips, the distribution was not as good a match visually, nor did the peak minute periods correSpond. It is strongly urged that, if the direct demand estimation process is applied, recalibration be accomplished. (5) Truck The mean times in the PM period are Shown to differ greatly (over one minute), showing that during the PM peak, the trips by truck are of a much longer duration. When the 24-hour "F" factors are applied, an even poorer match is achieved. Similar results are noted for the AM period. The elongation of the trip length frequency distribution compared to the 24- hour distribution results from the fact that many 24-hour truck trips are of short duration because of delivery schedules. Earlier in the day and in the afternoon, the trucks are returning to and coming from their terminal areas forcing longer trips. This purpose must be recalibrated. (6) Cordon Cordon trips contain all trips with one terminus outside the study area (crossing the cordon line). This purpose also reflects very closely the 24-hour distribution. The 24-hour total for this purpose is 35,155 trips (21, 881 trips for the purpose of work--62 percent, and 13, 274 for non-work 136 activities-38 percent).44 Cordon trips comprise a small percentage of the total area-wide trips (1. 4 percent work and . 8 percent non-work). During the PM peak, 6,190 trips were observed for the cordon purpose and 4, 719 for the AM period (Table 28). These totals will obviously reflect many work trips which are always made and would be reflected in the 24-hour distribu- tion. The mean trip length frequency of cordon (Table 28) shows a similarity to 24-hour distribution even though the peak periods have only a few observa- tions. (The PM peak is 18 percent of the 24-hour total and the AM peak is 13 percent of the total.) The 24-hour gravity model is appropriate for this purpose. 5. Conclusion The AM and PM peak hours in the Flint-Genesee County study area during which the most person trips are in motion, reflect for the most part, a very similar distribution to the 24-hour period. It cannot be assumed that this will hold true in all areas. This warning should eSpecially be heeded in cities of much greater population and physical area (i.e. , Detroit, etc. ). Because the longest possible trip in the area is only 60 minutes, it is possi- ble to maintain a very tight reign over any great variation. The major con- straint in any area is the basic land use configuration and street system. As long as they remain fairly constant, travel destinations will not be altered 44Michigan Department of State Highways and Transportation, Bureau of T ranSportation Flaming, Trip Generation Report, p. 22. 137 greatly (most of the industrial and commercial districts are in place and will continue to be active centers for the foreseeable future). Any major changes would require reevaluation of the models. It is recommended that the AM and PM peak work models, and the cordon gravity model and the PM peak Shopping ("F" factors) be the only ones used, without reservation, to develop projective models by distributing new trip ends. The others should be recalibrated with a new "other" category developed for all homebased trips except work. It Should be noted, however, that the basic fits are "good enough" to allow for the factoring of a 24-hour purpose matrix for these purposes, as the worst comparison is about 18 percent, and most are no more than 10 percent. The planner can now begin to determine the amount of reliability by which the model when applied can be judged. It is important to know which distribution models will not work before attempting to calibrate future trip ends by regression. Even if good equations were possible, the trip ends could not be distributed. Likewise, one should not start to calibrate com- bined gravity model purposes until it is proven that good regression equa- tions are possible. A good deal of effort and money can be saved in this fashion. CHAPTER VI EVALUATION OF TOTAL MODELS 1. Introduction Previous to this chapter, the Flint case study has emphasized the development and calibration of individually acceptable models that minimize, as much as possible, the error inherent in the prediction process. Under- standing of the basic weakness in individual models can be useful in the eval- uation of the effectiveness of the total models. The total model, as general- ly referred to, is the combination of the calibrated network, the distribution, and generation models. The usual form of the total model is a traffic assignment. Normally, this assignment would include all travel purposes and modes within a region. For the description in this chapter, however, only the work purpose for the PM will be examined. The need to restrict the sc0pe of the analysis (concentrating on the work purpose) makes this limita- tion necessary. Analysis of the AM period is not included. The PM period was chosen over the AM period because of the stronger generation models developed (see Chapter IV). The models used are the synthetic model with p0pu1ation and total employment as the independent variables and the 35 per— cent production, 65 percent attraction split factored matrix table. 138 139 2 . Matrix Comparison Tools One of the key elements of a total analysis would be the use of the matrix and trip end comparison program discussed earlier in the study. This has already been accomplished in the calibration of the matrix models and results are included in that particular analysis (Chapter IV). Problems with effectively utilizing the analysis procedure were discussed during calibration of the matrix models. These problems were the low mean for each interchange cell and the many zero interchanges. For this reason, it has been dropped at this time for total analysis. However, if all purposes were summed and then tested, it is quite possible that the comparison at the cellular level would be beneficial in making decisions as to model effectiveness. The pro- gram that would be applied for this analysis is "TP Comp". Trip end comparisons are useful and should be examined for each purpose as well as for the total model. Because of the limited scope of the case study (work purpose), this has already been accomplished for the matrix model during the calibration phase of development. The comparison is likewise implicit in the development of the synthetic generation model equa- tion. Each regression or rate analysis is tied to the comparison of observed trip ends with the estimated trip ends. The decision as to the best model is partially based on this procedure. The residual plot analysis results are discussed in Chapter IV. If a total model was being tested instead of a single purpose model, the com- parison would require residual analysis to test for predictability. By accep- tance of the work models (matrix and synthetic) the trip end analysis is 140 completed and the model is assumed to be predicting adequately. In a com- parison of the matrix and synthetic models, the synthetic appears to be the better predictor, but both models do indicate tendencies to overpredict at the high volumes. The shape of the plot would indicate that another variable should be added to increase predictability (Appendix C). 3. Evaluation of Assignment As indicated during the Introduction, the key to usefulness of the model is how good a job it is doing at the network link level (i. e. , Does it reproduce the actual travel pattern on the network ?). The total model must be tested to compare actual assigned volumes to predicted volumes on a link- by-link basis. The program "TP Eval" is utilized for this purpose. The program outputs comparisons based on volume groupings established by the planner. Outputs of the program also include actual versus residual normal deviate plots of these volume groups. (A normal deviate is a standardized residual or the residual/standard error of estimate.) These plots indicate the relative predictability of the total model for different link volumes. This is most important if the assigned volumes are to be used for project level decisions. Figures 28 and 29 Show only the comparison of all link volumes. Individual groups can also be plotted. Review of Figures 28 and 29 indicate that the synthetic model is doing a much better job of predicting link volumes than the matrix model. As indicated earlier, it appears that both models are underpredicting traffic on the higher volume links and overpredicting on the lower volumes. It would 141 Hmcoz xfiupmz mumpw> Sm Duo pom meESHo> xsfiq Hepoe “SOHm Hmsvflmmm mm mpsmflm wDN an chzupmu no zoxaw oz¢az unbuczp noo.ooon ooo.n~nc ouo.on~n oooonaan coo-com“ ooo.muou ooo.om~« ooo.n~o 000.3 r.ooooooooXooooooooeVoouooooooxoococoooofiooooooooOXooooo0.0.1.cocooooofloooooooooxx x o o o o o x O .1. c on e. I c c u o O C C o a ma ac. magma. c at; xaax. . .m mxa.aaa. a on C L minus. c a c c .xaamlxx c a Laaxaaanx. a a a . ..uaatlmauum. . «cm a marginal. a c a mixmaiimnana. c c .. caaaxm.rlarnn1 o m a can lamnimnaa . . a a c cum .- axmamxxmtm . a cc acclaimxmxxxa . 9 m mxmxmnmLLLLx . c . a..a . mxaxmialmi. x ac.a11iox main. . c c a. a aura; a . a. aaa.m ma . - mmaamcoax . a c mam-c no . c m on. o. . CC CC. CC. m o I. a. a. o O C I. o O C O. K II C o c a c c . 1 C o C C o O C o a m o C C C o C C u C C. 0 030.0! cocoa! oaw.~v can." coo.n oaooo fi;§§-¢d-hl 133 Y3’4 M2 Hmcoz oapmcpC>m mSmSo> Duo pom mmesao> xcfiq Hepos «Poam Hmsuflmmm mm mhswflm gnu as up‘xuhmu no zorau arcaz N HN N mm o.“ om. SH.H on. N>m. N m: H as N a: a.m 00.N No.H 00.N Hmm. H m: N as N as N.N 00.N ON.H oo.H Nam. N mm H o> H mm oa.m.o em> 92> >m> 92> em> 92> em> 92> em> 92> mmeao Nana u e owpmgpc>m xNepms 9-0 xmmm ouo xmmm Qno xmmm ospwg>c>m xflepmz Amocmmsonp :Hv pcmscmfimmd ofipmnpszm new .xfispms nmuovowm .Quo xmmm an em> ncm 92> mmomhsm xpoz exp 90% xmmm 2m amcomfiummsoo mmmao xsflu Hm manna 148 oo.H MH.H oo.H Nmm. H HdH H oHH H mNH .oo sameneo 00.H 00. 00.H 020. N 90H 9 00H 9 00H >9H0 chHm 00.H 00.H 00.H 0:0. 0 000 0 0>N 0 nmN H3209 m0.H mNN. 0 0N 0 NH 0 0H mH 00.H H20. 0 0H 0 0H 0 9H SH HH.H >20. 0 HN 0 0H 0 0H HH 00.H mm.H 0 m 0 0 0 0 NH 00.H 00.H 0 NH 0 HH 0 HH HH 9N.H 00.H 0 0H 0 0 0 0 0H 0H.H mum. 0 0H 0 SH 0 0H 0 mm.H 00.H 0 a 0 m 0 n 0 00.H SH.H 00.H 000. H mm H 0N H 0N N 00.H mo.H 00.H 00.H H HN H 0N H 0N 0 00.H 0m. 00.H 0mm. H mN H mN H 0N m 00.H 00.H 00.H N20. H 0H H 0H H 0H 0 00.H 00. 00.H mHm. H NN H HN H mN m 00.H 00.H 00.H 000. H NN H 0N H NN N 00.H 00.H 00.H 0 m 0 m 0 m H 00.H 00. 00.H 000. N mm N Nm N mm 0 92> 92> 92> 92> 92> 92> 92> 92> 92> 92> cOHHOHemHese 0-0 2mm» 0-0 23.9 099029220 2H2902 0-0 2309 OHNoepc>m xH29m2 “002005029 :Hv pamscmem< OHHonvczm 0:0..xthmz 00909002 .030 xmmm hp 92> a 92> mmomssm 2903 029 909 xwom Em mcomemmsoo :oHpovaHusn Nm erwB 149 the factored matrix does a slightly better job overall. Basically, the two models did an excellent prediction job. A final test is the comparison of links crossing screenlines in the area. These screenlines are set up to intercept travel to and from rural and urban areas, into and out of the CBD, and along major traffic corridors. This is the finest level of analysis and is used to check how finely tuned the model is. For the model under analysis, it was decided to examine flow into and out of the city and the CBD. This is logical due to the central loca- tion of Flint in the county and the location of most industry in the city. Table 33 shows total comparisons along the screenlines developed. Table 33 Screenline Percentage Comparison ‘70 '70 PM Matrix Synthetic Screenlines PM-OD Matrix Synthetic PM-OD PM-OD 1 West City Limits 10, 553 9, 343 10, 621 . 89 l. 01 2 East City Limits 4, 735 4, 647 4, 666 . 98 . 99 3 South City Limits 3,159 2, 966 2, 576 . 94 . 81 4 North City Limits 4, 986 4, 427 4, 915 . 88 . 99 Total City Limits 23, 433 21, 413 22, 778 . 91 . 97 5 North CBD 3, 841 4,117 4, 391 l. 07 1.14 6 South CBD 2,146 2,705 2,724 1.26 1.27 7 East CBD 2, 202 2, 758 3, 283 l. 25 1. 49 8 West CBD 4, 962 3, 968 4, 390 . 80 . 88 Total CBD 13,151 13, 548 14, 778 l. 03 1.12 tion for both models. The screenlines comparisons indicated acceptable levels of predic- Both models overpredicted on the smaller volume 150 screenlines while the synthetic model predicted much better on the larger volume locations. Once again, both models are very acceptable. 4. Total Model Applicability Analysis of the models available resulted in the basic acceptance of the various models. If a total model was being tested, other tools are also available to the planner. Assignment of travel by average travel length or by a particular travel tree would be possible. More extensive analysis of screenline crossings to determine geographical bias in the models would be recommended. It is possible to check direction of movement by examination of the assignment at various screenlines. All these are based on the parti- cular need the planner has for the prediction model and the amount of time available for its use. It is felt that this is the most vital stage of the analysis as it gives the planner the key knowledge as to what is the accuracy with which the data can be utilized. Too often the calibration of the "tools" overshadows the reason for their existence. The knowledge received from the total model analysis procedure puts the "tools" in perSpective and leads the planner to acknowledge the problem solving nature (not model proliferation) of the pro- fession. Some parting notes on this chapter are critical before moving on to Chapter VIII and the conclusion of this study. This case study has basically concentrated only on the purpose of work for analysis of the develOpmental process. Each purpose must be critically examined if a total model is to be 151 deveIOped for a region. This concentration was based on nothing more than the need to single-mindedly develop solutions to developmental problems presented by this thesis. This author feels that the analysis activities tested using work trips are applicable to other purposes and to the testing of a true prediction modeling package. It should also be emphasized lest the reader feels that this author has fallen into his own trap of calibration "over-kill", that this is a developmental attempt and as such should have as its under- lying focus the visualization of as many possible approaches to solving the developmental problems within the assumption of this thesis as possible. The process is such that hard decisions have to be made at each step of the process by the practitioner as to whether a "broad brush" tool or a "corridor model" is necessary; or whether person travel or vehicle travel is most important; what levels of total model accuracy are appropriate to your decisions, etc. This thesis offers various possibilities as to the prob— lems that may develop (this tests only one purpose in one urban area thor- oughly), and where to turn if such problems do develop. It does, however, Show that in Flint-Genesee County, Michigan, that the matrix model and the synthetic work model deveIOped by use of this procedure are acceptable planning tools for both systems and corridor level decisions. a-n-u-x “FAA ‘ _.... 5 We?" CHAPTER VII CONC LUSIO NS This study has developed and evaluated two peak period modeling procedures. The parameters of the study have required the process to be based on existing data and programming situations. The process was tested by develOping fully a work trip model for both modeling approaches, and comparing and contrasting the results to an observed peak period. A strong systems approach to solving the problems has been initiated for both pro- cesses. This process allows the planner to use his new tools with a know- ledge based on the expected accuracy and applicability of the models to the particular problem under analysis. Thus, it is not the calibration of the models, but methods of evaluating the reliability of those models that has been stressed by the process deveIOped. Methods have been provided then, to institute needed peak period planning tools and to judge the reliability of those tools. The models deveIOped by the process for the Flint-Genesee County area for the work travel purpose, during both morning and afternoon peak periods, produced acceptable results. It is recommended that the process be utilized for other travel purposes, and for different time frames. More Specific conclusions are addressed on the following pages. 152 153 1. Speed of Delivery While this study took over one year to develOp, the actual institution of the procedure can be a rapid process. Given the existence of a completed 24-hour process, develOpmental time should not exceed 40 to 60 man-hours. Analysis time, of course, would be reflective of the needs of the study and skills of the individual planner. 2. Availability of Data and Programming Capabilities This study has confirmed that enough basic data is available to do more than a reliable job of predicting peak period travel (certainly for work). Some problems could develop concerning the lack of observations within the aggregated zones, but these can be overcome. If it is possible, as new studies are taken, data more directly tied to the peak period should be col- lected. Enough programming capabilities presently exist within the resources of the Michigan Department of State Highways and Transportation to more than adequately handle any data manipulation problems that could develop. The process has been designed around that system. 3. Trip Generation This effort concentrated on the develOpment of a peak model based on zonal aggegate data for all persons in motion over a one hour period. The time frame was an arbitrary decision on the part of the researcher and was made to control the sampling process. Another arbitrary decision concerns 154 which trip purpose models should be deveIOped. During the peak hour a trip purpose, which is not critical in the 24-hour period, can become very impor- tant. If school trips, for example, are numerous, an additional model should be considered for these trips. Successful regression equations were dev- eloped for both AM and PM work periods. In general, these were not as good statistically as the 24-hour models developed. This was due to the re- duced number of observations in each aggregate zone. In general, the PM peak did a much better predictive job than the AM peak. The matrix models did a less conclusive job of predicting, both graphically and statistically, than the regression models. Both models (AM and PM) tended to underpredict the actual values. Just as the regres- sion model was troubled by a lack of trip ends, the matrix model was affect- ed by the low mean of the interchanges deveIOped, making statistical anal- ysis difficult. Overall, the matrix model did not predict as well as the synthetic model. This was especially true for the overprediction of lower volume and the underprediction of higher volume levels. 4. Trip Distribution The process utilized the gravity model developed for the 24-hour period to distribute the peak period work trips. In general, for both AM and PM hour periods, the comparisons were very good. In this case, the other purposes were also briefly reviewed. Indications from this analysis were that no recalibrations would be necessary, or at best, very little effort “fi-“T’FTT a .. W .- 155 would be required. It is doubtful that work or cordon trip purposes will ever require major alterations. 5. Matrix Model versus Synthetic Model The most critical decision for the planner is which of the peak period processes should be chosen in the develOpment of his tools. Comparison as to the assignment capabilities of the model are critical to the planner in this choice. This thesis only tests the work model instead of a true total model wacfimmtfi' (all purposes added together and assigned), but certain results are evident which should assist the planner in his initial choice of methods. Comparison between the processes is based on five areas: (1) Matrix analysis, (2) trip end analysis, (3) link volume group analysis, (4) link class and jurisdiction analysis, and (5) screenline analysis. In all areas, the synthetic model performed better than the factored matrix model. Generally, the results showed that both models tended to underpredict at the larger volumes. For very fine prediction, the synthetic model should be considered for development. It is vitally important for the planner to avail himself of the oppor- tunities for analysis that peak period investigation offers. As mentioned in the Introduction, many areas are Opening to the planner for use of this tool. These range from practical needs for design hour traffic percentage for facilities planning or environmental impact statements to 20—year sketch planning activities in the smaller communities. 156 Many areas of research are alluded to by this effort. A need to logi- cally convert peak periods average traffic assignment volumes to design hour volumes is required to do project level estimates. Likewise, much needs to be done in the areas of the development of synthetic origin-destination studies based only on peak work travel information. Calibration of peak period net- works and the development of means to utilize peak data for transit develop- ment are two very pressing issues. A critical developmental concern is the need for more and better prediction variables. This then has only laid out the means of getting started down the path to these activities. An excellent opportunity is offered to the planner for new anal- ysis tools by this study. It is the sincere hope of the author that this Oppor- tunity will not be missed. It isalso the sincere commitment of this author to actively attempt to educate and instruct all concerned in the use of this procedure. The door to fully understanding travel patterns over all periods of the day and how these patterns have affected and are affecting not only the physical structure of the area, but the very socio-economic conditions of the resident, has been Opened. It is the basic duty of those in the field to see that this potential opening for understanding, and the interaction of these variables, be realized. LIST OF REFERENCES LIST OF REFERENCES Abend, Norman, and Levan, Melvin R. W: m inAreaJranspmtatieni’lanning Cambridge Mass M I T Press, 1971. Afifi, A. A., and Azen, S. P. Statistical Analysis. A Computer Oriented Annmaab. New York and London: Academic Press, Inc. , 1972. Agnew, Malcolm and McCallam, Douglas. "Estimation of Peak Hour Flow on Urban Roads", Traffic Engineering apd Control. Vol. 14, No. 3, (July, 1972), 122-125. Armbrister, Carl Shelburne. WWW Travel Patterns. Thesis for Master of Science in Civil Engineer- ing, University of Texas at Austin, 1970. Betz, Mathew J. A Method of Analysis of Peak Hour Traffic Demand for Warn in Urban Areas. Engineer- ing Research Center, Arizona State University, Tempe, 1964. Blunden, W. R. The Land Use/Transport System Analjpsis and Synthesis. Oxford: Pergamor Press, Ltd., 1971. Burroughs Corporation. Burrgpghs Adyancegl Statistical Ingug’ y System. Detroit, 1969. Carll, Richard R. , and Hamburger, Wolfgang S. W Peak Period 'E'affic. Highway Research Board, Washington, D. C. , 1962. DeNeufuille, R. , and Stafford, J. Systems Analysis for Engineers and Maps.- gers. New York: McGraw-Hill Book Co. , 1971. Eberlein, Michael, pt a_l_. 'I_‘1;ave1 in Genesee County in 1966: The Impact of Land Use Upon the Transmortation System. Unpublished Report, Michigan Mpartment of State Highways and Transportation, 1975. Federal-Aid Highway Act of 1962. Section 9. 1962. 157 158 Federal-Aid Highway Act of 1970. 82 Stat. 1970. Garrett, Henny E. Elementary Statistics. New York: David McKay Company, Inc. , 1966. Genesee County Planning Commission. Genesee County Education Facilities, Conclusions Report. 1970. Highway Research Board. Urban Development Models, Special Report 97. Washington, D. C. , 1968. Highway Research Board. Urban Travel Demand Forecasting, Special Report 1113. Washington, D.C., 1973. Horn, John W. _e_t a}. The Examination and Comparison of Peak Hour Gravipty Models. North Carolina State University School of Engineering, 1965. Lamb, G. M. "Introduction to TranSportation Planning, Context of Trans- portation Planning", Traffic Engipneeripg and Control. Vol. 11, No. 9. (January, 1970), 422-425. Loewenstein, Louis K. The Location of Residences and Work Places in Urban Areas. New York: The Scarecrow Press, Inc. , 1965. Meyerowitz, Wayne. Documentation Manual: Statistical Analysis in Trans- portation Planning. Michigan Department of State Highways, 1974. Michigan Department of State Highways and TranSportation, Bureau of Trans- portation Planning. Accuracy Checks and Adjustment Fggtors: Flint-Genesee County Transportation Study. 1970. Michigan Department of State Highways and TranSportation, Bureau of Trans- portation Planning. Ann-Arbor Ypsilanti Factual Data Report. 1969. Michigan Department of State Highways and Transportation, Bureau of T rans- portation Flaming, Coding Instructions and Master Codes: Flin_t_: Genesee County Transportation Stud . 1966. Michigan Department of State Highways and Transportation, Bureau of TranS- portation Planning. Cross—Tabulations of Survey Data: Flint— Genesee County Transportation Study. 1970. Michigan Department of State Highways and TranSportation, Bureau of Trans- portation Planning. Grand Rapids Factual Data Repert. 1971 159 Michigan Department of State Highways and Transportation, Bureau of Trans- portation Planning. I.A. S. Summary of O-D Tract: Area Socio- Economic Characteristics; Flint-Genesee County Tranpportatiqn my. 1966. Michigan Department of State Highways and T ranSportation, Bureau of Trans- portation Planning. Midland Factual Data Report. 1971. Michigan Department of State Highways and TranSportation, Bureau of Trans- portation Planning. Muskegon Factual Data Report. 1972. Michigan Department of State Highways and TranSportation, Bureau of T rans- portation Plaming. Procedural Manual for Use in Preparation of Traffic Analysis Report. 1969. Michigan Demrtment of State Highways and TranSportation, Bureau of Trans— portation Plaming. Summary Report: 1966 Origin—Destinatigp StudyL Flint. 1972. Michigan Department of State Highways and TranSportation, Bureau of Trans- portation Plaming. Trip Distribution Report: Flint-Genesee County Transportation Study. 1972. Michigan Department of State Highways and TranSportation, Bureau of Trans- portation Plaming. Trip Generation Report: Flint-Genesee County Transportation Study. 1973. Michigan Department of State Highways and TranSportation, Pennsylvania Department of State Highways, Burroughs Corporation, and Alan M. Voorhees and Associates, Inc. A System of TranSportatipp Planning Programs for the Burroughs B-550j. 1969. Ockert, William, Easler, Richard, and Spielberg, Franklin L. An Analysis of Travel Peaking. Baltimore, Maryland Regional Plaming Coun- cil and Alan M. Voorhees and Associates, Inc. , 1971. Smerk, George M. Urban Transportation: The Federal Role. Bloomington: Indiana University Press, 1965. Stone, John Richard. An Analysis of Three Methods of Trip Generation for LSmall Urban Area. Unpublished thesis for the Degree of Master of Urban Planning, Michigan State University, 1972. Tittemore, Lawrence, _e_t pl. An Analysis of Urban Area Travel by Time of M. Peat, Marwick, Mitchell and Company, and United States Department of TranSportation, Federal Highway Administration, Office of Planning, Washington, D.C., 1972. 160 Tomazinis, Anthony R. , and Gabbour, Iskandor. Trip Length Variations Within Urban Areas. University of Pennsylvania, Institute for Environmental Studies, Philadelphia, Pa., 1966. TranSportation Research Board. Census Data and Urban TranSportation Planning, Special Report 145. Washington, D. C. , 1974. United States Department of Transportation. Highway Needs Report of 1972. Part II, Washington, D. C. , 1972. United States Department of Transportation, Federal Highway Administration, 5 Bureau of Public Roads. Calibrating and Testing a Gravity Model 7 for Any Size Area. Washington, D. C. , 1968. United States Department of Transportation, Federal Highway Administration. Traffic Assignment Manual. Washington, D. C. , 1973. United States Department of Transportation, Federal Highway Administration. Urban Transportation PlamingL General Information. Washington, D. C. , 1972. United States Department of Transportation, Federal Highway Administration, Office of Highway Plaming. Development of Design Volumes Usig Traffic Assignment Data. Washington, D. C. , 1972. United States Department of Transportation, Federal Highway Administration, Office of Highway Plaming. Guideunes for Trip Generation Anal- ys's. Washington, D.C., 1967. United States Department of Transportation, Federal Highway Administration, Office of Planning. Modal Split. Washington, D.C. , 1966. United States Department of Transportation, Federal Highway Administration, Office of Highway Plaming. "3lst. Planning Course". Washington, D.C., 1972. Wheeler, Philip Hampton. An Evaluation of Models for Modal Choice. The- sis for the Degree of Master of Urban Planning, Michigan State University, 1973. Wohl, Martin. "A Methodology for Forecasting Peak and Off-Peak Travel Volumes", mghway Research Record Number 332, Travel Anal- ysis. Washington, D. C. , 1970. Zevim, Isreal. Relationship of Peak Hour _Vo_lurpe§_and__Peak Ratespfjlpw to Peak Clock Volumes on Expressway Rampsp Turning Roadways, and City Streets. Connecticut Highway Department, Division of Planning, Traffic Plaming Section, 1969. APPENDICES APPENDIX A TRANSPORTATION PLANNING PROGRAM PACKAGE UTILITY PROGRAMS AND B.A.S.I.S. . -’- -’ w—aFeC— _ In? I. II. SUMMARY OF TP PROGRAMS TRAFFIC ASSIGNMENT PROGRAMS 1. 2. 3. 5. TRIP 1. 5. 7. TP NET, NETWORK BUILDER. Builds from cards or up- dates from tape a traffic assignment network. TP TREE, TREE BUILDER. Calculates minimum time paths from each traffic generating unit (zone, dis- trict, etc.) to all other units in the area via the traffic assignment network. TP LOAD, ASSIGNMENT. Assign trip table volumes to minimum time paths and accumulates resulting volumes on each link of the network. TP TURN, COLLECT ASSIGNMENT TURNS. Accumulates and prints out turning movement volumes around network intersections. TP SELECT, SELECTED LINK ANALYSIS. Creates a trip table of only those trips passing through a speci- fied network link in a traffic assignment. TABLE PROGRAMS TP TRIP, TRIP TABLE BUILDER. Builds zone-to-zone trip interchange matrices from O-D master tape based upon user specified stratifications (vehicle trips. P&A trips, trips by purpose, mode, etc.). TP MOD, TRIP TABLE MODIFIER. Multiplies, divides, adds to, subtracts from or replaces trip table en- tries with user Specified quantities or constants. TP MNIP, TRIP TABLE MANIPULATOR. Factors, combines, divides or modifies entire trip tables. Output is a single table with all factoring, Operations. scal— ing and rounding accomplished. TP FLOP, TRIP TABLE SPLITTER. Transposes, or changes the direction of trips in a matrix at the origin and destination ends by the use of directional split per- centages input by the user. Commonly used to go from productions and attractions to origins and destina- tions for assignment purposes in the forecast year. TP SQEZ, TRIP TABLE COMPRESSOR. Compresses to a lower level (districts, sectors) a zonal trip table by the use of user coded equivalency cards. TP PRIN, TRIP TABLE PRINTER. Prints out the inter- change values of a trip table of skim tree matrix with one zone per page. Selected ranges of zones are available. TP FMTR, TRIP TABLE FORMATTER. Prints out trip ta- ble entries from multiple tables in juxtaposition for ease of presentation. Selected ranges of zones and tables are also available. 161 III. IV. 9. TRIP 1. 7. 162 TP XPDR, TRIP TABLE EXPANDER. Using zone to district equivalence cards and district proportioning factors, it eXpands a district trip table to a zonal trip ta- ble. TP STAB, SUB-AREA TRIP TABLE GENERATOR. Generates a sub-area trip table for a section of the assign- ment network as specified by the user. DISTRIBUTION PROGRAMS TP GM, GRAVITY MODEL. Using a "gravity" concept, it distributes forecasted trip ends between zones, .creating a future year trip table. TP SKIM, SKIM TREE BUILDER. Outputs travel times between zones in matrix form to be used as input to the gravity model. TP TERM, AND TERMINAL TIMES. Adds additional times to the skim tree matrix to reflect non-driving por- tion of trips for use in the gravity model. TP TLD, TRIP LENGTH FREQUENCY DISTRIBUTION. Gener- ates graphs and statistics of frequency distributions, most commonly of trips volumes and travel times for use in calibrating the gravity model. TP FRAT, FRATAR EXPANSION. Expands a base year trip table into a future year trip table by iterating up to new trip end totals or growth factors provided by the user. TP PALM, POPULATION ALLOCATION MODEL. Allocates fu- ture population growth to zones based upon residen— tial characteristics and accessibilities on the transportation network. TP WST, BUILD WEIGHTED SKIM TREES. Reads skim trees, a trip table, and zone to district equivalence cards and creates district skim trees whose times are the average of travel times over the corresponding zones. ANALYSIS, PLOTTING, TABULATION PROGRAMS 1. 2. 3. TP VOLA, VOLUME WORD ADDER. Capable of adding new volume fields to a network tape or modifying eaist- ing fields. - TP TESM. PUNCH TRIP ENDS. Punches out on cards or writes on disk zonal trip end data (number of ori- gins and destinations) from trip tables for input to other programs. TP COMP, TRIP TABLE COMPARISON. Compares two trip tables on an interchange basis and outputs a statis- tical analysis. Commonly used to compare actual and predicted trip tables during the calibration of travel forecasting models. 7. 9. 10. 11. 12. 163 TP ACC, ACCESSIBILITY CALCULATOR. Calculates the accessibility of each zone to the other zones of the study area and punches or prints the resulting accessibility indices. TP CPA, CALCULATE PRODUCTIONS AND ATTRACTIONS. Calculates future production and attraction trip ends by the use of trip generating equations and future zonal socio-economic data. Output on cards and printer. TP NAPS, NETWORK ARITHMETIC PROGRAM. Performs sim— ple arithmetic on distance, time and volume fields of a network tape, outputing the results in new vol- ume fields. Commonly used to calculate volume to capacity ratios on tape for plotting. TP EVAL, NETWORK EVALUATION PROGRAM. Reads network tape and produces statistical summary of comparison of two volume fields. Summaries may be stratified by the use of control ranges. TP GPSP, GENERAL PURPOSE SUMMARY PROGRAM. Reads an O-D master tape and produces cross tabulations of data from user specified fields according to user specified criteria of selection or exclusion. Tab- ulations, in rows and columns with headings, are printed out. Up to 6-dimensions, or types of data, may appear in each run. TP MERG, GENERAL PURPOSE RECORD MERGE. Merges trip interchange data, skim tree data and zonal genera- tion variables onto a composite record for use in trip generation analysis. TP ADDR, GENERAL PURPOSE FIELD ADDER. Prints or punches summations of fields specified by user from O-D master tape or card image on tape. Control fields are available. Q01151, PRE-PLOT PROGRAM. Prepares a tape for the plotting program from a loaded network tape (i.e.. with volume fields filled). QOllSB, PLOT PROGRAM. Graphically plots networks and trees and annotates along the links either 1) the contents of up to six volume fields or 2) a bandwidth scaled to the volume field chosen. B.A.S.I.S. 1. 3. LOOK, Data Inspection Program. This program allows the user to display or print the values of selected variables from an input data set. STAT, Basic Statistics. This program computes the basic statistics for each variable named in an oper- ation. REGRESSION, Simple Linear Regression. This program computes the regression statistics on any number of x and y variable pairs. VI. 7. 164 Output are the regression equation, adjusted obser- vations, standard error, correlation coefficient and fraction of removed variance. PIOT, Data Plotting Program. This program gener- ates a set of (x, y) plots for each variable pair entered. CORRELATION, Linear Correlation Analysis. This program computes "Pearson product-moment"correla- tion coefficients among all entered variables. BUILD, Create a new Data File. This program allows the user to build a new data file on cards, tape or ' diSk 0 STEP R, Stepwise Multiple Regression. This program can be used to solve a sequence of multiple linear regression equations by a stepwise application of the Least Squares Method. UTILITY PROGRAMS 1. TAPE SELECT. This program allows the user to oper- ate on a basic data file and to create a new file based on the data fields selected. P&A OR O-D SORT. The program sorts the origin- destination ZOO Character Record into a destination within origin configuration or sorts the record by attracting zone within producing zone. A producing zone is the zone of origin or the home zone of a trip which either started or ended in the home ori- gin. APPENDDC B DATA FILE SUMMARIES ZOOCHARACTER RECORD MERGED PERSON RECORD COMBINED O&D FORMAT 200 CHARACTER RECORD Field Description Key-word City # Form # Residence Month Day Structure Cars at address Car Mileage Persons at address Persons over 5 Years at residence Own or rent Value of residence Education of household head Number of persons employed Income Occupation of trip makers Person # Sample # Trip # Mode of travel Number of vehicle Travel purpose Station # Direction Starting time Ending time Origin Land use origin Destination Land use destination Exit or entrance station Stops in area Purpose for stops Stop location Registration of auto Industry and business Total trips Capacity Garaged Parking Cross screenline Car pool Section of residence District of residence Zone of residence Section of origin District of origin Zone of origin Section of destination District of destination Zone of destination One hour expansion factor I.A.S. expansion factor 165 Location 19- 21 22- 25 26- 27 28- 29 30- 31 33- 35 36- 37 40- 79 80- 81 82- 85 86- 88 90- 91 92- 93 94- 95 97-100 101-104 105-110 111-112 113-118 119-120 121-122 123 12# 125-130 131 132-13h 135-137 138-1G0 141 1b2 143 144 165-166 167-168 169-171 172-173 17h-175 176-178 179-180 183-185 186-188 193-196 197-200 Characters :- kkwuNUNNUNNHHHHUUUHmF-‘HNNONOvfi'i-‘HNNNHUJ—‘NOHHNkfiHNNNkWI—‘HMONA)? 166 Merged I.A.S. and Internal Trip Records 200 Character Record Field Description Location Characters City # 1- 2 2 Form # 3- 4 2 Residential tract 5- 7 3 Residential block 8- 10 3 Person # ll- 12 2' Sample # 13- 16 4 Trip # 17- 19 3 Mode of travel 20 1 Number of vehicle 21- 22 2 Trip purpose from 23 1 Trip purpose to 24 1 Starting time 25- 28 4 Ending time 29- 32 4 Trip origin tract 33- 35 3 Trip origin block 36- 38 3 Origin land use 39- 4O 2 Trip destination tract 41- 43 3 Trip destination block 44- 46 3 Destination land use 47- 48 2 Type of parking 49 l Crossing screenline 50 1 Car pool 51 1 Month 52- 53 2 Day 54 1 Structure type 55 1 Auto at address 56- 58 3 Person at address 63- 64 2 Person over age of 5 65- 66 2 Years at residence 67- 68 2 Own or rent 69 1 Residential value 70- 72 3 Education of household head 73- 74 2 Person employed 75 1 Income 76 1 Industry 78 1 Occupation 79- 80 2 Age 81 1 Total trips at household 82- 84 3 Zone of residence 85- 87 3 Zone of trip origin 88- 9O 3 Zone of trip destination 91- 93 3 Trip expansion factor 94- 97 4 I.A.S. zonal expansion factor 98- 101 4 Transit-index 102 1 APPENDIX C NUNEE RICA L AND GRAPHICA L STATISTICA L STANDARDS FOR TRANSPORTATION MODEL BUILDING Numerical and Graphical Statistical Standards for TranSportation Model Building This appendix eXpands the brief descriptions of Chapter II, Section C, l a: concerning statistical tests used in model development. Numerical and graphical tests will be defined. It is important that both tests be used in that the numerical test may indicate that a model is statistically quite good, while the graphical test can indicate that its predictability is not adequate. Thus, these two testing parameters must be used in congress. The defini- tions are generally based on the manual deveIOped by Wayne Meyerowitz for the Michigan Department of State Highways and Transportation.45 Numerical Test 1. Mean: The mean is the arithmetic average of a population. The formula is computed as follows: Mean=X1+X2+X3. . . x N n N = Number of Observations 2. Standard Deviation: The Standard Deviation is used to mea- sure the dispersion of variables about the mean. The stand- ard Deviation is calculated as follows: 45Wayne Meyerowitz, Documentation Manual: Statistical Anal- ysis In TranSportation Planning, Michigan Department of State Highways and Transportation (1974). 167 168 Standard Deviation = \ (Xi _ §)2 N-l R = Mean Xi = Each Observation N = Number of Observations The statistic is a measure of how far, on the average, the data in a distribution is scattered about the mean. Standard Error of Estimate: The Standard Error of Esti— . . mate is a measure of diSpersion. To be Specific, it is the Standard Deviation of the residuals. (Residuals are the results of subtracting the predicted value from the actual observed value.) In other words, the Standard Error of Estimate provides information on how well the model fits the actual situation. A perfect fit occurs when the value of this statistic reaches zero. The standard error of estimate is used to detect residuals (observations that lie within at least three or four standard errors of estimate from the mean of the residuals). The value of the standard error of estimate is computed as follows: 8. E. of Estimate = \JKActual - Predicted)7 n-k n = Number of Observations k = Number of Parameters (i.e. , p0pu1ation characteristics) In some situations, this identification of outlier will culmi- nate in the application of a special factor or rate to the ob- servation. For example, in trip generation, it is often necessary to apply a Special trip rate to outlier zones. Correlation Coefficient and Coefficient of Multiple Determi- nation: Two important numerical techniques used for model verification are the correlation coefficient (R), and coeffi- cient of multiple determination (R2). The "R" value pro- vides an index of the degree to which the actual values are related to the values predicted by the model. The theoreti- cal upper limit of a correlation coefficient is +1 whereas the lower limit is -1. That is, a perfect direct (positive) rela- tionship has a value of +1 and a perfect indirect (negative) relationship has an associated correlation coefficient of —1. In model verification, the value of this statistic is to be maximized. The coefficient of multiple determination (Rz) indicates what proportion of the total variation of the 169 dependent variables observed value is accounted for by the regression line. A high R2 is desired. Statistically, R2 is defined as the ratio of the sum of the squares of deviation due to regression to the sum of squares of de— viation about the mean. T-Value: The T -value test gives an indication as to the significance or lack of significance of the regression coefficient of each independent variable in the equation. The "t—value" of an independent variable is computed by dividing the regression coefficient of that variable by its standard error. With a "t" value approximately of two, i..- the statement can be made that the regression coefficient is significant (non-zero) at the 95% level of confidence. All coefficients must be significant at the 95% level for this analysis. I a‘ql. Lit 9.4".” Graphical Test 1. Actual versus Residual and Actual versus Predicted Plots; Residual versus actual plots are used to graphically de- scribe the model. For example, numerical descriptive measures might indicate that our model is statistically recreating the actual data in an effective manner but this says little in relation to how well the model is predicting. This information on predictability is acquired through analysis of the actual versus residual plot or actual ver- sus predicted plots. The following general situations can occur: The model is predicting perfectly. Here, all the residuals are equal to zero since the model exactly recreates the actual situation. The model is predicting "well". In this situation, the model predicts the actual phenom- enon in an unbiased fashion. That is, the plot will be characterized by an even Spread of observations in both sides of the "0" line. The model is predicting in a "biased" fashion. Here we usually find that the model is both consistently overpredicting the lower values of the actual data as well as underpredicting the higher values of the actual data. 170 Normal Deviate Plots: The usefulness of the residual ver- sus actual plot in describing model predictability can be maximized by standardizing the residuals. This transfor- mation of the residuals into standard error of estimate units is accomplished as shown below. Normalized Deviate = Residuals S. E. of Estimate These standardized residuals, called normal deviates, provide a means for detecting potential outliers in a residual versus dependent and independent variable plots. Plots of residuals versus independent variables are used to help discover if basic assumptions of re- gression or rate analysis have been violated and to give further data as to value of the variable to the equation. Furthermore, normal deviates optimize the analysis of predictability in a dependent variable plot. Establishment of predictability is a key element in the development of transportation planning models. The planner can more confidently utilize the model for future tranSportation plans if efficient predictability has been substantiated. For example, normal deviate versus dependent variable plot will show how well a calibrated model is recreating the actual volumes. If a particular model is predicting in a biased manner, modification must be seriously con- sidered before pursuing further transportation objectives. Model Predictability: It has been assumed that the pre- dictability of a model can be determined. As mentioned, one of three actual versus residual situations is possible for a model. The question arises as to what one should do if situation 3--"biased prediction"--should occur. Figure 30 from Statistical Analysis, A Commter Oriented Approach46 indicates what measures one might take to better fit the model. Graph A, an aggegate model, is predicting well while Graph B shows heteroscedasticity (i. e. , lack of constant variance). This situation calls for transformation of the "Y" variable. Plot C shows a more typical situation as seen by the Transportation Planner, 1. e. , need for another variable. Often our models are best attempts to minimize this prediction situation. Plot D reveals a need for a quadratic term to be added to the equation. 46A. A. Afifi and s.p. Azen, Statistical Analysis, A Computer Oriented Approach (New York and London: Academic Press, Inc. , 1972), p. 366. 171 PCoUCmmmccH efleeflee> capmhcmzo so smmcwq mpwofipmmcmomOSmpmm :3 mpoam asapflmom yo moagsmxm om mpsmfim mapmflsm> pCmoCommch pmmcfiq Abv poam mpmsvmo< Amy APPENDD( D 24 HOUR TRIP GENERATION EQUATION FOR HOMEBASED WORK PRODUCTIONS AND ATTRACTIONS INTERNAL HOMEBASED WORK PRODUCTION (H B WORK P) The average number of work production trips per zone is 741 with a standard deviation of 634. Only one independent variable is included in the final equation for this work trip variable. Resident labor force per zone was judged to be the most logical predictor. The almost perfect correlation between the dependent variable and the independent variable along with the T-test, indicates that the relationship between resident labor force and work trips is estimated very precisely. Each resident labor generates approxi- mately 1. 7 work production trips per day. Analysis of the statistical data obtained from this model reveals that the equation is well within the acceptable limits of statistical validity and reliability set forth previously. Based on the above, it is recommended that the model be applied in the future projection of internal homebased work production trips. 172 u..— 173 Table 34 Regression Summary - Homebased Work Production (Resident Labor Force) Dependent Variable (Y): Homebased Work Production Mean Value (Y) 741.61 Standard Deviation (Y) 634.49 Multiple Correlation Coefficient (R) 0.98 Coefficient of Multiple Determination (R2) 0.96 Standard Error of Estimate 123.01 Constant Term - 22.47 Number of Observations 315 Special Generators (None) 0 Unit of Analysis Zone The Equation: Y = 1.71 (X1) - 22.47 Simple Independent Regression T Mean Correlation Variable Coefficient Value Value With Y (X1) Resident Labor/ 1.71 89.67 448.02 .98 Zone 174 INTERNAL HOMEBASED WORK AT TRACTION ( H B WORK A) Only one independent variable is included in the final regression equa- tion predicting internal homebased work attraction. Total zonal employment is very highly correlated with work trips and R2 is . 99. The "T" value of 175 indicates that the relationship between total employment and work attrac— tion is estimated very precisely. Approximately 1.4 work trips were gener- ated by each employee on an average weekday. Each of the 313 zones in- cluded, generated about 703 work trips per weekday. The standard deviation for work attraction is 2215. This indicates that there exists a wide variation in the distribution of these trips by zone. This is to be eXpectcd since some of the zones contain very large automotive manufacturing plants which employ thousands of workers and consequently result in thousands of work trips for these zones. However, such zones are not considered as Special generators because the ratio of work attraction trips to total employment is the same for these as it is for the other zones (i.e. , about 1.4). Plotting of the residuals was done in two parts; one for those zones of less than 3000 work attraction trips (actual) per zone and the second for zones of more than 3000 trips. The two residual plots (obtained from the same 175 regression equation) provide a more informative visual presentation than could have been offered by a single plot. The equation satisfies the requirement for statistical validity and re- liability and is therefore, recommended for use in future projections of work attraction trips in Genesee County. 176 Table 35 Regression Summary - Homebased Work Attraction (Total Employment) Dependent Variable (Y): Homebased Work Attraction Mean Value (Y) 703.15 Standard Deviation (Y) 2215.08 Multiple Correlation Coefficient (R) .99 Coefficient of Multiple Determination (R2) .99 Standard Error of Estimate 222.58 Constant Term 61.14 Number of Observations 313 Special Generators (Zones 105, 186) 2 Unit of Analysis Zone The Equation: Y = 61.14 + 1.43 (X1) Simple Independent Regression T Mean Correlation Variable Coefficient Value Value With Y (X1) Total Employment/ Zone 1.43 175 449 .99