Skip to content

Destinations and stops

Location models of MAST over the full choice set of 1,588 zones, estimated on the survey estimation half. Each model has one attraction variable, entered as its own log term, β ln(1 + x), and ln zone area where a density term would otherwise be needed.

Reading the tables

Coefficient: maximum-likelihood estimate, with significance stars *** p < 0.01, ** p < 0.05, * p < 0.10; robust (sandwich) standard error; two-sided p-value. N: observations; K: estimated parameters; LL(β̂): log-likelihood at the estimates; ρ̄²: adjusted rho-square. Hold-out: fit on the held-out half of the data, which was also used in the specification search. "Reference": the omitted category; "—": removed from the model.

Main destination of discretionary tours

One logit for shopping, recreation, accompanying and other errands, from home, with a scale for each purpose. Level of service: piecewise car distance and car generalised time.

N = 14,341K = 62LL(β̂) = −49,427.4ρ̄² = 0.532Hold-out: LL per record −3.370
Variable Shopping Recreation Accompanying Other
Distance and impedance
Distance 0–1 km (per km) −3.8632*** (shared)
(0.2474) <0.001
Distance 1–2 km −1.0004*** (shared)
(0.0763) <0.001
Distance 2–5 km −0.5475*** (shared)
(0.0215) <0.001
Distance 5–10 km −0.2281*** (shared)
(0.0106) <0.001
Distance 10–20 km −0.1539*** (shared)
(0.0057) <0.001
Distance 20–50 km −0.0828*** (shared)
(0.0043) <0.001
Distance > 50 km 0 (fixed)
Car generalised time (min) −0.0390***a
(0.0023) <0.001
−0.0275***
(0.0025) <0.001
−0.0537***
(0.0091) <0.001
−0.0390***a
(0.0023) <0.001
Same zone
Intrazonal (destination = home zone) −1.0335***
(0.0604) <0.001
−0.5869***
(0.0714) <0.001
−0.2913***
(0.1125) 0.010
−0.6037***
(0.0835) <0.001
Area
Rest of Maryland reference
Baltimore City −1.0854***
(0.1177) <0.001
−0.1145
(0.1446) 0.428
−0.3444
(0.2655) 0.194
0.1178
(0.1564) 0.451
Washington DC 0.0738
(0.1472) 0.616
0.9144***
(0.1412) <0.001
0.8009***
(0.2191) <0.001
0.5635***
(0.1465) <0.001
High job-density zone −0.6292***
(0.0911) <0.001
0.6412***
(0.0983) <0.001
1.1656***
(0.1964) <0.001
0.5052***
(0.1203) <0.001
Virginia 0.2944***
(0.1028) 0.004
0.5959***
(0.1085) <0.001
0.8229***
(0.2493) <0.001
0.9775***
(0.1137) <0.001
Delaware, Pennsylvania, West Virginia 1.2452***
(0.1729) <0.001
0.2491
(0.2322) 0.283
0.3116
(0.7931) 0.694
0.4927*
(0.2585) 0.057
Land use
Ln(zone area, km²) 0.2463***
(0.0260) <0.001
0.4519***
(0.0288) <0.001
0.3969***
(0.0563) <0.001
0.3013***
(0.0325) <0.001
Land-use mix (entropy) −0.0247
(0.1154) 0.831
0.7290***
(0.1350) <0.001
1.2163***
(0.1991) <0.001
0.3604**
(0.1571) 0.022
Boundaries
Outside the origin county −0.5356*** (shared)
(0.0363) <0.001
Outside the origin state −0.6326*** (shared)
(0.0611) <0.001
Interactions with ln(distance)
Ln(distance) × cars ≥ drivers reference
Ln(distance) × no car −0.3730*** (shared)
(0.0533) <0.001
Ln(distance) × cars < drivers −0.1526*** (shared)
(0.0321) <0.001
Ln(distance) × person 18–64 reference
Ln(distance) × person < 18 −0.0027 (shared)
(0.0404) 0.947
Ln(distance) × person 65+ −0.0748** (shared)
(0.0298) 0.012
Ln(distance) × household with children −0.2069*** (shared)
(0.0295) <0.001
Ln(distance) × worker −0.0525* (shared)
(0.0277) 0.058
Ln(distance) × male −0.0071 (shared)
(0.0219) 0.746
Ln(distance) × one-person household −0.0822** (shared)
(0.0320) 0.010
Ln(distance) × midday 9–15 reference
Ln(distance) × AM 6–9 −0.2476*** (shared)
(0.0392) <0.001
Ln(distance) × PM 15–19 −0.1183*** (shared)
(0.0263) <0.001
Ln(distance) × evening 19–6 −0.0011 (shared)
(0.0348) 0.975
Ln(distance) × tour with at least one stop 0.4577*** (shared)
(0.0257) <0.001
Attraction
Ln(1 + retail jobs) 1.0496***
(0.0216) <0.001
0.4880***
(0.0253) <0.001
– 0.5222***
(0.0265) <0.001
Ln(1 + school enrolment) – – 0.1858***
(0.0208) <0.001
–
Scale
Scale μ_p (p against 1) 1 (normalised) 0.8252***
(0.0204) <0.001
0.8197***
(0.0367) <0.001
0.7449***
(0.0114) <0.001

Observations (estimation half): shopping 5,058; recreation 4,159; accompanying 1,593; other 3,531 tours; alternatives: 1,588 zones (full choice set); parameters: 62. LL(β̂) = −49,427.4; BIC = 99,448.2. Hold-out half (14,177 records, used before in the specification search): LL = −47,778.8. Utility of zone j for purpose p: Upj = μ_p Vpj, μ = 1 for shopping; a coefficient marked shared takes the same value in Vp of every purpose that has the variable, so its effect on utility differs across purposes only through the scale; a shared by shopping and other. –: not in the model of this purpose. Estimation: maximum likelihood on the full choice set, robust sandwich covariance, unweighted; estimation half of the Regional Travel Survey. Locations: the survey's traffic zones, spread over the model zones by housing units (home ends) or jobs (activity ends); the likelihood is averaged over the possible zones of the destination and of home (location mixture). Attraction: β_pln(1+xpj), xpj the retail jobs of zone j (school enrolment for accompanying). Zone area in km², at least 0.05 km². Distance: car distance from home (km), piecewise linear with the slope per km in each band; ln(distance) interactions with person, household, arrival-time and tour flags; high job-density zone: at least 10,000 jobs per km²; land-use mix: entropy of the job and population shares (0–1); car generalised time: car time plus cost over the household's value of time (min).

Each cell: coefficient with stars; below it, the robust standard error in brackets and the p-value.

Main destination: hold-out records by PUMA, observed and model, and model minus observed.

Main destination: hold-out records by PUMA, observed and model, and model minus observed.

Stop location: shopping

Location of an intermediate shopping stop by its detour between the previous location and the next anchor, with the logsum of the tour mode models over the detour.

N = 3,979K = 23LL(β̂) = −9,982.6ρ̄² = 0.659Hold-out: LL per record −2.527
Variable Coefficient Robust s.e. p
Detour and impedance
Detour 1–2 km −0.5052*** 0.0781 <0.001
Detour 2–5 km −0.2115*** 0.0341 <0.001
Detour 5–10 km −0.1436*** 0.0218 <0.001
Detour 10–20 km −0.1566*** 0.0163 <0.001
Detour > 20 km −0.0386*** 0.0084 <0.001
Mode-choice logsum, detour 0.4845*** 0.0686 <0.001
Car generalised time, detour (min) −0.0179*** 0.0069 0.009
Anchor zones
Stop in the zone of the previous location 0.3417*** 0.0584 <0.001
Stop in the zone of the next anchor 0.2971*** 0.0614 <0.001
Area
Rest of Maryland reference
Baltimore City −0.8146*** 0.1662 <0.001
Washington DC −0.3281** 0.1462 0.025
High job-density zone −0.2294** 0.1037 0.027
Virginia 0.1816 0.1522 0.233
Delaware, Pennsylvania, West Virginia 0.7625 0.5170 0.140
Land use
Ln(zone area, km²) 0.3688*** 0.0364 <0.001
Land-use mix (entropy) −0.4951*** 0.1372 <0.001
Boundaries
Outside the counties of both anchors −0.7378*** 0.0899 <0.001
Outside the states of both anchors −0.5983*** 0.1896 0.002
Interactions with ln(1 + detour)
Ln(1 + detour) × cars ≥ drivers reference
Ln(1 + detour) × no car −0.0797 0.1326 0.548
Ln(1 + detour) × cars < drivers −0.2765*** 0.0674 <0.001
Ln(1 + detour) × household size (persons) 0.0537*** 0.0168 0.001
Ln(1 + detour) × tour mode not car −0.5520*** 0.0900 <0.001
Attraction
Ln(1 + retail jobs) 0.9202*** 0.0268 <0.001

Observations: 3,979; alternatives: 1,588 zones (full choice set); parameters: 23. LL(β̂) = −9,982.6; BIC = 20,155.9. Hold-out half (3,846 records, used before in the specification search): LL = −9,719.2. Estimation: maximum likelihood on the full choice set, robust sandwich covariance, unweighted; estimation half of the Regional Travel Survey. Locations: the survey's traffic zones, spread over the model zones by housing units (home ends) or jobs (activity ends); the likelihood is averaged over the possible zones of the destination and of the anchors (location mixture). Attraction: βln(1+xj), xj the retail jobs of zone j. Zone area in km², at least 0.05 km². Dependent variable: chosen zone of the stop. Detour: d(a,j) + d(j,b) - d(a,b) by car (km) between the origin a of the trip to the stop and the next anchor b of the tour (the primary destination for an outbound stop, home for an inbound stop), piecewise linear with the slope per km in each band; interactions with person, household, tour and arrival-time flags; land-use mix: entropy of the job and population shares (0–1); logsum: mode-choice logsum of the detour; transit available: bus door-to-door time below 120 min.

Shopping stops: hold-out records by PUMA, observed and model, and model minus observed.

Shopping stops: hold-out records by PUMA, observed and model, and model minus observed.

Stop location: recreation

Location of an intermediate recreation stop.

N = 5,064K = 20LL(β̂) = −15,874.9ρ̄² = 0.574Hold-out: LL per record −3.205
Variable Coefficient Robust s.e. p
Detour and impedance
Detour 1–2 km −0.4963*** 0.0672 <0.001
Detour 2–5 km −0.1625*** 0.0277 <0.001
Detour 5–10 km −0.0884*** 0.0175 <0.001
Detour 10–20 km −0.1560*** 0.0097 <0.001
Detour > 20 km −0.0349*** 0.0029 <0.001
Mode-choice logsum, detour 1.2926*** 0.0813 <0.001
Anchor zones
Stop in the zone of the previous location 0.4821*** 0.0541 <0.001
Stop in the zone of the next anchor 0.3177*** 0.0559 <0.001
Area
Rest of Maryland reference
Baltimore City −0.1632 0.1098 0.137
Washington DC 0.5772*** 0.1036 <0.001
High job-density zone 0.8835*** 0.0688 <0.001
Virginia 0.1881* 0.1074 0.080
Delaware, Pennsylvania, West Virginia 0.8807*** 0.2259 <0.001
Land use
Ln(zone area, km²) 0.4465*** 0.0253 <0.001
Land-use mix (entropy) 0.0155 0.1075 0.886
Boundaries
Outside the counties of both anchors −0.8589*** 0.0698 <0.001
Outside the states of both anchors −0.4712*** 0.1235 <0.001
Interactions with ln(1 + detour)
Ln(1 + detour) × tour mode not car −0.3436*** 0.0491 <0.001
Ln(1 + detour) × inbound stop 0.1343*** 0.0300 <0.001
Attraction
Ln(1 + retail jobs) 0.4456*** 0.0190 <0.001

Observations: 5,064; alternatives: 1,588 zones (full choice set); parameters: 20. LL(β̂) = −15,874.9; BIC = 31,920.5. Hold-out half (5,184 records, used before in the specification search): LL = −16,616.0. Estimation: maximum likelihood on the full choice set, robust sandwich covariance, unweighted; estimation half of the Regional Travel Survey. Locations: the survey's traffic zones, spread over the model zones by housing units (home ends) or jobs (activity ends); the likelihood is averaged over the possible zones of the destination and of the anchors (location mixture). Attraction: βln(1+xj), xj the retail jobs of zone j. Zone area in km², at least 0.05 km². Dependent variable: chosen zone of the stop. Detour: d(a,j) + d(j,b) - d(a,b) by car (km) between the origin a of the trip to the stop and the next anchor b of the tour (the primary destination for an outbound stop, home for an inbound stop), piecewise linear with the slope per km in each band; interactions with person, household, tour and arrival-time flags; land-use mix: entropy of the job and population shares (0–1); logsum: mode-choice logsum of the detour; transit available: bus door-to-door time below 120 min.

Recreation stops: hold-out records by PUMA, observed and model, and model minus observed.

Recreation stops: hold-out records by PUMA, observed and model, and model minus observed.

Stop location: accompanying

Location of an intermediate drop-off or pick-up stop.

N = 3,515K = 22LL(β̂) = −12,488.1ρ̄² = 0.517Hold-out: LL per record −3.396
Variable Coefficient Robust s.e. p
Detour and impedance
Detour 0–1 km (per km) 0.4902*** 0.1173 <0.001
Detour 1–2 km −0.3241*** 0.0940 <0.001
Detour 2–5 km −0.1870*** 0.0307 <0.001
Detour 5–10 km −0.0920*** 0.0180 <0.001
Detour 10–20 km −0.1316*** 0.0098 <0.001
Detour > 20 km −0.0537*** 0.0039 <0.001
Mode-choice logsum, detour 1.0272*** 0.1208 <0.001
Anchor zones
Stop in the zone of the previous location 0.5685*** 0.0662 <0.001
Stop in the zone of the next anchor 0.3115*** 0.0690 <0.001
Area
Rest of Maryland reference
Baltimore City −0.2207* 0.1330 0.097
Washington DC −0.1387 0.1139 0.223
High job-density zone 1.1162*** 0.0972 <0.001
Virginia 0.2596** 0.1140 0.023
Delaware, Pennsylvania, West Virginia −0.1532 0.4161 0.713
Land use
Ln(zone area, km²) 0.5396*** 0.0290 <0.001
Land-use mix (entropy) 0.5783*** 0.1014 <0.001
Boundaries
Outside the counties of both anchors −1.0163*** 0.0791 <0.001
Outside the states of both anchors −0.1847 0.1359 0.174
Interactions with ln(1 + detour)
Ln(1 + detour) × cars ≥ drivers reference
Ln(1 + detour) × no car −0.0656 0.2623 0.802
Ln(1 + detour) × cars < drivers −0.1051** 0.0488 0.031
Ln(1 + detour) × tour mode not car −0.3201*** 0.0723 <0.001
Attraction
Ln(1 + school enrolment) 0.1108*** 0.0088 <0.001

Observations: 3,515; alternatives: 1,588 zones (full choice set); parameters: 22. LL(β̂) = −12,488.1; BIC = 25,155.7. Hold-out half (3,554 records, used before in the specification search): LL = −12,070.0. Estimation: maximum likelihood on the full choice set, robust sandwich covariance, unweighted; estimation half of the Regional Travel Survey. Locations: the survey's traffic zones, spread over the model zones by housing units (home ends) or jobs (activity ends); the likelihood is averaged over the possible zones of the destination and of the anchors (location mixture). Attraction: βln(1+xj), xj the school enrolment of zone j. Zone area in km², at least 0.05 km². Dependent variable: chosen zone of the stop. Detour: d(a,j) + d(j,b) - d(a,b) by car (km) between the origin a of the trip to the stop and the next anchor b of the tour (the primary destination for an outbound stop, home for an inbound stop), piecewise linear with the slope per km in each band; interactions with person, household, tour and arrival-time flags; land-use mix: entropy of the job and population shares (0–1); logsum: mode-choice logsum of the detour; transit available: bus door-to-door time below 120 min.

Accompanying stops: hold-out records by PUMA, observed and model, and model minus observed.

Accompanying stops: hold-out records by PUMA, observed and model, and model minus observed.

Stop location: other errands

Location of an intermediate stop for other errands.

N = 8,055K = 20LL(β̂) = −24,653.3ρ̄² = 0.584Hold-out: LL per record −3.028
Variable Coefficient Robust s.e. p
Detour and impedance
Detour 1–2 km −0.3645*** 0.0515 <0.001
Detour 2–5 km −0.2202*** 0.0221 <0.001
Detour 5–10 km −0.0761*** 0.0140 <0.001
Detour 10–20 km −0.1517*** 0.0083 <0.001
Detour > 20 km −0.0257*** 0.0043 <0.001
Mode-choice logsum, detour 1.5476*** 0.1019 <0.001
Car generalised time, detour (min) −0.0063* 0.0034 0.065
Anchor zones
Stop in the zone of the previous location 0.5396*** 0.0411 <0.001
Stop in the zone of the next anchor 0.1943*** 0.0430 <0.001
Area
Rest of Maryland reference
Baltimore City −0.0798 0.0867 0.358
Washington DC 0.0757 0.0887 0.394
High job-density zone 0.6557*** 0.0612 <0.001
Virginia 0.1931** 0.0909 0.034
Delaware, Pennsylvania, West Virginia −0.2740 0.2351 0.244
Land use
Ln(zone area, km²) 0.3882*** 0.0210 <0.001
Land-use mix (entropy) 0.2763*** 0.0874 0.002
Boundaries
Outside the counties of both anchors −1.0293*** 0.0586 <0.001
Outside the states of both anchors −0.2536** 0.1010 0.012
Interactions with ln(1 + detour)
Ln(1 + detour) × tour mode not car −0.3264*** 0.0458 <0.001
Attraction
Ln(1 + retail jobs) 0.4635*** 0.0147 <0.001

Observations: 8,055; alternatives: 1,588 zones (full choice set); parameters: 20. LL(β̂) = −24,653.3; BIC = 49,486.5. Hold-out half (8,010 records, used before in the specification search): LL = −24,256.6. Estimation: maximum likelihood on the full choice set, robust sandwich covariance, unweighted; estimation half of the Regional Travel Survey. Locations: the survey's traffic zones, spread over the model zones by housing units (home ends) or jobs (activity ends); the likelihood is averaged over the possible zones of the destination and of the anchors (location mixture). Attraction: βln(1+xj), xj the retail jobs of zone j. Zone area in km², at least 0.05 km². Dependent variable: chosen zone of the stop. Detour: d(a,j) + d(j,b) - d(a,b) by car (km) between the origin a of the trip to the stop and the next anchor b of the tour (the primary destination for an outbound stop, home for an inbound stop), piecewise linear with the slope per km in each band; interactions with person, household, tour and arrival-time flags; land-use mix: entropy of the job and population shares (0–1); logsum: mode-choice logsum of the detour; transit available: bus door-to-door time below 120 min.

Other-errand stops: hold-out records by PUMA, observed and model, and model minus observed.

Other-errand stops: hold-out records by PUMA, observed and model, and model minus observed.

Outing from the workplace: first stop

Location of the first stop of an outing that starts and ends at the workplace.

N = 1,180K = 10LL(β̂) = −3,112.3ρ̄² = 0.663Hold-out: LL per record −2.707
Variable Coefficient Robust s.e. p
Distance and impedance
Distance 0–1 km (per km) −5.9496*** 0.4628 <0.001
Distance 1–2 km −1.2190*** 0.2130 <0.001
Distance 2–5 km −0.3562*** 0.0810 <0.001
Distance 5–10 km −0.3237*** 0.0484 <0.001
Distance 10–20 km −0.1093*** 0.0384 0.004
Distance 20–50 km −0.0505* 0.0304 0.096
Car generalised time (min) −0.0691*** 0.0239 0.004
Same zone
Intrazonal (destination = workplace zone) −0.2151 0.1564 0.169
Boundaries
Outside the workplace county −1.3915*** 0.1828 <0.001
Attraction
Ln(1 + retail jobs) 0.6934*** 0.0476 <0.001

Observations: 1,180; alternatives: 1,588 zones (full choice set); parameters: 10. LL(β̂) = −3,112.3; BIC = 6,295.3. Hold-out half (1,155 records, used before in the specification search): LL = −3,126.1. Estimation: maximum likelihood on the full choice set, robust sandwich covariance, unweighted. Destinations: the survey's census tract, matched to the model zones by the area of their overlap. Estimation half of the Regional Travel Survey. Attraction: βln(1+xj), xj the retail jobs of zone j. Same form and estimator as the other location models with the usual workplace as the anchor: distance from the workplace (km), piecewise linear with the slope per km in each band.

Outing first stops: hold-out records by PUMA, observed and model, and model minus observed.

Outing first stops: hold-out records by PUMA, observed and model, and model minus observed.

Work business: workers with a usual workplace

Location of work-business episodes, with home and workplace as anchors.

N = 1,539K = 18LL(β̂) = −8,432.1ρ̄² = 0.310Hold-out: LL per record −5.554
Variable Coefficient Robust s.e. p
Detour and impedance
Detour 0–1 km (per km) 0.8718*** 0.1929 <0.001
Detour 1–2 km −0.7337*** 0.1747 <0.001
Detour 2–5 km −0.0933 0.0604 0.123
Detour 5–10 km −0.0145 0.0316 0.646
Detour 10–20 km −0.0719*** 0.0132 <0.001
Detour > 20 km −0.0178*** 0.0047 <0.001
Car generalised time, detour (min) −0.0088** 0.0042 0.037
Same zone
Stop in the home zone 1.1300*** 0.1225 <0.001
Stop in the workplace zone −0.5239*** 0.1640 0.001
Land use
Ln(zone area, km²) 0.1646*** 0.0290 <0.001
Land-use mix (entropy) 0.2989 0.1949 0.125
Boundaries
Outside the counties of both anchors −0.9945*** 0.0819 <0.001
Interactions with ln(1 + detour)
Ln(1 + detour) × cars ≥ drivers reference
Ln(1 + detour) × no car −0.1520 0.1165 0.192
Ln(1 + detour) × cars < drivers −0.2870*** 0.0712 <0.001
Ln(1 + detour) × midday 9–15 reference
Ln(1 + detour) × AM 6–9 0.0403 0.0528 0.445
Ln(1 + detour) × PM 15–19 −0.2480*** 0.0582 <0.001
Ln(1 + detour) × evening 19–6 −0.2369** 0.0994 0.017
Attraction
Ln(1 + office jobs) 0.5580*** 0.0251 <0.001

Observations: 1,539; alternatives: 1,588 zones (full choice set); parameters: 18. LL(β̂) = −8,432.1; BIC = 16,996.3. Hold-out half (1,529 records, used before in the specification search): LL = −8,492.8. Estimation: maximum likelihood on the full choice set, robust sandwich covariance, unweighted. Destinations: the survey's census tract, matched to the model zones by the area of their overlap. Estimation half of the Regional Travel Survey. Attraction: βln(1+xj), xj the office jobs of zone j. Zone area in km², at least 0.05 km². Detour: d(h,j) + d(j,w) - d(h,w) by car between home h and the usual workplace w (km), piecewise linear with the slope per km in each band; interactions with household and arrival-time flags; land-use mix: entropy of the job and population shares (0–1); logsum: mode-choice logsum of the stop.

Work business, usual-workplace workers: hold-out records by PUMA, observed and model, and model minus observed.

Work business, usual-workplace workers: hold-out records by PUMA, observed and model, and model minus observed.

Work business: other persons

Location of work-business episodes of persons without a usual workplace.

N = 1,259K = 16LL(β̂) = −6,819.3ρ̄² = 0.292Hold-out: LL per record −5.625
Variable Coefficient Robust s.e. p
Distance and impedance
Distance 1–2 km −2.1253*** 0.2528 <0.001
Distance 2–5 km −0.1183 0.0814 0.146
Distance 5–10 km −0.1706*** 0.0325 <0.001
Distance 10–20 km −0.1145*** 0.0138 <0.001
Distance 20–50 km −0.0811*** 0.0054 <0.001
Distance > 50 km −0.0392*** 0.0033 <0.001
Same zone
Intrazonal (destination = home zone) 0.7504*** 0.1792 <0.001
Area
Rest of Maryland reference
Baltimore City −0.2364** 0.0983 0.016
Washington DC 0.3742*** 0.1269 0.003
Virginia 0.2043* 0.1197 0.088
Delaware, Pennsylvania, West Virginia −0.2775 0.2094 0.185
Boundaries
Outside the origin county −0.3124*** 0.0861 <0.001
Interactions with ln(distance)
Ln(distance) × income $50–100k reference
Ln(distance) × income < $50k −0.1206* 0.0664 0.069
Ln(distance) × income ≥ $100k 0.1157* 0.0634 0.068
Ln(distance) × worker 0.3775*** 0.0556 <0.001
Attraction
Ln(1 + office jobs) 0.5440*** 0.0239 <0.001

Observations: 1,259; alternatives: 1,588 zones (full choice set); parameters: 16. LL(β̂) = −6,819.3; BIC = 13,752.9. Hold-out half (1,348 records, used before in the specification search): LL = −7,583.1. Estimation: maximum likelihood on the full choice set, robust sandwich covariance, unweighted. Destinations: the survey's census tract, matched to the model zones by the area of their overlap. Estimation half of the Regional Travel Survey. Attraction: βln(1+xj), xj the office jobs of zone j. Distance: car distance from the home (km), piecewise linear with the slope per km in each band; ln(distance) interactions with person, household and arrival-time flags; land-use mix: entropy of the job and population shares (0–1).

Work business, other persons: hold-out records by PUMA, observed and model, and model minus observed.

Work business, other persons: hold-out records by PUMA, observed and model, and model minus observed.

Work trips of persons who work at home

Location of the work trips of persons whose usual workplace is home.

N = 366K = 5LL(β̂) = −1,640.9ρ̄² = 0.430Hold-out: LL per record −4.389
Variable Coefficient Robust s.e. p
Distance and impedance
Distance (km) −0.0183*** 0.0050 <0.001
Ln(distance, km) −1.4001*** 0.1388 <0.001
Same zone
Intrazonal (destination = home zone) 1.3040*** 0.2615 <0.001
Boundaries
Outside the origin county −0.4580** 0.1903 0.016
Attraction
Ln(1 + office jobs) 0.5299*** 0.0466 <0.001

Observations: 366; alternatives: 1,588 zones (full choice set); parameters: 5. LL(β̂) = −1,640.9; BIC = 3,311.3. Hold-out half (427 records, used before in the specification search): LL = −1,874.0. Estimation: maximum likelihood on the full choice set, robust sandwich covariance, unweighted. Destinations: the survey's census tract, matched to the model zones by the area of their overlap. Estimation half of the Regional Travel Survey. Attraction: βln(1+xj), xj the office jobs of zone j. Distance: car distance from home (km).

Work trips of home workers: hold-out records by PUMA, observed and model, and model minus observed.

Work trips of home workers: hold-out records by PUMA, observed and model, and model minus observed.

School and daycare without a school record

One pooled model for school episodes of persons without a school record and daycare of children under 5, with a purpose scale.

N = 846K = 9LL(β̂) = −3,798.5ρ̄² = 0.440Hold-out: LL per record −4.396
Variable Non-student school Daycare
Distance and impedance
Ln(distance, km) −1.0600*** (shared)
(0.0618) <0.001
Car generalised time (min) −0.0447*** (shared)
(0.0043) <0.001
Land use
Ln(zone area, km²) 0.1238*** (shared)
(0.0463) 0.007
Land-use mix (entropy) 1.2281*** (shared)
(0.2069) <0.001
Boundaries
Outside the origin county −0.5883*** (shared)
(0.1177) <0.001
Outside the origin state −0.4893*** (shared)
(0.1660) 0.003
Attraction
Ln(1 + school enrolment) 0.5437***
(0.0814) <0.001
0.1261***
(0.0334) <0.001
Scale
Scale (p against 1) 1 (normalised) 1.2389***
(0.0453) <0.001

Observations (estimation half): non-student school 567, daycare 279 episodes; alternatives: 1,588 zones (full choice set); parameters: 9. LL(β̂) = −3,798.5; BIC = 7,657.6. Hold-out half (839 records: non-student school 557, daycare 282; used before in the specification search): LL = −3,688.3. Utility of zone j for purpose p: Upj = μ_p Vpj, μ = 1 for non-student school; a coefficient marked shared takes the same value in Vp of both purposes. –: not in the model of this purpose. Estimation: maximum likelihood on the full choice set, robust sandwich covariance, unweighted. Destinations: the survey's census tract, matched to the model zones by the area of their overlap. Estimation half of the Regional Travel Survey. Attraction: β_pln(1+xj), xj the school enrolment of zone j, one coefficient per purpose. Zone area in km², at least 0.05 km². Distance: car distance from home (km); land-use mix: entropy of the job and population shares (0–1).

Each cell: coefficient with stars; below it, the robust standard error in brackets and the p-value.

School and daycare without a record: hold-out records by PUMA, observed and model, and model minus observed.

School and daycare without a record: hold-out records by PUMA, observed and model, and model minus observed.