Incorporation of ing Children to School in Individual Daily Activity Patterns of the Household Members Peter Vovsha, Surabhi Gupta, Binny Paul, PB Americas Vladimir Livshits, Petya Maneva, Kyunghwi Jeon, Maricopa Association of Governments (MAG) Motivation of the Research ing children to school by adult household members (workers and/or non workers) is a frequent travel arrangement. The need in escorting is characterized by a fixed schedule constraint on the school activity side, thus this arrangement requires a schedule consolidation on the side of the chauffeur. ing children to school is a complicated choice phenomenon where several considerations are intertwined including possible alternative mode options for children and chauffeurs, schedule constraints, as well as possible forms of escorting such as dropping off children at school on the way to work (ride sharing) vs. escorting them in a separate tour (pure escort)). The resulted choice construct were previously considered too complicated for an operational Activity Based Model (ABM) and the published research approaches were in general limited to modeling only one child at a time. In this paper, we present a more general approach that can effectively handle the associated choice complexity for the entire household day. The presented model included as part of the operational ABM developed for MAG. Placement of the ing Sub Model in the ABM Model Chain The proposed model is applied in the ABM after the following sub models: Population synthesis, Long term choices of work locations for workers and school locations for students, Mid term choices for individual mobility attribute such as household car ownership and person transit pass holding, Daily Activity Pattern (DAP) type for each household member that defines if the person has an outof home work or school activity on the given day, or has some other out of home activities but not work or school, or is not travel active on the given day. Preferred time of day choice for all work and school activities planned for the day (tours). In the framework of the escorting choice sub models, the outcomes of the prior sub models are considered as inputs. In particular, all school tours and corresponding escorting needs in terms of school activity start and end times are known as well as potential chauffeurs (adult household members with an travel active day) are known. Additionally, all chauffeurs are broken into two groups: 1) workers and students who have a work or school activity on the given day and could incorporate escorting as an 1
additional stop on the way to work or school, and 2) non workers who do not have a work or school activity on the given day and could escort children to and/from school as a separate home based tour. For a chauffeur who is a worker or student, his (preferred) work or school schedule is known that is an important constraint in the schedule consolidation process. Engagement in an escorting activity either as the chauffeur or as an escorted child has an impact on the subsequent modeled choices: All non work tours and activities modeled for chauffeurs are subject to that fact that an escorting activity is taken by the person; if the chauffeur is a worker, the escorting activity inserted as a stop on the work tour; if the chauffeur is a non worker, escorting activity forms a separate non work tour. The mode for the work tour involving escorting or for escorting tour is assumed auto (except for very short tours where non motorized modes are also considered). A schedule consolidation procedure for workers is applied that synchronizes the original work schedule with the school schedule of the escorted child. For children who are escorted to school the assumed mode is auto passenger (except for short distances to school where non motorized modes are also considered). If a child is escorted, no additional stops are modeled on the corresponding school half tour except for cases where several children are escorted by the same chauffeurs on the same tour. Children who are not escorted by the household members can choose a different mode for the school tour (transit, school bus, nonmotorized) or be escorted by a non household member. Choice Model Structure and Dimensions The choice model is applied for the entire household day and involves the following household members who are modeled simultaneously: Up to 3 school children with escorting need including day care, preschool children and school children under 16. If a household has more than 3 children with escorting needs, the youngest are consider in the model. This covers more than 99% of the observed cases. Up to 2 potential adult chauffeurs with an active travel pattern on the given day including workers, non workers, retirees, university students, and school children of driving age. If a household has more than 2 chauffeurs available, a rule based algorithm was applied to choose the most probable ones based on the person type (for example, female non workers and part time workers proved to be the most frequent child chauffeurs). In the available dataset, this covers 100% of the observed cases. The choice model includes the following dimensions as shown in Table 1: 2
Table 1: Number of ing Alternatives by Choice Dimensions Alternatives ing need and bundling calculated for outbound and inbound directions separately 1 st Child 2 nd Child 3 rd Child Number of escorting tasks after bundling Number of chauffeur assignments and escort types 1 with 2 nd and 3 rd with 1 st and 3 rd with 1 st and 2 nd 1 4 2 with 2 nd with 1 st separately 2 4 4=16 3 with 3 rd separately with 1 st 2 4 4=16 4 separately with 3 rd with 2 nd 2 4 4=16 5 separately separately separately 3 4 4 4=64 6 with 2 nd with 1 st No escort 1 4 7 separately separately No escort 2 4 4=16 8 with 3 rd No escort with 1 st 1 4 9 separately No escort separately 2 4 4=16 10 No escort with 3 rd with 2 nd 1 4 11 No escort separately separately 2 4 4=16 12 separately No escort No escort 1 4 13 No escort separately No escort 1 4 14 No escort No escort separately 1 4 15 No escort No escort No escort 0 1 Total 189 The total number of choice alternatives of this model is formidable when all escorting alternatives with respect to children need are combined with the possible assignment of chauffeurs and escorting type (on the way to work/school vs. pure escort as a separate home based tour). For each escorting task there are up to 4 possible chauffeur assignments: 1 st chauffeur, escorting on the way to/from work, 1 st chauffeur, pure escorting as a separate home based tour, 2 nd chauffeur, escorting on the way to/from work, 2 nd chauffeur, pure escorting as a separate home based tour. For a household with 3 children and 2 available chauffeurs, the number of unique alternatives considering both directions will constitute 189 189. The choice tree is truncated significantly for smaller households with respect to number of children (1 or 2 instead of one) and/or chauffeurs (1 instead of 2). In the model estimation and application, we consider the entire structure to cover all possible cases where smaller households are treated by making irrelevant alternatives unavailable. Since a simultaneous estimation of a joint choice model with all possible alternatives is technically cumbersome, a decomposition method is applied. It should be noted that the escorting decisions in outbound and inbound directions can be in many respects considered independently with a limited 3
number of linking factors. The linkage between outbound and inbound escorting decisions includes the fact that the same child who relies on escorting when going to school might be more relying on escorting from school since he would less likely be a transit and school bus user. On the chauffeur side, the same person would probably take care of escorting in both directions because of the willingness to adjust the schedule and/or having the school on the way to/from work. This linkage creates some symmetry in outbound and inbound escorting decisions that cannot be completely ignored. To address this, the model is decomposed into the following sequence choices applied iteratively: 1. Choice of outbound escorting arrangements independently of inbound arrangements, 2. Choice of inbound escorting arrangements conditional upon the chosen outbound arrangements, 3. Choice of outbound escorting arrangements conditional upon the chosen inbound arrangements, 4. Go to 2 for several iterations. Statistical Evidence on ing Children to School Some key statistical data on escorting children to school in the Phoenix, AZ, Metropolitan Region from the add on NHTS survey, 2008 are shown in Tables 2 (focus on the escorted child characteristics) and 3 (Focus on the chauffeur characteristics). 4
Table 2: Observed ing Frequency by Child Person Type Type Number of School tours % by escort type Driving Age School Child Predriving age school child Prescho ol Child Total Driving Age School Child Predriving age school child Prescho ol Child Outbound half tour: Ride Sharing 23 186 69 278 10.9% 19.6% 40.1% 20.8% Pure 40 271 65 376 19.0% 28.5% 37.8% 28.2% No 148 494 38 680 70.1% 51.9% 22.1% 51.0% Total 211 951 172 1,334 100.0% 100.0% 100.0% 100.0% Inbound half tour: Ride Sharing 10 128 58 196 4.8% 13.5% 35.8% 14.8% Pure 35 299 68 402 16.7% 31.5% 42.0% 30.5% No 165 521 36 722 78.6% 55.0% 22.2% 54.7% Total 210 948 162 1,320 100.0% 100.0% 100.0% 100.0% Outbound/Inbound Combinations: Ride Sharing/Ride Sharing 3 83 53 139 1.4% 8.8% 32.7% 10.5% Ride Sharing/Pure 7 52 8 67 3.3% 5.5% 4.9% 5.1% Ride Sharing/No 12 49 3 64 5.7% 5.2% 1.9% 4.8% Pure /Ride Sharing 4 10 2 16 1.9% 1.1% 1.2% 1.2% Pure /Pure 21 186 51 258 10.0% 19.6% 31.5% 19.5% Pure /No 15 75 7 97 7.1% 7.9% 4.3% 7.3% No /Ride Sharing 3 35 3 41 1.4% 3.7% 1.9% 3.1% No /Pure 7 61 9 77 3.3% 6.4% 5.6% 5.8% No /No 138 397 26 561 65.7% 41.9% 16.0% 42.5% Total 210 948 162 1,320 100.0% 100.0% 100.0% 100.0% Symmetric Outbound/Inbound Combinations with same chauffer: Ride Sharing/Ride Sharing 3 70 49 122 1.4% 7.4% 30.2% 9.2% Pure /Pure 18 167 45 230 8.6% 17.6% 27.8% 17.4% Total 5
Table 3: Observed Distribution of ed School Tours by Chauffeur Person Type type Number of escorted school tours by chauffeur person type Total Unknown Person Type Full Time Worker Part Time Worker University Student Nonworker (under 65) Retired (65 or older) School Child (16 or older) Outbound half tour: Ride sharing 216 32 37 6 291 Pure escort 124 69 22 175 6 396 Inbound half tours: Ride sharing 41 109 22 25 4 160 Pure escort 77 88 68 18 162 4 340 Number of symmetric outbound/inbound combinations with the same chauffeur: Ride/ride 85 17 21 4 127 /escort 60 44 7 129 3 243 Outbound half tours: % by chauffeur type Ride sharing 0.0% 74.2% 11.0% 12.7% 2.1% 0.0% 0.0% 100.0% Pure escort 0.0% 31.3% 17.4% 5.6% 44.2% 1.5% 0.0% 100.0% Inbound half tours: % by chauffeur type Ride sharing 25.6% 68.1% 13.8% 15.6% 2.5% 0.0% 0.0% 100.0% Pure escort 22.6% 25.9% 20.0% 5.3% 47.6% 1.2% 0.0% 100.0% Symmetric outbound/inbound combinations with the same chauffeur: % by chauffeur type Ride/ride 0.0% 66.9% 13.4% 16.5% 3.1% 0.0% 0.0% 100.0% /escort 0.0% 24.7% 18.1% 2.9% 53.1% 1.2% 0.0% 100.0% Overall, escorting proved to be a very frequent case that affect close to 50% of school trips. It can be seen that there is a logical tendency in terms of escorting needs by child age (the younger the child is the more he relies on escorting). There is also a significant differentiation by person type in terms of performing the chauffeur role. Logically, the most frequent chauffeurs for escorting on the way to and from work are workers (ride sharing) while non workers and part time workers are the most frequent chauffeurs for pure escorting. There is also a certain logical bias towards escorting in the outbound direction compared to inbound direction that can be explained by easier schedule synchronization between school children and workers in the morning. 6
Structure of the Utility Function The utility function includes the following components: For each child, a utility of being escorted vs. going to school on his own (by transit, school bus, nonmotorized mode, or as a passenger of a non household carpool) is formulated as function of the corresponding level of service characteristics and person variables. This utility component is calculated separately for inbound and outbound direction. For each chauffeur, a (dis)utility of taking an escorting task is formulated. For workers and (university or high school) students, this disutility is primarily associated with a detour compared to a strait commuting without stopping at the school. For non workers, this disutility is primarily associated with the distance and time to implement the escorting task as well as the work schedule adjustment needed to accommodate escorting. This utility component is calculated separately for inbound and outbound direction. Bundling of several children on one half tour reduces the total detour or travel time from home compared to escorting each of them separately. For each child, a symmetry component for escorting need is included to address the fact that it is in general inconvenient to have escorting in only one direction. This component creates a linkage between escorting decisions made for outbound and inbound directions. For each child and chauffer combination, a symmetry component for escorting task allocation is included to address the fact that escorting tasks for the same child is more frequently implemented by the same chauffeur than by different chauffeurs. This component creates a linkage between escorting decisions made for outbound and inbound directions. Each component includes many explanatory variables such as person and household characteristics as well as level of service variables by different modes for both chauffeurs and school children. The model does not have non behavioral flat constants for each particular escorting alternative. Thus, the utility of each alternative is combined of the corresponding components that reflect the worth of each particular arrangement from the perspective of all household members involved. Model Estimation Results The short paper format does not allow for presentation of all estimation results in detail. We present as an example the sub model estimated for inbound direction (escorting children from school to home) in Table 4. Each variable relates to a certain utility component. The presentation and full paper will include a detailed analysis and impact of each variable. 7
Table 4: Estimation Results for Inbound Direction Utility Parameters Coefficient T stat Chauffer person and Commuting Characteristics: Female Ride Sharing 0.393 0.87 Female Pure ing 1.644 6.72 Male Ride Sharing 0.705 1.55 Male Pure ing 1.818 7.81 PT Worker Ride Sharing 0.209 0.38 PT Worker Pure ing 0.175 0.60 University Student Ride Sharing 0.589 1.01 University Student Pure ing 0.963 1.97 Non Working Adult Pure ing 0.587 2.08 Retiree Pure ing 0.650 0.98 Age 35 or younger Pure ing 2.187 10.36 Distance Pure ing 0.028 2.00 Absolute Distance Deviation Ride Sharing 0.015 0.53 Walking Distance to Work/School for Chauffer: Ride Sharing 0.849 1.78 Pure ing 0.364 1.40 Travel Time to Work/School for Chauffer: Ride Sharing 0.028 2.10 Child person and school tour characteristics: No ing: Driving Age (16 years +) 0.351 0.74 Pre driving Age (6 to 9 years) 0.648 2.04 Pre driving Age (10 to 15 years) 1.131 3.61 Pre driving Age (6 to 15 years) 0.000 Pre school Age (under 5 years) 2.012 4.65 Distance to school 0.067 0.49 Distance to school Squared 0.002 0.60 Distance to school Log 0.821 1.77 Distance to school Squared Age 0 to 5 years 0.026 1.32 Distance to school Squared Age 6 to 9 years 0.001 0.31 Distance to school Age 16+ years 0.048 0.82 Travel Time to School for Child: Pure ing 0.045 1.18 Household Variables Income No ing 8
Utility Parameters Coefficient T stat $49,999 or Less 0.074 0.41 Car Ownership Zero cars No ing 0.538 0.76 Cars less than Workers Ride Sharing 3.175 2.30 Cars less than Workers Pure 0.573 0.71 Outbound Direction ing Type for Child No ing in both Direction 1.919 10.41 Same Driver in both direction 0.987 6.84 Number of Observations 718 Likelihood with Constants only 897.9741 Final likelihood 812.0995 Conclusions and Further Research The model is included as part of the MAG ABM currently being developed. Inclusion of this sub model enhances the behavioral realism of the entire model system by creating more linkages between different household members and accounting for associated constraints. Several new research dimensions have become clear in the process of developing the model: ing children is not bound to school activities. There are many cases of escorting to non school discretionary activities ( soccer moms ). These cases require a different choice structure compared to the structure described in this paper. ing decisions are closely intertwined with mode choice decisions. In particular, commuting mode choice for workers and mode choice for school trips are linked through escorting. This linkage is somewhat taken into account in the current model formulation through level of service variables for non escorting options that compete with escorting. However, there could possibly be a more integrated formulation where commuting mode choice decisions of workers and school children are modeled simultaneously. This is beneficial for certain types of projects and policies. For example, promoting transit commuting could be properly related not only to the transit improvement for workers but also to improvements in walk ability, transit service, and school bus for school trips. ing children to school is closely intertwined with scheduling of work and school activities. In the current research, certain simplified assumptions were made regarding possible schedule adjustments that workers could made to consolidate their schedules with school start and end times (30 min). In reality, these adjustments are highly individual and depend on the schedule flexibility. Understanding and modeling schedule consolidations within the household (of which escorting children to school is just one particular component) is yet another important layer of intrahousehold interactions that is currently missing in most models. 9