Longitudinal Analysis of Activity Generation in Greater Toronto and Hamilton Area (GTHA)

TTS 1996 TTS 2006 TTS 2016 TTS 2001 TTS 2011 Longitudinal Analysis of Activity Generation in the GTHA TMG Meeting November 7, 2018 Dr. Eric J. Miller & Gozde Ozonder

Outline Objective Data Methodology Summary of Analysis Results & Discussion Activity Types: Work School Marketing/Shopping Other

WARNING! WARNING!

Objective Analyze activity generation behaviour in the GTHA over a 20-year period Take advantage of multiple surveys Benefit from the utility concept Develop policy-sensitive models Try to capture the extent of survey bias Improve TASHA s activity generation module

TTS Data 1996 2001 2006 2011 2016 Data Cleansing Aim: obtain data sets that can be effectively used in activity/travel- related research Tool: rule-based code in RStudio Tasks: Dealing with inconsistencies Missing individuals, missing respondents, no W but R trips, etc. Removing some unknowns Employment status unknown, sex unknown, etc. Filtering out unusual behaviour 12 year-old full-time worker (with a regular work place) in Manufacturing sector Post-Cleansing Statistics: (Incomplete HHs) (% of initial data points conserved) Households: 552,839 (99%) Persons: 1,272,174 (99%) Trips: 2,969,374 (98%) Trip Makers: 1,006,775 (99%) Trip Making Households: 493,066 (99%)

Methodology For each activity type: Estimate year- specific models & assess the outcomes (goodness of fit, signs, order of magnitudes, etc.) Estimate four* types of joint models, investigate the scaled estimates & evaluate the results Define set(s) of choice alternatives (binomial, trinomial, etc.) Select model structure(s) to be tested Choose levels/sectors/ categories *: NP, NPDA, JOSI, JO

Work Work Model structures: Logit (multinomial, binomial logit, hierarchical) Sets of choice alternatives: {A, B, C} {0,1} {A} & {B,C} Sectors: Sector G Sector M Sector P Sector S Categories: Full-time Part-time Year-specific & joint models Population: F: full-time workers with a regular office location P: part-time workers with a regular office location H: full-time workers at home J: part-time workers at home

WORK WORK Specific Models WIM - Estimates WIG - Estimates Year Year- -Specific Models Parameter Estimates & Discussion 10.00 10.00 8.00 8.00 By looking at the estimates two main conclusions can be drawn for all four sectors: Most variables (trip day of week, sex, count of children in household, etc.) have stable parameter estimates over the years (20 year period). This indicates that the behaviour can be forecast confidently. Dummy for work-at-home employment status seems to have varying parameter estimates over the fluctuation in propensity for working at home. However, the sum of the intercept for choice B and dummy for work-at-home seems to be stable over time. 6.00 6.00 4.00 4.00 2.00 2.00 0.00 0.00 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -2.00 -2.00 -4.00 -4.00 -6.00 -6.00 -8.00 -8.00 years. Statistical B:(intercept) C:(intercept) B:(intercept) C:(intercept) Count_CHI_C dumF_C1 dumF_C1 dumHorJ_B1 dumHorJ_B1 ForP_dist_C ForP_dist_C ForP_TripDayOfWeek_F_C ForP_TripDayOfWeek_F_C PorJ_stu_S_C PorJ_stu_P_C PorJ_stu_S_C Sex_M_C1 It can be concluded that (through pairwise t- tests, goodness of fit tests, etc.) it is essential to model the four occupational groups separately, as they have different explanatory variables, different estimates. Sex_M_B1 WIP - Estimates WIS - Estimates 10.00 10.00 8.00 8.00 6.00 6.00 Fluctuation increasing/decreasing trend is observed over the years across the sectors. However the range of the values taken by the estimates are very narrow in most cases. Hence, not observing a clear pattern might not be an indicator of the changes in travel behaviour, this might be a side- product of having independent survey samples each year. of estimates No strictly 4.00 4.00 2.00 2.00 0.00 0.00 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -2.00 -2.00 -4.00 -4.00 -6.00 -6.00 -8.00 -8.00 B:(intercept) C:(intercept) B:(intercept) C:(intercept) Count_CHI_C Survey respondent, survey method (phone, online)) are not found to have significant impacts on choice probabilities! instruments (respondent/non- Count_CHI_C dumF_C1 dumF_C1 dumHorJ_B1 PorJ_stu_P_C dumHorJ_B1 ForP_dist_C PorJ_stu_S_C Sex_M_B1 Sex_M_C1 ForP_Loc_EmpZone_PD1_C1 ForP_TripDayOfWeek_F_C PorJ_stu_S_C Sex_M_C1

On average we are able to explain 30% of variability with these models. When different sectors are compared, it is seen that Manufacturing sector has the lowest fit due to its highly heterogeneous nature. Evaluating the trends in goodness of fit of work models, it can be underlined that we are able to explain more of the variability as we move from past to present: Is the survey picking more homogeneous people (selectivity bias)? Are people getting more & more systematic, displaying less variability due to increased stress, congestion, etc.? BOTH? WORK Year-Specific Models Comparison of Goodness of Fit 2against const. - across sectors 2 - WIG 2 - WIM 0.65 0.65 0.65 0.64 0.63 0.70 0.63 0.70 0.70 0.61 0.61 0.61 0.59 0.60 0.60 0.60 0.50 0.50 0.50 0.40 0.36 0.40 0.33 0.40 0.40 0.26 0.24 0.30 0.23 0.30 0.30 0.22 0.21 0.17 0.17 0.20 0.20 0.20 0.10 0.10 0.10 0.00 0.00 0.00 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 G M P S 2 against const. Sector G 2 against null 2 against const. 2 against null Sector M 2 - WIS 2 - WIP 2against null - across sectors 0.70 0.66 0.70 0.61 0.70 0.60 0.58 0.57 0.54 0.53 0.60 0.53 0.52 0.50 0.60 0.60 0.50 0.50 0.50 0.37 0.38 0.35 0.36 0.40 0.32 0.40 0.32 0.40 0.25 0.24 0.30 0.30 0.21 0.30 0.19 0.20 0.20 0.20 0.10 0.10 0.10 0.00 0.00 0.00 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 2 against const. Sector P 2 against null 2 against const. Sector S 2 against null G M P S

WORK Joint Models In all four types of joint models estimated, parameter estimates are close to the values obtained from year-specific models. Almost the same goodness of fit values are obtained from four different joint models in each sector. Models of sectors G and S perform better ( 2>0.31) than the models of sector P ( 2~0.29), and the models of sector M have the lowest fit ( 2~0.19). All the year-specific scales are close to 1 (the highest value < 1.5). We observe a slightly upward trend in scales over the years in all sectors (though with exceptions, there are slight fluctuations), meaning variance is getting lower. Either people s behaviour is getting more systematic, less random over time, or, surveys are starting to pick similar people. NP -3.83 NPDA JOSI -3.76 JO NP -2.67 NPDA JOSI -2.60 JO NP NPDA JOSI -2.94 JO NP NPDA JOSI -3.76 JO B:(intercept) B:(intercept) - - B:(intercept) -3.08 - - B:(intercept) -3.88 - - int_96_B int_01_B int_06_B int_11_B int_16_B -3.73 -3.45 -4.36 -4.07 -3.77 -3.54 -3.31 -4.18 -3.93 -3.65 int_96_B int_01_B int_06_B int_11_B int_16_B - - - - -2.65 -2.50 -3.10 -2.91 -2.61 - - - - -2.41 -2.26 -3.04 -2.98 -2.70 int_96_B int_01_B int_06_B int_11_B int_16_B - - - - - -3.22 -2.88 -3.43 -3.24 -2.86 - - - - - -2.71 -2.37 -3.25 -3.14 -2.85 int_96_B int_01_B int_06_B int_11_B int_16_B - - - - - -3.72 -3.76 -4.20 -3.93 -3.74 - - - - - -3.37 -3.42 -3.87 -3.67 -3.50 C:(intercept) 0.89 0.87 C:(intercept) 1.52 - 1.46 - C:(intercept) 1.01 - 0.94 - C:(intercept) 0.81 - 0.78 - int_96_C int_01_C 0.89 0.91 0.95 1.06 int_96_C int_01_C - - 1.48 1.54 - - 1.56 1.65 int_96_C int_01_C int_06_C - - - 0.96 1.04 0.99 - - - 1.15 1.31 0.66 int_96_C int_01_C - - 0.88 0.90 - - 0.91 0.90 int_06_C 0.93 0.78 int_06_C - 1.52 - 1.16 int_06_C - 0.87 - 0.76 int_11_C - 0.91 - 0.58 int_11_C int_16_C 0.80 0.93 1.42 6.93 -1.35 -0.10 0.13 -0.41 0.00 0.66 0.77 1.30 6.45 -1.29 -0.09 0.12 -0.38 0.00 0.94 1.16 1.16 1.17 int_11_C int_16_C - - 1.53 1.54 0.76 5.38 -0.92 -1.81 -0.97 -0.40 0.00 - - 1.06 1.09 0.63 4.99 -0.82 -1.65 -0.88 -0.34 0.00 0.98 1.30 1.40 1.38 int_11_C int_16_C - - 0.73 0.76 0.90 6.45 -0.64 -1.59 -0.07 -0.64 0.20 - - 0.61 0.63 0.79 5.86 -0.58 -1.44 -0.06 -0.57 0.18 1.00 1.15 1.18 1.19 int_16_C - 1.13 1.14 5.57 -1.36 -0.08 0.28 -0.40 -0.01 - 0.69 0.89 4.85 -1.09 -0.06 0.22 -0.32 -0.01 0.95 1.41 1.44 1.49 dumF_C1 dumHorJ_B1 PorJ_stu_S_C1 Count_CHI_C Sex_M_C1 ForP_TripDayOfWeek_F_C1 ForP_dist_C mu_01 mu_06 mu_11 mu_16 2 against const. 2 against null. LL model LL against const. LL against null 1.42 6.78 -1.34 -0.10 0.13 -0.41 0.00 1.38 6.60 -1.31 -0.10 0.12 -0.40 0.00 0.97 1.07 1.02 1.09 dumF_C1 dumHorJ_B1 PorJ_stu_P_C1 PorJ_stu_S_C1 Sex_M_B1 ForP_TripDayOfWeek_F_C1 ForP_dist_C mu_01 mu_06 mu_11 mu_16 2 against const. 2 against null. LL model LL against const. LL against null 0.76 5.32 -0.92 -1.81 -1.02 -0.40 0.00 0.73 5.14 -0.88 -1.74 -0.98 -0.38 0.00 1.01 1.08 1.09 1.10 dumF_C1 dumHorJ_B1 PorJ_stu_S_C1 Count_CHI_C Sex_M_C1 ForP_TripDayOfWeek_F_C1 ForP_dist_C mu_01 mu_06 mu_11 mu_16 2 against const. 2 against null. LL model LL against const. LL against null 1.15 5.53 -1.35 -0.08 0.28 -0.40 -0.01 1.07 5.22 -1.25 -0.07 0.26 -0.37 -0.01 0.99 1.11 1.09 1.20 dumF_C1 dumHorJ_B1 PorJ_stu_P_C1 PorJ_stu_S_C1 Count_CHI_C Sex_M_B1 Sex_M_C1 mu_01 mu_06 mu_11 mu_16 2 against const. 2 against null. LL model LL against const. LL against null 0.90 6.40 -0.63 -1.57 -0.06 -0.65 0.20 0.86 6.17 -0.60 -1.50 -0.06 -0.63 0.19 0.99 1.07 1.05 1.08 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 0.3181 0.3190 0.3144 0.3159 0.1971 0.1980 0.1976 0.2001 0.2908 0.2919 0.2907 0.2945 0.3154 0.3165 0.3129 0.3146 0.6249 -40938 -60032 -109138 0.6254 -40882 -60032 -109138 G 0.6253 -40898 -59653 -109138 0.6261 -40811 -59653 -109138 0.6201 -43560 -54253 -114661 0.6205 -43510 -54253 -114661 0.6205 -43519 -54234 -114661 0.6217 -43380 -54234 -114661 0.6010 -123453 -174079 -309432 0.6016 -123266 -174079 -309432 0.6023 -123059 -173483 -309432 0.6045 -122387 -173483 -309432 0.5226 -98540 -143938 -206391 0.5233 -98387 -143938 -206391 0.5227 -98501 -143347 -206391 0.5239 -98255 -143347 -206391 M P S

2against const. 2against null. 0.7000 0.7000 0.6000 0.6000 0.5000 0.5000 0.4000 0.4000 0.3000 0.3000 0.2000 0.2000 0.1000 0.1000 0.0000 0.0000 WORK WORK Joint Models Joint Models Goodness of Fit NP NPDA JOSI JO NP NPDA JOSI JO S P M G S P M G Log-Likelihood AIC 0 300000 NP NPDA JOSI JO -20000 250000 -40000 200000 -60000 150000 -80000 100000 -100000 50000 -120000 0 -140000 NP NPDA JOSI JO S P M G S P M G

* Scaled JO Estimates - WIG Scaled JOSI Estimates - WIG 8.00 8.00 6.00 6.00 4.00 4.00 Joint Models WORK Joint Models 2.00 2.00 the variation of estimates show a similar pattern to the variation of estimates in year-specific models (for JO model parameters). 0.00 0.00 When the parameter estimates from joint models are scaled: Sc'd JO 96 Sc'd JO 01 Sc'd JO 06 Sc'd JO 11 Sc'd JO 16 Sc'd JOSI 96 Sc'd JOSI 01 Sc'd JOSI 06 Sc'd JOSI 11 Sc'd JOSI 16 Scaled Estimates -2.00 -2.00 -4.00 -4.00 -6.00 -6.00 B:(intercept) C:(intercept) B:(intercept) C:(intercept) dumF_C1 dumHorJ_B1 dumF_C1 dumHorJ_B1 PorJ_stu_S_C1 Count_CHI_C PorJ_stu_S_C1 Count_CHI_C Sex_M_C1 ForP_TripDayOfWeek_F_C1 Sex_M_C1 ForP_TripDayOfWeek_F_C1 ForP_dist_C ForP_dist_C Scaled JO Estimates - WIM Scaled JOSI Estimates - WIM 8.00 8.00 6.00 6.00 4.00 4.00 WORK 2.00 2.00 0.00 0.00 Sc'd JO 96 Sc'd JO 01 Sc'd JO 06 Sc'd JO 11 Sc'd JO 16 Sc'd JOSI 96 Sc'd JOSI 01 Sc'd JOSI 06 Sc'd JOSI 11 Sc'd JOSI 16 -2.00 -2.00 -4.00 -4.00 -6.00 -6.00 B:(intercept) C:(intercept) B:(intercept) C:(intercept) dumF_C1 dumHorJ_B1 dumF_C1 dumHorJ_B1 PorJ_stu_P_C1 PorJ_stu_S_C1 PorJ_stu_P_C1 PorJ_stu_S_C1 Sex_M_B1 ForP_TripDayOfWeek_F_C1 Sex_M_B1 ForP_TripDayOfWeek_F_C1 ForP_dist_C ForP_dist_C

* Scaled JOSI Estimates - WIP Scaled JO Estimates - WIP 8.00 8.00 information that is not captured by the models varies from year to year. This 6.00 6.00 WORK Joint Models Joint Models 4.00 4.00 No clear trend in year-specific intercepts of the models: The portion of 2.00 2.00 0.00 0.00 Scaled Estimates Sc'd JOSI 96 Sc'd JOSI 01 Sc'd JOSI 06 Sc'd JOSI 11 Sc'd JOSI 16 Sc'd JO 96 Sc'd JO 01 Sc'd JO 06 Sc'd JO 11 Sc'd JO 16 -2.00 -2.00 -4.00 -4.00 -6.00 -6.00 might, again, be attributed to the survey characteristics. B:(intercept) C:(intercept) B:(intercept) C:(intercept) dumF_C1 dumHorJ_B1 dumF_C1 dumHorJ_B1 PorJ_stu_S_C1 Count_CHI_C PorJ_stu_S_C1 Count_CHI_C Sex_M_C1 ForP_TripDayOfWeek_F_C1 Sex_M_C1 ForP_TripDayOfWeek_F_C1 ForP_dist_C ForP_dist_C Scaled JOSI Estimates - WIS Scaled JO Estimates - WIS 8.00 8.00 6.00 6.00 WORK 4.00 4.00 2.00 2.00 0.00 0.00 Sc'd JOSI 96 Sc'd JOSI 01 Sc'd JOSI 06 Sc'd JOSI 11 Sc'd JOSI 16 Sc'd JO 96 Sc'd JO 01 Sc'd JO 06 Sc'd JO 11 Sc'd JO 16 -2.00 -2.00 -4.00 -4.00 -6.00 -6.00 B:(intercept) C:(intercept) dumF_C1 B:(intercept) C:(intercept) dumF_C1 dumHorJ_B1 PorJ_stu_P_C1 PorJ_stu_S_C1 dumHorJ_B1 PorJ_stu_P_C1 PorJ_stu_S_C1 Count_CHI_C Sex_M_B1 Sex_M_C1 Count_CHI_C Sex_M_B1 Sex_M_C1

WORK WORK Coefficient of Variation G - COV M - COV 0.300 0.300 0.200 0.200 0.100 0.100 0.000 0.000 C.O.V.-JOSI C.O.V.-JO C.O.V.-JOSI C.O.V.-JO -0.100 -0.100 -0.200 -0.200 -0.300 -0.300 B:(intercept) C:(intercept) Other Variables (+) Other Variables (-) B:(intercept) C:(intercept) Other Variables (+) Other Variables (-) Very low values!!! P - COV S - COV 0.300 0.300 0.200 0.200 0.100 0.100 0.000 0.000 C.O.V.-JOSI C.O.V.-JO C.O.V.-JOSI C.O.V.-JO -0.100 -0.100 -0.200 -0.200 -0.300 -0.300 B:(intercept) C:(intercept) Other Variables (+) Other Variables (-) B:(intercept) C:(intercept) Other Variables (+) Other Variables (-)

Likelihood Ratio Tests Log-Likelihood TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 Sum Sector S Sector P Sector M Sector G -14072.1 -21643.9 -16882.9 -32882.9 -25163.4 -21710.9 -25301.5 -20259.5 -16630.4 -25566.2 -98050 -122064 ISU -9488 -11139 -129240 -108580 -188030 -148670 -10875 -130730 -109150 -181590 -144490 -10042 -142360 -117670 -187070 -148840 -9488 -131060 -108500 -171490 -137840 -51032 -629868 -525261 -868670 -691260 MI - 3 CH MI - 2 CH OI - 3 CH OI - 2 CH -96478 -81361 -140490 -111420 -10279 -13153 -8280 -6305 -5278 -43295 -6912 -8604 -8211 -9339 -7663 -40728 Log-Likelihood NP NPDA JOSI JO Sector S Sector P Sector M Sector G -98540 -123453 -98387 -123266 -98501 -123059 -98255 -122387 ISU -51398 -630840 -526110 -871310 -693700 -51135 -630630 -525940 -870580 -693070 -51289 -630643 -525883 -870532 -693025 -51117 -630561 -525856 -870492 -692998 MI - 3 CH MI - 2 CH OI - 3 CH OI - 2 CH -43560 -43510 -43519 -43380 -40938 -40882 -40898 -40811 LRT Sector S Sector P Sector M Sector G 978 2779 674 2404 901 1990 410 646 ISU MI - 3 CH MI - 2 CH OI - 3 CH OI - 2 CH 1944 1698 1524 1358 1550 1243 1386 1190 2*LL(joint-NP)-2*LL( total year specific) 2*LL(joint-NPDA)-2*LL( total year specific) 2*LL(joint-JOSI)-2*LL( total year specific) 2*LL(joint-JO)-2*LL( total year specific) 531 430 449 170 418 308 339 165 733 207 514 170 5280 3820 3724 3643 4880 3620 3530 3477 In all models null hypothesis is rejected at all confidence levels (even at 0.995)! (Highest value in Chi-Square Distribution Table (with 100 degrees of freedom) is 140, here the minimum value is 165!) In our case, degrees of freedom are in the orders of 20. REJECT THE NULL HYPOTHESIS THAT JOINT MODEL FITS BETTER THAN YEAR-SPECIFIC MODEL!

School School Model structure: Logit (binomial) Set of choice alternatives: {0,1} Levels: Elementary Secondary University + Graduate Year-specific & joint models

SCHOOL SCHOOL Year Parameter Estimates Estimates ISU_1996 1:(intercept) 0.00 1:StudentStatus_S 1.61 1:dumF -0.69 1:OccupationG -0.40 1:OccupationM -0.32 1:OccupationP -0.51 1:OccupationS -0.27 1:TripDayOfWeek_F -0.66 1:DistHHSchZone -0.02 1:SexM 0.16 1:Count_0_1 -0.27 1:Age_18_21 0.57 1:Age_22_25 0.17 1:Age_26_30 - 2 against const. 0.25 2 against null. 0.29 LL model -9488 LL against const. -12603 LL against null -13367 Year- -Specific Models Specific Models ISU - Estimates 2.00 ISU_2001 ISU_2006 ISU_2011 ISU_2016 0.06 1.52 -0.70 -0.43 -0.30 -0.57 -0.21 -0.64 -0.02 0.10 -0.30 0.47 -0.11 1.41 -0.61 -0.54 -0.47 -0.84 -0.33 -0.63 -0.01 0.13 -0.23 0.55 0.24 0.21 -0.40 1.50 -0.59 -0.47 -0.35 -0.70 -0.37 -0.58 -0.01 0.13 -0.33 0.52 0.19 0.17 -0.31 1.42 -0.86 -0.43 -0.19 -0.63 -0.19 -0.49 -0.01 0.12 -0.29 0.16 1.50 1.00 0.50 0.00 ISU_1996 ISU_2001 ISU_2006 ISU_2011 ISU_2016 - - - - -0.50 0.22 0.20 0.18 0.16 -1.00 0.27 0.23 0.20 0.17 -9488 -11257 -11390 -11139 -14347 -15232 -10875 -13594 -14151 -10042 -12249 -12615 1:(intercept) 1:StudentStatus_S 1:dumF 1:OccupationG 1:OccupationM 1:OccupationP 1:OccupationS 1:TripDayOfWeek_F 1:DistHHSchZone 1:SexM 1:Count_0_1 1:Age_18_21 ISU - 2 0.35 For elementary school and secondary school, it is the intercept of the model that mainly defines the choice probabilities, rather than the explanatory variables. The parameter estimates remain relatively stable over time, indicating that the models can be used for forecasting. In contrast to what we observed in work models, goodness of fit of the university and graduate level models improve as we go back in time, which might indicate that either the behaviour is getting more random as we move from past to present, or, the survey is perhaps not doing a good job at picking a representative sample for this group. Survey effects are not found to have significant impacts on choice probabilities. 0.3 0.25 0.2 0.15 0.1 0.05 0 ISU_1996 ISU_2001 ISU_2006 ISU_2011 ISU_2016 2 against const. 2 against null.

Scaled Estimates - JOSI - ISU 2.00 SCHOOL SCHOOL Joint Models Joint Models 1.50 1.00 0.50 0.00 Sc'd JOSI_96 Sc'd JOSI_01 Sc'd JOSI_06 Sc'd JOSI_11 Sc'd JOSI_16 The parameter estimates obtained from different types of joint models have nearly the same values, which might enable us to choose one structure and use that for further analyses. There is a decreasing trend in the year- specific intercepts of the joint models (NPDA and JOSI). This suggests that university & graduate students are getting less likely to go to school on a random day (since not going to school is the base alternative). All year-specific scales are less than 1, and there is a decreasing trend over the years, going from past to present, which might suggest that the variation is getting higher in time. The goodness of fit obtained from four types of joint models are almost the same ( 2~0.20) , as we observed in the case of work models. The decreasing trend is continued to be observed in the intercepts when the estimates are scaled. -0.50 -1.00 ISU_NP ISU_NPDA ISU_JOSI ISU_JO 1:(intercept) -0.09 - -0.10 - 1:(intercept) 1:StudentStatus_S 1:dumF int_96_1 int_01_1 int_06_1 int_11_1 int_16_1 - - - - - 0.10 0.05 -0.12 -0.27 -0.38 1.53 -0.70 -0.44 -0.33 -0.65 -0.26 -0.60 -0.01 - - - - - 0.07 0.05 -0.13 -0.29 -0.37 1.62 -0.73 -0.27 -0.69 -0.34 -0.47 -0.64 -0.01 1:OccupationG 1:OccupationM 1:OccupationP 1:OccupationS 1:TripDayOfWeek_F 1:DistHHSchZone 1:SexM 1:Count_0_1 1:Age_18_21 * 1:StudentStatus_S 1:dumF 1:OccupationG 1:OccupationM 1:OccupationP 1:OccupationS 1:TripDayOfWeek_F 1:DistHHSchZone 1.52 -0.69 -0.42 -0.27 -0.64 -0.27 -0.60 -0.01 1.69 -0.75 -0.30 -0.70 -0.30 -0.48 -0.68 -0.02 Scaled Estimates - JO - ISU 2.00 1.50 1.00 0.50 0.00 1:SexM 0.13 0.13 0.14 0.14 Sc'd JO_96 Sc'd JO_01 Sc'd JO_06 Sc'd JO_11 Sc'd JO_16 -0.50 1:Count_0_1 1:Age_18_21 1:Age_22_25 1:Age_26_30 -0.28 0.39 -0.30 0.38 -0.32 0.45 -0.31 0.41 -1.00 - - - - - - - - - - - - - - - - 1:(intercept) 1:StudentStatus_S 1:dumF 1:OccupationG 1:OccupationM 1:OccupationP mu_01 mu_06 mu_11 mu_16 0.95 0.89 0.84 0.74 0.95 0.94 0.95 0.86 1:OccupationS 1:TripDayOfWeek_F 1:DistHHSchZone 1:SexM 1:Count_0_1 1:Age_18_21 2 against const. 2 against null. LL model LL against const. LL against null Uni+Gad - COV 0.1995 0.2036 0.1992 0.2019 0.2301 -51398 -64205 -66756 0.2340 -51135 -64205 -66756 0.2317 -51289 -64049 -66756 0.2343 -51117 -64049 -66756 0.500 0.000 1:(intercept) C.O.V.-JOSI C.O.V.-JO -0.500 Other Variables (+) Other Variables (-) -1.000 -1.500

Marketing/Shopping Marketing/Shopping Model structures: Logit family: Binomial, trinomial, hierarchical and ordered logit Count family: Hurdle with Poisson, hurdle with negative binomial Sets of Choice Alternatives: {0,1,2+} {0,1+} {0,1,2, ,n} Year-specific & joint models

MARKETING/SHOPPING MARKETING/SHOPPING Logit Family - Comparison of Estimates, 2and LL MNL Estimates Marketing/Shopping Trinomial Logit Marketing/Shopping - LL Comparison - MNL vs OL 1.50 0.50 0.569 0.561 0.555 0.551 0.600 -80000 0.523 -0.50 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -90000 0.500 -1.50 -100000 0.400 -2.50 -110000 -3.50 0.300 -120000 0.200 0.080 0.071 0.070 0.066 -130000 0.064 1:(intercept) 2+:(intercept) 1:Respondent1 0.100 2+:Respondent1 1:dumF 2+:dumF -140000 0.000 1:dumH 2+:dumH 1:StudentStatus_S TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -150000 2+:StudentStatus_S 1:SexM 2+:SexM 2 against Null 2 against Const. only LL model - MNL LL model - OL 1:PersMob_SV_YL 2+:PersMob_SV_YL 1:Reside_in_PD1 2+:Reside_in_PD1 1:Reside_in_PD2_6 2+:Reside_in_PD2_6 1:Count_TRE 2+:Count_TRE 1:TripDayOfWeek_F Marketing/Shopping Binomial Logit 2+:TripDayOfWeek_F OL Estimates BL Estimates 0.600 1.50 1.50 0.500 0.50 0.50 0.424 0.415 0.415 0.405 0.375 -0.50 0.400 -0.50 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -1.50 0.300 -1.50 -2.50 -2.50 0.200 0.093 0.080 0.080 0.077 0.073 -3.50 -3.50 0.100 Respondent dumF dumH 1+:(intercept) 1+:Respondent1 1+:dumF 0.000 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 StudentStatus_S SexM PersMob_SV_YL 1+:dumH 1+:StudentStatus_S 1+:SexM Reside_in_PD1 Reside_in_PD2_6 Count_TRE 2 against Null 2 against Const. only 1+:PersMob_SV_YL 1+:Reside_in_PD1 1+:Reside_in_PD2_6 TripDayOfWeek_F 1+:Count_TRE 1+:TripDayOfWeek_F

MARKETING/SHOPPING MARKETING/SHOPPING Count Family - Comparison of Estimates and Fit HRD PO Estimates 1.50 0.50 -0.50 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -1.50 -2.50 -3.50 count_(Intercept) count_Respondent count_dumF count_StudentStatus_S count_PersMob_SV_YL count_Reside_in_PD1 count_TripDayOfWeek_F zero_(Intercept) zero_Respondent zero_dumF zero_dumH zero_StudentStatus_S zero_SexM zero_PersMob_SV_YL zero_Reside_in_PD1 zero_Reside_in_PD2_6 zero_Count_TRE zero_TripDayOfWeek_F Example: Year 2016 HRD NB Estimates 1.50 0.50 -0.50 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -1.50 -2.50 -3.50 count_(Intercept) count_Respondent count_dumF count_StudentStatus_S count_PersMob_SV_YL count_Reside_in_PD1 count_TripDayOfWeek_F zero_(Intercept) zero_Respondent zero_dumF zero_dumH zero_StudentStatus_S zero_SexM zero_PersMob_SV_YL zero_Reside_in_PD1 zero_Reside_in_PD2_6 zero_Count_TRE zero_TripDayOfWeek_F

MARKETING/SHOPPING MARKETING/SHOPPING Year Comparison of Goodness of Fit, Interpretation of Results AIC Comparison The models are not able to explain the behaviour very well, in other words, we obtain low fit values ( 2~0.08). This could be attributed to several reasons: More variability in behaviour when compared to work/school activities Cannot capture the difference between milk shopping and shoe shopping due to the nature of the survey Lack of preference/taste related explanatory variables, lack of accessibility variables Only having a single day s information for an activity that potentially has a longer planning range (e.g., planned on a weekly basis) Year- -Specific Models Specific Models 350000 300000 250000 200000 150000 100000 50000 0 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 Multinomial Logit Binary Logit Ordered Logit Hurdle_Poisson Hurdle_Negative Binomial BL Better Than (expected due to aggregation, variance to be explained is reduced) MNL & OL Better Than HRD PO & HRD NB (useful if we want to estimate the exact number of activities) LL Comparison It is found that survey effects have significant impact on choice probabilities for Marketing/Shopping activities. This is an important finding as the results support the hypothesis that the marketing/shopping trips of non-respondents are under-reported. In addition, if the survey is filled online, then there is also more chance of under- reporting, when compared to phone surveys. Might be due to survey fatigue, since the individual has to fill the survey herself in the online version. 0 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -20000 -40000 -60000 -80000 -100000 -120000 -140000 -160000 MNL BL OL Hurdle PO Hurdle NB

Joint Estimates 3 Choices 1:(intercept) NP_MI NPDA_MI JOSI_MI JO_MI -1.57 - -1.58 - MARKETING/SHOPPING MARKETING/SHOPPING Joint Models Joint Models Comparison of Estimates int_1_96 int_1_01 int_1_06 int_1_11 int_1_16 - - - - - -1.59 -1.54 -1.55 -1.53 -1.66 - - - - - -1.59 -1.47 -1.51 -1.51 -1.47 2+:(intercept) -3.08 - -3.10 - int_2p_96 int_2p_01 int_2p_06 int_2p_11 int_2p_16 - - - - - -3.14 -3.09 -3.07 -3.00 -3.11 0.59 0.81 -0.81 -1.26 -0.40 -0.47 -1.25 -1.78 -0.17 -0.20 0.40 0.64 -0.47 -0.78 -0.17 -0.26 -0.07 -0.07 0.17 0.31 - - - - - -3.13 -2.96 -2.98 -2.95 -2.78 0.56 0.77 -0.78 -1.21 -0.38 -0.45 -1.20 -1.71 -0.16 -0.19 0.38 0.61 -0.44 -0.73 -0.17 -0.25 -0.07 -0.06 0.16 0.29 1.04 1.03 1.01 1.12 0.0694 0.5501 Joint Estimates 2 Choices 1:(intercept) NP_MI NPDA_MI JOSI_MI JO_MI -1.37 - -1.37 - 1:Respondent1 2+:Respondent1 1:dumF 2+:dumF 1:dumH 2+:dumH 1:StudentStatus_S 2+:StudentStatus_S 1:SexM 2+:SexM 1:PersMob_SV_YL 2+:PersMob_SV_YL 1:Reside_in_PD1 2+:Reside_in_PD1 1:Reside_in_PD2_6 2+:Reside_in_PD2_6 1:Count_TRE 2+:Count_TRE 1:TripDayOfWeek_F 2+:TripDayOfWeek_F 0.59 0.81 -0.81 -1.26 -0.39 -0.47 -1.24 -1.78 -0.17 -0.20 0.40 0.64 -0.48 -0.78 -0.17 -0.26 -0.07 -0.07 0.17 0.31 0.59 0.81 -0.82 -1.27 -0.40 -0.48 -1.26 -1.80 -0.17 -0.20 0.40 0.64 -0.47 -0.77 -0.17 -0.27 -0.07 -0.07 0.17 0.31 0.98 0.98 0.96 1.03 0.0693 0.5500 -630643 -630561 -677569 -677569 -1401515 -1401515 int_1_96 int_1_01 int_1_06 int_1_11 int_1_16 - - - - - -1.39 -1.34 -1.35 -1.31 -1.44 0.63 -0.90 -0.41 -1.33 -0.18 0.45 -0.53 -0.19 -0.07 0.20 - - - - - -1.39 -1.28 -1.31 -1.29 -1.26 0.60 -0.86 -0.39 -1.28 -0.17 0.42 -0.49 -0.18 -0.07 0.19 1.04 1.03 1.02 1.14 Estimates from different types of joint models are very close to each other. 1+:Respondent1 1+:dumF 1+:dumH 1+:StudentStatus_S 1+:SexM 1+:PersMob_SV_YL 1+:Reside_in_PD1 1+:Reside_in_PD2_6 1+:Count_TRE 1+:TripDayOfWeek_F 0.63 -0.90 -0.41 -1.33 -0.18 0.44 -0.53 -0.19 -0.07 0.20 0.63 -0.91 -0.41 -1.35 -0.18 0.45 -0.52 -0.19 -0.07 0.20 0.98 0.98 0.96 1.06 Joint models have similar 2values to what is obtained in the year- specific models. They are also (almost) the same for all four types of joint models. mu_01 mu_06 mu_11 mu_16 - - - - - - - - No clear pattern is observed in terms of the changes in scales over the years. The reasons could be similar to the ones explained for the poor fit of the year-specific models. mu_01 mu_06 mu_11 mu_16 - - - - - - - - 2against const. 2against null. LL model LL against const. LL against null 0.0799 0.0802 0.0799 0.0799 0.4050 -526110 -571816 -884258 0.4052 -525940 -571816 -884258 0.4053 -525883 -571533 -884258 -884257.6 0.4053 -525856 -571533 2against const. 2against null. LL model LL against const. LL against null 0.0694 0.5499 -630840 -677916 -1401515 0.0698 0.5500 -630630 -677916 -1401515

MARKETING/SHOPPING MARKETING/SHOPPING Joint Models Comparison of Goodness of Fit Joint Models Joint Models - 3 Choices - AIC Joint Models - 2 Choices - AIC 1261800 1261700 1052300 1261600 1052200 1261500 1052100 1261400 1052000 1261300 1051900 1261200 1051800 1261100 1051700 1261000 1051600 1260900 1051500 2 Comparison Na ve Na ve Pooling w/ Year-Specific Intercepts 3 Ch. Joint Model w/ Year- Specific Scales 3 Ch. Joint Model w/ Year- Specific Scales & Intercepts 3 Ch. Na ve Na ve Pooling w/ Year-Specific Intercepts 2 Ch. Joint Model w/ Year- Specific Scales 2 Ch. Joint Model w/ Year- Specific Scales & Intercepts 2 Ch. Pooling 3 Ch. Pooling 2 Ch. 0.55 0.55 0.55 0.55 0.60 0.50 0.41 0.41 0.41 0.41 0.40 0.30 0.20 Joint Models - 3 Choices - LL Joint Models - 2 Choices - LL 0.10 0.08 0.08 0.08 0.08 0.07 0.07 0.07 0.07 -630400 -525700 0.00 NP 3 Ch. NPDA 3 Ch. JOSI 3 Ch. JO 3 Ch. -630450 NP 2 Ch. NPDA 2 Ch. JOSI 2 Ch. JO 2 Ch. NP NPDA JOSI JO -525750 -630500 -525800 3 ch. - 2 against const. 3 ch. - 2 against null. -630550 -525850 2 ch. - 2 against const. 2 ch. - 2 against null. -630600 -525900 -630650 -525950 -630700 -526000 -630750 -526050 -630800 -526100 -630850 -526150 -630900

The scaled values of parameter estimates seem to be stable over time both for JO and JOSI type of joint models. This is valid both in the case of trinomial models and binomial models. MARKETING/SHOPPING MARKETING/SHOPPING Joint Models Scaled Parameter Estimates & Coefficient of Variation * Joint Models There is no trend in scaled year-specific intercepts. However, it should be noted that the values are very close to each other, the coefficient of variation for the intercept values is in the order of 0.02-0.04 (absolute value) considering both trinomial joint models and binomial joint models (JO & JOSI), meaning the extent of variability in relation to the mean of the population is very low. Scaled Estimates - JOSI - MI - 3 Choices Scaled Estimates - JO - MI - 3 Choices 2.00 2.00 1.00 1.00 0.00 0.00 Sc'd JO_96 Sc'd JO_01 Sc'd JO_06 Sc'd JO_11 Sc'd JO_16 Sc'd JOSI_96 Sc'd JOSI_01 Sc'd JOSI_06 Sc'd JOSI_11 Sc'd JOSI_16 -1.00 -1.00 -2.00 -2.00 -3.00 -3.00 -4.00 -4.00 1:(intercept) 2+:(intercept) 1:Respondent1 1:(intercept) 2+:(intercept) 1:Respondent1 2+:Respondent1 2+:Respondent1 1:dumF 2+:dumF 1:dumF 2+:dumF 1:dumH 2+:dumH 1:dumH 2+:dumH 1:StudentStatus_S 1:StudentStatus_S 2+:StudentStatus_S 1:SexM 2+:SexM 2+:StudentStatus_S 1:SexM 2+:SexM 1:PersMob_SV_YL 2+:PersMob_SV_YL 1:Reside_in_PD1 2+:Reside_in_PD1 1:PersMob_SV_YL 2+:PersMob_SV_YL 1:Reside_in_PD1 1:Reside_in_PD2_6 2+:Reside_in_PD2_6 1:Count_TRE 2+:Count_TRE 2+:Reside_in_PD1 1:Reside_in_PD2_6 2+:Reside_in_PD2_6 1:TripDayOfWeek_F 2+:TripDayOfWeek_F 1:Count_TRE 2+:Count_TRE 1:TripDayOfWeek_F Marketing/Shopping - 3 Choices - COV Marketing/Shopping - 2 Choices - COV Scaled Estimates - JO - MI - 2 Choices Scaled Estimates - JOSI - MI - 2 Choices 1.00 1.00 0.070 0.070 0.00 0.00 1 2 3 4 5 Sc'd JOSI_96 Sc'd JOSI_01 Sc'd JOSI_06 Sc'd JOSI_11 Sc'd JOSI_16 -1.00 0.020 -1.00 0.020 -2.00 -2.00 C.O.V.-JOSI C.O.V.-JO -0.030 C.O.V.-JOSI C.O.V.-JO -0.030 1:(intercept) 1+:Respondent1 1:(intercept) 1+:Respondent1 -0.080 1+:dumF 1+:dumH 1+:dumF 1+:dumH -0.080 1:(intercept) 1+:StudentStatus_S 1+:SexM 1+:StudentStatus_S 1+:SexM 1:(intercept) 2+:(intercept) 1+:PersMob_SV_YL 1+:Reside_in_PD1 1+:PersMob_SV_YL 1+:Reside_in_PD1 Other Variables (+) Other Variables (+) 1+:Reside_in_PD2_6 1+:Count_TRE 1+:Reside_in_PD2_6 1+:Count_TRE Other Variables (-) Other Variables (-) 1+:TripDayOfWeek_F 1+:TripDayOfWeek_F

Other Other Model structures: Logit family: Binomial, trinomial, hierarchical and ordered logit Count family: Hurdle with Poisson, hurdle with negative binomial Sets of Choice Alternatives: {0,1,2+} {0,1+} {0,1,2, ,n} Year-specific & joint models

OTHER OTHER Logit Family - Comparison of Estimates, 2and LL 2- 3 Choices "Other" - MNL LL vs OL LL MNL Estimates 0.417 0.450 -130000 2.00 0.376 0.373 0.373 0.361 0.400 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 1.00 -140000 0.350 0.00 0.300 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -150000 -1.00 0.250 -2.00 0.200 -160000 -3.00 0.150 -170000 -4.00 0.100 0.044 0.030 0.029 0.028 0.028 0.050 1:(intercept) 2+:(intercept) 1:Respondent1 -180000 0.000 2+:Respondent1 1:dumF 2+:dumF TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -190000 1:PersMob_SV_YL 2+:PersMob_SV_YL 1:TripDayOfWeek_F 2 const 2 null LL MNL LL OL 2+:TripDayOfWeek_F 1:TripDayOfWeek_M 2+:TripDayOfWeek_M 2- 2 Choices OL Estimates BL Estimates 0.450 2.00 2.00 0.400 1.00 1.00 0.350 0.300 0.00 0.257 0.00 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 0.213 0.212 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 0.250 0.209 0.199 -1.00 -1.00 0.200 -2.00 -2.00 0.150 0.100 -3.00 0.051 -3.00 0.034 0.033 0.032 0.032 0.050 -4.00 -4.00 0.000 1+:(intercept) 1+:Respondent1 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 Respondent dumF PersMob_SV_YL 1+:dumF 1+:PersMob_SV_YL 2 const 2 null TripDayOfWeek_F TripDayOfWeek_M 1+:TripDayOfWeek_F 1+:TripDayOfWeek_M

OTHER OTHER Count Family - Comparison of Estimates and Fit HRD PO Estimates 1.50 0.50 -0.50 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -1.50 -2.50 -3.50 count_(Intercept) count_Respondent count_dumF count_PersMob_SV_YL count_TripDayOfWeek_F zero_(Intercept) zero_Respondent zero_dumF zero_PersMob_SV_YL zero_TripDayOfWeek_F zero_TripDayOfWeek_M Example: Year 2016 HRD NB Estimates 1.50 0.50 -0.50 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 -1.50 -2.50 -3.50 count_(Intercept) count_Respondent count_dumF count_PersMob_SV_YL count_TripDayOfWeek_F zero_(Intercept) zero_Respondent zero_dumF zero_PersMob_SV_YL zero_TripDayOfWeek_F zero_TripDayOfWeek_M

OTHER OTHER Year Comparison of Goodness of Fit, Interpretation of Results Year- -Specific Models Specific Models AIC Comparison The values of parameter estimates are relatively less stable when compared to the models estimated for Marketing/Shopping, which might be because of the highly random nature of the Other activities. 500000 400000 300000 200000 The models for Other activities perform worse than Marketing/Shopping models. This could be attributed to their highly heterogeneous nature. This is clearly a drawback of the survey, as everything other than work, school and marketing/shopping is collected under the heading of other . 100000 0 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 Multinomial Logit Binary Logit Ordered Logit Hurdle_Poisson Hurdle_Negative Binomial BL Better Than (expected due to aggregation, variance to be explained is reduced) MNL & OL Better Than HRD PO & HRD NB (useful if we want to estimate the exact number of activities) LL Comparison 0 TTS1996 TTS2001 TTS2006 TTS2011 TTS2016 It is found that survey effects have significant impact on choice probabilities for Other activities. The hypothesis that the Other activities are under-reported for the non- respondents is supported by the model results. Similar to Marketing/Shopping models, if the survey is completed online, it is more likely to observe no other activities. This could be a result of survey fatigue. -50000 -100000 -150000 -200000 -250000 MNL BL OL Hurdle PO Hurdle NB

OTHER OTHER Joint Models Joint Models Comparison of Estimates Joint Estimates 3 Choices 1:(intercept) NP_OI NPDA_OI JOSI_OI JO_OI -1.64 - -1.59 - int_1_96 int_1_01 int_1_06 int_1_11 int_1_16 - - - - - -1.61 -1.57 -1.63 -1.63 -1.75 - - - - - -1.61 -1.60 -1.68 -1.70 -1.60 Joint Estimates 2 Choices 1:(intercept) NP_OI NPDA_OI JOSI_OI JO_OI -1.40 - -1.35 - int_1p_96 int_1p_01 int_1p_06 int_1p_11 int_1p_16 - - - - - -1.36 -1.32 -1.39 -1.41 -1.54 0.54 -0.55 0.55 0.20 -0.09 - - - - - -1.36 -1.35 -1.43 -1.44 -1.42 0.54 -0.55 0.54 0.20 -0.09 0.97 0.96 0.96 1.11 2+:(intercept) -2.97 - -2.88 - int_2p_96 int_2p_01 int_2p_06 int_2p_11 int_2p_16 - - - - - -2.88 -2.84 -2.95 -3.00 -3.18 0.46 0.79 -0.46 -0.82 0.46 0.79 0.19 0.24 -0.07 -0.12 - - - - - -2.88 -2.89 -3.04 -3.11 -2.93 0.46 0.79 -0.46 -0.83 0.46 0.79 0.18 0.23 -0.07 -0.12 0.98 0.96 0.95 1.11 0.0297 0.3789 -870492 -897141 Joint models have similar 2values to what is obtained in the year- specific models. 1+:Respondent1 1+:dumF 1+:PersMob_SV_YL 1+:TripDayOfWeek_F 1+:TripDayOfWeek_M 0.54 -0.54 0.54 0.20 -0.08 0.52 -0.53 0.52 0.19 -0.08 0.96 1.02 1.03 1.17 1:Respondent1 2+:Respondent1 1:dumF 2+:dumF 1:PersMob_SV_YL 2+:PersMob_SV_YL 1:TripDayOfWeek_F 2+:TripDayOfWeek_F 1:TripDayOfWeek_M 2+:TripDayOfWeek_M 0.45 0.78 -0.45 -0.81 0.46 0.78 0.19 0.24 -0.07 -0.12 0.44 0.77 -0.45 -0.80 0.45 0.77 0.18 0.23 -0.07 -0.12 0.98 1.02 1.03 1.12 0.0297 0.3789 -870532 -897141 -1401515 -1401515 No clear pattern is observed in terms of the changes in scales over the years. This is assumed to be a consequence of characteristics. the survey mu_01 mu_06 mu_11 mu_16 - - - - - - - - mu_01 mu_06 mu_11 mu_16 - - - - - - - - 2against const. 0.0333 0.0342 0.0338 0.0339 2against null. LL model LL against const. LL against null 0.2155 -693700 -717600 -884258 0.2162 -693070 -717600 -884258 0.2163 -693025 -692998 -717278 -717278 -884258 -884258 0.2163 2against const. 2against null. LL model LL against const. LL against null 0.0292 0.3783 -871310 -897514 -1401515 0.0300 0.3788 -870580 -897514 -1401515

OTHER OTHER Joint Models Joint Models Comparison of Goodness of Fit Joint Models - 3 Choices - AIC Joint Models - 2 Choices - AIC 1743000 1388000 1742500 1387500 1742000 1387000 1741500 1386500 1741000 Goodness of Fit Comparison 1386000 1740500 0.38 0.38 0.38 0.38 0.40 1385500 1740000 Na ve Pooling 3 Ch. Na ve Pooling w/ Year-Specific Intercepts 3 Ch. Joint Model w/ Year-Specific Scales 3 Ch. Joint Model w/ Year-Specific Scales & Intercepts 3 Ch. 0.35 1385000 Na ve Pooling 2 Ch. Na ve Pooling w/ Year- Specific Intercepts 2 Ch. Joint Model w/ Year- Specific Scales 2 Ch. Joint Model w/ Year- Specific Scales & Intercepts 2 Ch. 0.30 0.25 0.22 0.22 0.22 0.22 0.20 0.15 Joint Models - 3 Choices - LL 0.10 Joint Models - 2 Choices - LL 0.03 0.03 0.03 0.03 0.05 -870000 -692600 NP 3 Ch. NPDA 3 Ch. JOSI 3 Ch. JO 3 Ch. 0.00 0.03 0.03 JO_OI 0.03 JOSI_OI 0.03 NP_OI -870200 NP 2 Ch. NPDA 2 Ch. JOSI 2 Ch. JO 2 Ch. NPDA_OI -692800 -870400 Binomial - 2 against const. -693000 -870600 Binomial - 2 against null. Trinomial - 2 against const. -870800 -693200 Trinomial - 2 against null. -871000 -693400 -871200 -693600 -871400 -693800

Similar to Marketing/Shopping joint model outputs, the scaled values of parameter estimates seem to be stable over time both for JO type of models and JOSI type of joint models. This is valid both in the case of trinomial models and binomial models. OTHER OTHER Joint Models Joint Models Scaled Parameter Estimates & Coefficient of Variation * There is no clear trend in scaled year-specific intercepts. Although the values are close to each other, the coefficient of variation for the intercept values is between 0.05 and 0.07 (absolute value) (considering both trinomial joint models and binomial joint models (JO & JOSI)), meaning the extent of variability in relation to the mean of the population is higher when compared to Marketing/Shopping models. In year-specific models, it was concluded that the estimates are relatively less stable, and the higher range obtained from coefficient of variation confirms that remark. Scaled Estimates - JOSI - OI - 3 Choices Other - 3 Choices - COV Scaled Estimates - JO - OI - 3 Choices 2.00 0.080 2.00 1.00 0.060 1.00 0.00 0.040 0.00 Sc'd JOSI_96 Sc'd JOSI_01 Sc'd JOSI_06 Sc'd JOSI_11 Sc'd JOSI_16 -1.00 Sc'd JO_96 Sc'd JO_01 Sc'd JO_06 Sc'd JO_11 Sc'd JO_16 0.020 -1.00 -2.00 0.000 -2.00 -3.00 C.O.V.-JOSI C.O.V.-JO -0.020 -3.00 -4.00 -0.040 -4.00 1:(intercept) 2+:(intercept) 1:Respondent1 -0.060 1:(intercept) 2+:(intercept) 1:Respondent1 2+:Respondent1 1:dumF 2+:dumF -0.080 2+:Respondent1 1:dumF 2+:dumF 1:PersMob_SV_YL 2+:PersMob_SV_YL 1:TripDayOfWeek_F 1:(intercept) 2+:(intercept) 1:PersMob_SV_YL 2+:PersMob_SV_YL 1:TripDayOfWeek_F 2+:TripDayOfWeek_F 1:TripDayOfWeek_M 2+:TripDayOfWeek_M Other Variables (+) Other Variables (-) 2+:TripDayOfWeek_F 1:TripDayOfWeek_M 2+:TripDayOfWeek_M Scaled Estimates - JO - OI - 2 Choices Scaled Estimates - JOSI - OI - 2 Choices Other - 2 Choices - COV 1.00 1.00 0.080 0.50 0.060 0.50 0.040 0.00 0.00 Sc'd JO_96 Sc'd JO_01 Sc'd JO_06 Sc'd JO_11 Sc'd JO_16 0.020 Sc'd JOSI_96 Sc'd JOSI_01 Sc'd JOSI_06 Sc'd JOSI_11 Sc'd JOSI_16 -0.50 -0.50 0.000 -1.00 C.O.V.-JOSI C.O.V.-JO -1.00 -0.020 -1.50 -1.50 -0.040 -2.00 -0.060 -2.00 1:(intercept) 1+:Respondent1 1+:dumF -0.080 1:(intercept) 1+:Respondent1 1+:dumF 1+:PersMob_SV_YL 1+:TripDayOfWeek_F 1+:TripDayOfWeek_M 1:(intercept) Other Variables (+) Other Variables (-) 1+:PersMob_SV_YL 1+:TripDayOfWeek_F 1+:TripDayOfWeek_M

Key Findings For each activity type, year-specific models provide stable results in terms of the parameter estimates. The parameter estimates and the performances of various joint models are similar. The trade-off between computational burden and the amount of detail desired in a model should be considered before selecting a single structure to proceed with. FORECASTING Year-specific scales are around 1, indicating that year-to-year variations are low. Some results (increasing (in time) goodness of fit in work models versus decreasing fit in school models, etc.) make us question the sampling methods of the survey. Survey instruments are not significant for work/school related activities, whereas models support the idea that the chances of under-reporting shopping/other trips are higher for non-respondents. Moreover, under-reporting of shopping/other trips are more likely if the survey is filled online.

References Badoe, D. A. & Miller, E. J. (1998). Modelling Mode Choice with Data from two Independent Cross-Sectional Surveys: An Investigation. Transportation Planning and Technology, Vol. 21, pp. 235-261 Badoe, D. A. (1994). An Investigation into the Long Range Transferability of Work-Trip Discrete Mode Choice Models(PhD Dissertation). University of Toronto, Toronto, Canada. Data Management Group [University of Toronto]. (1997). 1996 Transportation Tomorrow Survey Data Guide Version 2.1. Retrieved July 18, 2018, from http://dmg.utoronto.ca/transportation-tomorrow-survey/tts-reports Data Management Group [University of Toronto]. (2003). 2001 Transportation Tomorrow Survey Data Guide. Retrieved July 18, 2018, from http://dmg.utoronto.ca/transportation-tomorrow-survey/tts-reports Data Management Group [University of Toronto]. (2008). 2006 Transportation Tomorrow Survey Data Guide Version 1.0. Retrieved July 18, 2018, from http://dmg.utoronto.ca/transportation-tomorrow-survey/tts-reports Data Management Group [University of Toronto]. (2013). 2011 Transportation Tomorrow Survey Data Guide Version 1.0. Retrieved July 18, 2018, from http://dmg.utoronto.ca/transportation-tomorrow-survey/tts-reports Data Management Group [University of Toronto]. (2018). 2016 Transportation Tomorrow Survey Data Guide. Retrieved July 18, 2018, from http://dmg.utoronto.ca/transportation-tomorrow-survey/tts-reports Data Management Group [University of Toronto]. (2008). 2006 TTS Working Paper Series: Interview Manual. Retrieved Miller, E. J. & Ozonder, G. (2018) Meeting Notes/Discussions Ozonder, G. (2018). Technical Documentation on Cleansing Transportation Tomorrow Survey Data (Technical Report). University of Toronto, Canada. RStudio Team. (2018). RStudio: Integrated Development for R. RStudio, Inc., Boston, MA URL http://www.rstudio.com/. Transportation Tomorrow Survey (1996). Travel Modelling Group Database. [Data File] Transportation Tomorrow Survey (2001). Travel Modelling Group Database. [Data File] Transportation Tomorrow Survey (2006). Travel Modelling Group Database. [Data File] Transportation Tomorrow Survey (2011). Travel Modelling Group Database. [Data File] Transportation Tomorrow Survey (2016). Travel Modelling Group Database. [Data File]

WORK SCHOOL THANK YOU THANK YOU Questions / Comments SHOPPING OTHER