1. Real-time Tracking and Analysis of The Dynamics in Activity Scheduling and Schedule Execution By Jianyu (Jack) Zhou 8/08/06 Advisor: Reginald Golledge Committee members: Jack Loomis, Keith Clarke, and Richard Church.

2. Outline 1.Problem Statement 2.Research Assumptions 3.Research Hypotheses 4.Theoretical Relevance 5.Methodology  Survey Design  Data processing  Data Analysis and Modeling 6.Conclusion 7.Further Research

3. Problem Statement Activity scheduling is a continuous process of spatial and temporal choice over time. Activity execution represents the process that the planned schedule is converted into the sequence of implemented activities that are continuous in space and time.

4. Problem Statement (cont’) Activity scheduling study in Transportation and Geography research context focuses on two aspects: The temporal-spatial decision-making structure embedded in the scheduling process. The temporal-spatial decision-making structure embedded in the scheduling process. The linkage of schedule to actual activity execution. The linkage of schedule to actual activity execution. Two existing approaches for analyzing and predicting the dynamic process of activity scheduling: Econometric approach Econometric approach Cognitive approach Cognitive approach

5. Problem Statement (cont’) Objectives of this research are two-fold: Develop the systematic techniques for tracking and recording the interlaced process of real-life activity scheduling and execution. Develop the systematic techniques for tracking and recording the interlaced process of real-life activity scheduling and execution. Reveal the critical factors that affect the relations between activity schedules and their actual execution based on “revealed” in-field data. Model the relationship and quantify the effects of the factors changes on people’s activity temporal-spatial choices with respect to their schedules. Reveal the critical factors that affect the relations between activity schedules and their actual execution based on “revealed” in-field data. Model the relationship and quantify the effects of the factors changes on people’s activity temporal-spatial choices with respect to their schedules.

6. Research Assumptions Obligatory vs. Discretionary Activities Stochastic decision making Continuous revisions of activity scheduling

7. Research Hypotheses The congruence and deviation relations between individual activity schedules and their actual execution can be consistently described in a series of relevant factors -- socio-demographic characteristics, spatial- temporal constraints, etc. These factors affect the schedules and their executions in different ways. Not every type of activity is thoroughly planned ahead of time. Mobile real-time system constitutes a powerful tool to capture the asynchronous activity decision-making and execution process with the least time and location constraints.

8. Theoretical Relevance Two existing propositions by Hayes- Roth and Hayes-Roth (1979) reflected different views and understanding about the scheduling process - Successive refinement model and Opportunistic model. Hagerstrand’s (1970) time-space geography

9. Methodology: Survey Design -Integrated Activity Scheduling/Execution Data Collection This research implemented a data collection system for pilot study. The system offers unique advantages for travel/activity Survey.

10. Methodology: Survey Design Start-up form presents four modules that constitutes the main function of the system devices. Module 1 – Personal Info and Week Schedule It helps the survey respondents to set up personal demographic background and establish a preliminary week schedule that came up to them at the interview time.

11. Methodology: Survey Design (cont’) Module 2 - Schedule Activities or Refine Schedules Capture schedule- related information with Schedule-an- activity form. The accomplished schedules are listed on the weekday tab pane with a brief description

12. Methodology: Survey Design (cont’) Module 3- Trace Activity Implementation Module Trace travel and activity Execution. Capture the travel route by drawing tool when most of GPS points are invalid. Identify the relevant schedule to the current activity and their congruence / deviation relationship

13. Methodology: Pilot Survey A total of 20 volunteers (13 males, 7 females) recruited for one-week survey. The ages of the survey respondents fall within the range of 20-35, with the average being 28.75. Each survey respondent commonly uses 4 types of travel modes - at most 6 and at least 2. Each survey respondent indicated 20.3 visited or frequently-visited locations over the survey period.

14. Methodology: Pilot Survey The non-response rate for activities and scheduling tracking were 13.75% and 4.1% respectively. Survey Feedbacks: Most survey participants consider the survey questions clearly and concisely organized. Most survey participants consider the survey questions clearly and concisely organized. Most of the survey participants (90%) have no problem with data uploading at the end of the survey day. Most of the survey participants (90%) have no problem with data uploading at the end of the survey day. Fatigue Effects: Respondents’ activity counts and their data entry steps are highly correlated. Respondents’ activity counts and their data entry steps are highly correlated. Average entry time for the travel/activity tracking module - 26.18 seconds per form. The average data entry steps in the module – 35.34 steps per day. Average entry time for the travel/activity tracking module - 26.18 seconds per form. The average data entry steps in the module – 35.34 steps per day.

15. Methodology: Survey Results Highlight- Schedule and Activity Intensities No strong correlation between activity intensity and the scheduling steps were revealed. Relative scheduling intensity -- measured as the ratio of schedule count against the total activity count. Recreation and Entertainment activities-- the most actively scheduled. Seconded by Social activities. Household Obligation activities -- least planned before execution.

16. Methodology: Survey Results Highlight – Activity Constraints Spatial-temporal constraints: spatial constraints along the path are more rigid than their temporal counterparts Coupling constraints: Household Obligation and Work/school activities are subject to the least coupling constraints (about 65-75% completed alone), About 75% Social activities are expected to be completed in group.

17. Methodology: Data Processing – A general three-step map matching algorithm Three-step map matching Data Preprocessing— Cluster Reduction and Density Leverage Data Preprocessing— Cluster Reduction and Density Leverage Multiple-Hypothesis Matching with Rank Aggregation Multiple-Hypothesis Matching with Rank Aggregation Dempster Belief Test for Travel Off-Road/Noise Discernment Dempster Belief Test for Travel Off-Road/Noise Discernment On average the map matching algorithm reaches a matching accuracy of 95.74% Travel Mode Total Number of Travels Average Matching Accuracy (in percentage) Walk10494.75 Car23497.04 Bicycle61 91.19 * Carpool3497.17 Vanpool3100.00 Local bus 4596.17 Total48195.75 Map Matching Accuracy by Travel Modes

18. Methodology: Data Analysis – Travel optimization Travel Mode Total Numbe r of Travels Average Diff between Actual Route and Shortest Time Route Average Diff between Actual Route and Shortest Path Route Walk110 4.32 * 4.36 * Car23414.4420.96 Bicycle55 1.35 * Carpool3410.7121.38 Vanpool3 37.00 # 45.33 # Local Bus 45 48.40 ^ 64.73 ^ Total48113.6219.05 Activity Category Total Number of Travels Average Diff between Actual Route and Shortest Time Route Average Diff between Actual Route and Shortest Path Route Eat/sleep/person al hygiene 1088.8418.84 Household obligation 9119.6325.86 Recreation/entert ainment 537.5714.70 Services and errands 18 5.33 * 5.22 * Shopping3912.2818.33 Social1212.175.08 Work/school160 16.79 ^ 19.55 ^ Total48113.6219.05 Diff of actual travel route and the shortest distance/time route by activity types Diff of actual travel route and the shortest distance/time route by travel modes

19. Methodology: Data Analysis – Route Choice Analysis Binary Logistic analysis is used to analyze the traveler’s route choice preference between the shortest time path and the shortest distance path. Route distance in miles. Route distance in miles. Travel time in minutes. Travel time in minutes. Number of street links (extracted from the GIS base map). Number of street links (extracted from the GIS base map). Number of intersections encountered during the travel. Number of intersections encountered during the travel. Off-road Ratio. Off-road Ratio. Gender of the traveler. Gender of the traveler. Travel Mode. Travel Mode. Male travelers tend to choose travel path that is relatively time- optimized compared to female travelers. As the travel distance on a route increases, travelers will shift their routing aim toward time-optimization rather than distance optimization. P = exp(U)/(1+ exp (U) )

20. Methodology: Data Analysis - Schedule Horizon Analysis Shopping activities and Services and Errands activities-- short schedule horizon; Work and School activities and Household Obligation--most distant schedule horizon. Activities with short duration are less likely to be planned out early. Activities with longer durations tend to be associated with more distant schedule horizons. Cluster 1234 Activity Duration (in minutes ) 1477.47462.6718937.85 Schedule Horizon (in days) 3.772.541.791.73

21. Methodology: Data Analysis - Missing Value Analysis (Missing/Mismatch Percentage) Activity locations were planned out well in recorded schedules. End time tends to suffer the greatest degree of uncertainty. Start time and end time have the lowest mismatch percentage. Location Miss ing Date Miss ing Start Time Missi ng End Time Missing Location Missing 14.49 * Date Missing 33.6438.79 Start Time Missing 37.8545.3327.10 End Time Missing 51.4049.53 15.42 # 41.59 ^

22. Methodology: Data Analysis -Missing Value Analysis (cont’) n = 211 Row # Number of Cases Missing Patterns(a) Complete if...(b) Horizon day(c) Location Missing Date Missing Start Time Missing End Time Missing 168 681.6513 220 X88.4171 332 XX119.2348 430 X 982.7505 520 XX 1244.0085 66 X 742.1619 73 X XX127.1588 821 XXX1812.1066 911 X X1293.2968 Activity locations is prioritized over other schedule elements. Schedules with a short schedule horizon tends to have an undetermined start or end time ( by row 3 and 7). Schedules with a long schedule horizon are associated with the undetermined activity date (row 5 and 9).

23. Methodology: Data Analysis – Nested Logistic Modeling Estimation of a nested logistic model is used to study the potential deviation of the respondents’ schedule execution from their stated intention – the schedule Assume that schedule execution is an integrated decision-making process that conforms to a model in a decision tree form. Schedule Execution Choice Evaluation: The utility function-- the weighted linear addition of three vectors of attributes. The utility function-- the weighted linear addition of three vectors of attributes. Factors that affects activity participation choice (Aap), activity start time choice (Aas) and combinations of activity participation and start time choice (Aaps). Factors that affects activity participation choice (Aap), activity start time choice (Aas) and combinations of activity participation and start time choice (Aaps).

24. Methodology: Data Analysis – Nested Logistic Modeling (cont’) Only “Total work/school time duration” variable is statistically significant (5% level) in the two-level model. Continue to model the three discrete levels of activity start time choices under a MLM framework. “Travel duration”, “Travel distance”, “The ratio of Off-road travel”, “Work/School activity type” and “End time missing” are the significant factors that affect the activity start time choices at the significance level of 0.05. MLM model results offer sensitivity quantification: Given the same status of the other variables, for each 1 mile increase of travel distance, the odds of activity start on time decrease by 1- exp (0.12 * 1) = 12.7% and odds of activity start early decrease by 1- exp (0.2 * 1) = 22.1 % Given the same status of the other variables, for each 1 mile increase of travel distance, the odds of activity start on time decrease by 1- exp (0.12 * 1) = 12.7% and odds of activity start early decrease by 1- exp (0.2 * 1) = 22.1 %

25. Conclusion Innovative Data collection methodology : Conceptualized and implemented a real-time system tool that facilitates the study of the dynamic linkages between the activity scheduling and execution process. Conceptualized and implemented a real-time system tool that facilitates the study of the dynamic linkages between the activity scheduling and execution process. Small-scale pilot study by this research showed that the methodology was successful in achieving our goals without incurring significant survey fatigue effects. Small-scale pilot study by this research showed that the methodology was successful in achieving our goals without incurring significant survey fatigue effects. In-depth analysis of the routing behavior, scheduling pattern of various activity categories and the inter-relationship between scheduling and correlated activity execution: Using a nested logistic modeling approach, the research was able to identify the single factor that dominates the activity participation and start time choice decision making. Using a nested logistic modeling approach, the research was able to identify the single factor that dominates the activity participation and start time choice decision making. The further one-level multinomial logistic modeling efforts identified five factors that affect the activity start time choices at a significant level. The modeling results offer us the quantitative measures for effects of the factor changes on activity start time choices. The further one-level multinomial logistic modeling efforts identified five factors that affect the activity start time choices at a significant level. The modeling results offer us the quantitative measures for effects of the factor changes on activity start time choices.

26. Further Research Improve current data collection system Reduce the load of the survey task. Reduce the load of the survey task. Provide effective survey guidance. Provide effective survey guidance. Enhance the multi-task mode of the survey program. Enhance the multi-task mode of the survey program. Compare instrument bias and survey burden brought by the system with traditional activity/travel data collection methods Compare instrument bias and survey burden brought by the system with traditional activity/travel data collection methods Further modeling efforts Artificial neural network (ANN) provides us a method to learn and approximate the relationship with a discrete-valued function in the form of a network of interconnected neurons. Artificial neural network (ANN) provides us a method to learn and approximate the relationship with a discrete-valued function in the form of a network of interconnected neurons. Decision tree modeling to infers the hierarchical decision structure from the empirical data by induction with no prior model-structure assumption made. Decision tree modeling to infers the hierarchical decision structure from the empirical data by induction with no prior model-structure assumption made.