Research Collection
Presentation
Connecting time-use, travel and shopping behavior Results of a multi-stage household survey
Author(s):
Schmid, Basil Publication Date:
2019-09-24 Permanent Link:
https://doi.org/10.3929/ethz-b-000366670
Rights / License:
In Copyright - Non-Commercial Use Permitted
This page was generated automatically upon download from the ETH Zurich Research Collection. For more information please consult the Terms of use.
ETH Library
Connecting time-use, travel and shopping behavior: Results of a multi-stage household survey
Basil Schmid IVTETH Zurich
PhD Thesis Defence
Zurich, September 24, 2019
Introduction
• Post-Car World (SNF project): Travel behavior in a (quasi-)car-less hypothetical environment (Schmid et al., 2019a)
=⇒ Pre-Post-Car World
• How do respondents trade-off different attributes in their (short- and long-term) choice and scheduling behavior?
• How are individual preferences for time-use, travel and shopping behavior related to each other?
• What is the role innovative mode sharing systems such as carsharing (CS) and carpooling (CP)?
– Valuation of level-of-service attributes
Fieldwork and response behavior
• Canton of Zurich (2015/16)
• 7’500 households invited by mail
• 50 CHF incentive payment
≈ 5% net cooperation rate
≈ 55% response rate (after recruitment)
• Final dataset:
≈ 300 households (red dots)
≈ 450 respondents
Survey methods: Three-stage household survey
• Revealed preference (RP; stage I) data: One-week reporting period on ...
– Travel behavior and time-use
– ICT usage (shopping, entertainment, etc.) – Daily and long-term expenditures
• Stated preference (SP; stage II) data on ...
– Mode choice – Route choice
– Shopping channel choice
• Stated adaptation (SA; stage III) data on ...
– Daily scheduling – Mobility tool ownership
The value of (travel) time: VTTS, VoL and VTAT
• Decomposition of the value of travel time savings (VTTS) into two components (DeSerpa, 1971; Jara-Diaz and Guevara, 2003)
VTTSi,n
| {z }
typically>0
=VoLn
| {z }
>0
−VTATi,n
| {z }
<or>0
– VTTS: Money value of indirect utility obtained from decrease in travel time – VoL: Money value to reduce travel time in favor of other activities;
opportunity value of time
– VTAT: Money value of the (dis)pleasure of travel time itself; depends on the conditions/comfort of travel
VTTS, VoL and VTAT: Policy relevance
• VTTS: Biggest share of user benefits in cost-benefit analyses
– Decisive role in the ranking of investment projects that affect travel time – Empirical basis for modeling and forecasting of travel behavior
=⇒ Identify mode and user-type effects in the VTTS
• VoL: Assessing the options under a budget constraint
VTTS [CHF/h] VoL [CHF/h] VTAT [CHF/h] Invest in ... Implication
20 5 –15 Conditions VTTS↓
20 20 0 Cond./speed VTTS/travel time↓
20 35 15 Speed Travel time↓
VoL and VTTS: Overview and main model variables
• VoL (H¨ossinger et al., 2019): Time assigned to activities (incl. work) and expenditures assigned to goods consumption, wage rate and fixed income (RP)
• VTTS (Schmid et al., 2019b): Travel choices (RP/SP) among a set of available alternatives with corresponding attributes (travel cost, travel time, etc.)
Mode Individual (user) Trip Random
VTTS VTTS/VoL VTTS VTTS/VoL
MIV Sex Distance Error components (VTTS)
PT Age Trip purpose Scale (VTTS)
CS Kids in HH Daily weather Taste (VTTS/VoL)
CP Couple Weekend vs. weekday
Bike Residential location area Walk Education
Personal income
SP route choice: Example choice situation
Mode-specific VTTS and VoL [CHF/h]
0102030405060VTTS and VoL [CHF/h]
Walk
Bike
MIV
PT
CS CP
VoL
VoL, VTTS and VTAT: Main findings
• VoL ≈52% of wage rate (49.5 CHF/h)
– Zurich has strong preference for goods consumption relative to leisure
• VTAT negative for car modes (MIV, CS and CP), positive for PT and bike – Conditions/quality of travel higher in PT and bike
– E.g. MIV vs. PT: Mode effect always dominates user-type effects
• VoL and VTTS are only weakly related
– Income has no effect on the VTTS, but on VoL
– Individual travel preferences (VTTS) mostly uncoupled with the opportunity value of time (VoL)
– Anomaly for Zurich? $$$$-income region with small share of mobility expenditures ...
In-store or online shopping?
• Goals, methods and main hypotheses (Schmid and Axhausen, 2019):
– Exploring key determinants for choice btw. in-store and online shopping
=⇒ Comparison of the value of delivery time (VDTS) and the VTTS – Search goods (standard electronic appliances): More easily evaluated from
externally provided information
– Experience goods (groceries): Preferably inspected physically – Incorporating attitudes in decision process
• Experimental framing:
– Home-based round trip for in-store alternative
– No private cars; no multi-channel shopping; no social motives – Goods quality is assumed to be identical
In-store or online shopping: Example choice situations
Ordering me / shopping me Travel me to store
Size / weight of the goods basket
Your choice
Situaon 1 Order Travel to store
CHF 5.20
600.00 54 39
48
540.00 min. 1 week
CHF
CHF
min. min.
min.
CHF Delivery me (incl.
possible delays)
Shopping cost Delivery cost / travel cost
Situaon 2
5.00 CHF
0.00 9.10
600.00 66 21
58
600.00 2-3 days
CHF
CHF
min. min.
min.
CHF Purpose: Electronic appl.
Ordering me / shopping me Travel me to store
Size / weight of the goods basket
Your choice
Order Travel to
store
Delivery me (incl.
possible delays)
Shopping cost Delivery cost / travel cost Purpose: Electronic appl.
13
In-store or online shopping? Main findings I
• VTTS vs. VDTS:
Median VTTS shopping trips (groceries) [CHF/h] 54.4 VTTS shopping trips (electronics) [CHF/h] 29.3 VDTS delivery time (groceries) [CHF/day] 9.8 VDTS delivery time (electronics) [CHF/day] 1.2
• Context-dependent cost sensitivity: Delivery vs. shopping vs. travel cost
– Develop effective retailing strategies (i.e. delivery as part of shopping cost)
• Pro-online shopping attitudes higher shopping cost sensitivity:
Larger choice set when considering both shopping channels
• Socioeconomic profile of online/in-store shoppers
– Forecasting: (1) direct effects and (2) mediated via attitudes ...
Adaptations in car usage and ownership
• Goals, methods and main hypotheses:
– Assessing radical pricing effects from an activity-based perspective
=⇒ 4 adaptation scenarios (daily scheduling and mobility tool ownership experiment) with increasing variable MIV costs
– Understand and quantify the transition towards a Post-Car World
=⇒ Comparison of the MIV cost elasticities between the two experiments (1%increasein cost → x %decrease in distance traveled)
– Substitution effects may eventually lead to a decreasing cost elasticity
• Experimental framing:
– Future policies to reduce MIV usage; promotion of of shared mobility – Road tolls, congestion and fuel taxes
Daily scheduling experiment: Example scenario
Mobility tool ownership experiment: Example scenario
Adaptations in car usage: Main findings
• Daily scheduling experiment: ≈ −0.4→ medium-run (upper-bound?)
• Mobility tool ownership experiment: ≈ −0.15 → long-run (lower-bound?)
=⇒ Reflects substitution patterns towards more energy-efficient cars
• Hypothesized pathway of cost elasticities:
Summary and conclusions
• Main contributions of the thesis:
– Unique survey design
– Connecting the value of leisure, travel and delivery time using sophisticated modeling approaches with clear suggestions for policy and practice
– Quantitative assessment of car usage from a mobility pricing perspective
• Main limitations:
– Sampling bias (participation choice and response behavior) – Data resolution (activities) and quality (expenditures)
– Hypothetical bias (no private cars, experimental framing, etc.)
Questions?
Literature
DeSerpa, A. C. (1971) A theory of the economics of time,The Economic Journal,81(342) 828–846.
H¨ossinger, R., F. Aschauer, S. Jokubauskaite, B. Schmid, S. Peer, R. Gerike, S. R. Jara-Diaz and K. W. Axhausen (2019) A joint time assignment and expenditure allocation model: Value of leisure and value of time assigned to travel for specific population segments,Transportation(in press).
Jara-Diaz, S. R. and C. A. Guevara (2003) Behind the subjective value of travel time savings: The perception of work, leisure and travel,Journal of Transport Economics and Policy,37(1) 29–46.
Jara-Diaz, S. R., M. A. Munizaga, P. Greeven, R. Guerra and K. W. Axhausen (2008) Estimating the value of leisure from a time allocation model, Transportation Research Part B: Methodological,42(10) 946–957.
Molloy, J., B. Schmid, F. Becker and K. W. Axhausen (2019) mixl: An open-source R package for estimating complex choice models on large datasets, Working Paper, 1408, Institute for Transport Planning and Systems (IVT), ETH Zurich, Zurich.
Petrin, A. and K. Train (2010) A control function approach to endogeneity in consumer choice models,Journal of Marketing Research,47(1) 3–13.
Schmid, B. and K. W. Axhausen (2019) In-store or online shopping of search and experience goods: A hybrid choice approach,Journal of Choice Modelling,31, 156–180.
Schmid, B., M. Balac and K. W. Axhausen (2019a) Post-Car World: Data collection methods and response behavior in a multi-stage travel survey, Transportation,46(2) 425–492.
Schmid, B., F. Aschauer, S. Jokubauskaite, S. Peer, R. H¨ossinger, R. Gerike, S. R. Jara-Diaz and K.W. Axhausen (2019b) A pooled RP/SP mode, route and destination choice model to investigate mode and user-type effects in the value of travel time savings,Transportation Research Part A:
Policy and Practice,124, 262–294.
Appendix: Response behavior, by survey wave
Pre-Test Wave 1 Wave 2 Wave 3 Stage I survey period: Jan. 2015 Jul. 2015 Oct. 2015 Apr. 2016
Number of households invited: 800 1600 3500 1600
Contacted by phone: 270 676 1110 546
Rejected participation: 203 543 919 428
Accepted participation: 67 133 191 118
Response burden scores of stage I: 3500 1700 1700 1700
Response rate stage I: 52.2% 54.8% 64.9% 66.9%
Response burden scores of stage II: 530 350 350 350
Response rate stage II: 52.2% 54.8% 61.7 % 63.6%
Response burden scores of stage III: 270 500 500 –
Estimated total response time 360 min. 215 min. 215 min. 170 min.
Final response rate: 52.2% 54.1% 60.2% 63.6%
Final cooperation rate: 4.9% 5.2% 3.8% 5.3%
Appendix: Response behavior for different incentive levels
• Sample selection participation choice Probit model – Incentive payments were varied in the pre-test
– 100 vs. 50 CHF: Increase in initial participation probability by 28%-points – Negative effect on completion conditional on participation of –33%-points
=⇒ net-effect on completion not statistically different from zero
Reject Participate Drop-out Complete Total
N [%] [%] [%] [%] [%]
Incentive 50 CHF 989 57.9 42.1 37.5 63.5 26.7
70/80 CHF 58 62.1 37.9 50.0 50.0 19.0
100 CHF 34 44.1 55.9 52.6 47.4 26.5
Sample N 1081 624 457 156 284 284
Appendix: Response burden and response rates @ IVT
020406080100Response Rate [%]
0 500 1000 1500 2000 2500 3000 3500 4000
Response Burden Score [−]
Prior recruitment and incentive Fit & 95 % CI Prior recruitment, no incentive Fit
No prior recruitment, no incentive Fit
PCW (Pre−Test) PCW (Wave I)
PCW (Wave II) PCW (Wave III)
a)
a) a)
a) a) a)
b) b)
c) c)
Appendix: Time-use model specification
• Log-linear additive function with diminishing marginal utility (Jara-Diaz et al., 2008; H¨ossinger et al., 2018)
– Tw: Time assigned to work – Ti: Time assigned to activities
– Ej: Expenditures assigned to goods consumption
• θ andψ represent the baseline utility parameters max U = θwlog(Tw) +
I
X
i=1
θilog(Ti) +
J
X
j=1
ψjlog(Ej)
s.t. τ−Tw−
I
X
i=1
Ti = 0 (µ) Ti −Timin ≥0, ∀i ∈Ac (κi)
w ·Tw+Y −
J
XEj ≥0 (λ) Ej −Ejmin≥0, ∀j ∈Gc (ηj)
Appendix: Mode/route choice as utility maximization
• Assume linear function of travel time and cost that one wants to minimize (mode-independent)
• Slopes are given by individual preferences → VTTSn
• Choice = mode/route with highest utility
Appendix: Conditional indirect utility function
• An individual behaves according to maxU(Xj,Qi)
s.t.X
j
PjXj +ci ≤I
• Find optimal consumption Xj based on the discrete choicei with costci → conditional demand Xj(P,I−ci,Qi) with available incomeI−ci
• Optimize overi → the conditional indirect utility function Vi is the maximum utility the individual would achieve if alternative i was chosen
Vi ≡U(Xj(P,I−ci,Qi),Qi)
Appendix: Subjective value of a characteristic
• The marginal (unconditional) utility of income equals minus the marginal (conditional) dis-utility of costci
• The subjective value of an increasein quality characteristick in alternative i is defined as the marginal rate of substitution (MRS) between this characteristic and money (e.g. increase in cost ci) at constant utility
SVk,i = ∂Vi/∂Qk,i
∂V/∂I =−∂Vi/∂Qk,i
∂Vi/∂ci
• E.g. average willingness to pay for reductionin travel time VTTSi = βtt,i
βtc
Appendix: Time-use data
Category PCW [h] MAED [h] ATUS [h] Variable
Working 36.2 37.8 36.9 Tw
Wage rate [CHF/h] 49.5 14.5 – w
Leisure – 28.9 32.3 Tf1
Out-of-home leisure 16.4 – – Tf1
In-home leisure 9.6 – – Tf2
Eating – 9.3 8.6 Tf2
Shopping – 2.1 2.0 Tf2
Committed time – 80.6 80.1 Tc
Travel 10.6 9.3 8.1 Tc
Shopping 1.9 – – Tc
Other 0.8 – – Tc
At home 92.5 – – Tc
Total (time constraint) 168.0 168.0 168.0 τ
Appendix: Expenditures data
Category PCW [%] MAED [%] HABE [%] Variable
Hotel, restaurants and holidays 11.4 6.2 7.5 Ef1
Leisure 2.9 7.8 3.9 Ef1
Clothes and accessories 5.3 5.6 3.1 Ef2
Electronics 2.2 3.6 2.0 Ef2
Taxes 23.1 – 10.8 Ec
Housing 18.6 23.2 18.7 Ec
Food 10.2 17.3 9.1 Ec
Health (incl. insurance) 7.0 2.4 9.8 Ec
Mobility 5.0 12.7 6.8 Ec
Communication 1.6 – 2.2 Ec
Furniture 1.6 2.4 1.3 Ec
Education 1.3 2.0 1.6 Ec
Services 1.8 3.1 2.0 Ec
Insurances1 3.0 8.2 17.6 Ec
Other expenditures 4.9 4.7 3.5 Ec
Avg. weekly expenditures [CHF] 1931.0 560.1 2309.7 PEf
j+Ec Avg. weekly labor income [CHF] 1995.6 550.7 2296.8 w·Tw
Appendix: VoL model estimation
• Four baseline utility parameters to be estimated: Λx,n∈ {θw, θTf1, ψEf1,Ψ} Λx,n=±exp(ζx +Znβx+ρx,n)
• The VoL for each individual is given by
VoLdn= ∂U/∂Ti
∂U/∂Ej = µ
λ = w·Tw∗+Y −Ec Ψbn(τ−Tw∗−Tc)
• Ψbn: Conditionals of the preference parameter for freely chosen goods relative time
• µ andλ: Marginal utility of increasing available time/money (s.t. satiation)
• Decomposition of the VoL in a preference and data-driven component:
log(VoLdn) = −logΨbn
+ logw·Twd∗+Y −Ec−logτ −Twd∗−Tc
Appendix: Discrete choice model specification
• Including time-use residuals to account for endogeneity (Petrin and Train, 2010)
• Mixed Logit model estimated in willingness-to-pay space (Schmid et al., 2019b)
VTTS^i,n,t = (VTTSi +Pn,tρVTTS,i+ZnκVTTS,i+ηVTTS,i,n)distn,t
dist
θVTTS,i
• The scale parameter is defined as
ψen,t = exp (βscale +Znκscale+ηscale,n)distn,t dist
θscale
>0 ∀ n,t
• The VTTS for each mode and individual is given by the conditionals VTTS\i,n
• Estimated in R 3.4.1 using mixl-package (Molloy et al., 2019)
Appendix: Travel choice data
• Discrete choice model: Pooled RP/SP panel data
• 12’594 choice observations, 362 individuals
02468Percent [%]
0 20 40 60 80 100
a) # choice observations per respondents
0510152025Percent [%]
Walk (MC_RP)Bike (MC_RP)MIV (MC_RP)PT (MC_RP)Walk (MC_SP)Bike (MC_SP)CP (MC_SP)CS (MC_SP)PT (MC_SP)
Alt. 1 (RC_CS)Alt. 2 (RC_CS)Alt. 3 (RC_CS)Alt. 1 (RC_PT)Alt. 2 (RC_PT)Alt. 3 (RC_PT)
b) Choice rates, by alternative (pooled data set)
Appendix: Time-use model estimation results
• All baseline utility main effects significant (p <0.01)
• θw <0: Utility of working time negative
• Utility of out-of-home leisure significantly higher than in-home leisure
BASE INTER TUMIX
# estimated parameters 4 24 28
# respondents 362 362 362
# MLHS draws − − 2000
R2: Tw∗ equation 0.73 0.73 0.77 R2: Tf1∗ equation 0.64 0.67 0.75 R2: Ef1∗equation 0.30 0.41 0.57
LLfinal −5592 −5528 −5510
AICc 11193 11109 11080
Appendix: VoL decomposition
log(money) log(time) log(exprate) log(pref.) log(VoL) Age [years] +0.34∗∗∗ −0.12∗∗ +0.39∗∗∗ +0.78∗∗∗ −0.19∗∗∗
Male +0.33∗∗∗ −0.09 +0.36∗∗∗ +0.49∗∗∗ +0.00
High educ. +0.27∗∗∗ −0.07 +0.29∗∗∗ +0.19∗∗∗ +0.16∗∗∗
Urban res. loc. +0.04 +0.05 +0.00 −0.00 +0.01
Couple +0.03 −0.09∗ +0.09 −0.02 +0.10∗
Kids −0.10∗ −0.12∗∗ +0.00 −0.55∗∗∗ +0.43∗∗∗
Inc. [CHF] +0.62∗∗∗ −0.10∗ +0.63∗∗∗ +0.38∗∗∗ +0.37∗∗∗
Money = available money component; time = available time component; exprate = expenditure rate;
pref. = preference component
Significance levels: ∗∗∗:p<0.01,∗∗:p<0.05,∗:p<0.1
Appendix: Choice model estimation results
• All mode-specific VTTS main effects significant (p<0.01)
• Endogeneity: LR test significant (8 df; increase in LL by 34.8 units)
• Trip explain much more than respondent characteristics
RMNL TMNL UMNL MIXL
# est. parameters 30 52 66 79
# respondents 356 356 356 356
# choice observations 12594 12594 12594 12594
# Sobol draws − − − 5000
LLfinal −10875 −8791 −8617 −6548
AICc 21816 17704 17397 13301
Appendix: Mode effects in VTTS
• MIV: Reference mode
• Residential location area: Strongest power in disentangling the total mode effect
MIV–PT MIV–CS MIV–CP
Total ∆VTTSi−j 15.8 –0.1 –2.1
Interquartile range (IQR) (19.0) (20.1) (17.3)
Ntotal 233 173 98
Female ∆VTTSi−j 21.2 10.9 2.2
Na 117 82 49
Male ∆VTTSi−j 8.7 –7.9 –5.7
Nb 116 91 49
Agglo./rural ∆VTTSi−j 19.3 5.1 1.2
Na 147 111 70
Urban ∆VTTSi−j 5.9 –6.5 –10.8
Appendix: VTTS and VoL correlation analysis
VTTS walk VTTS bike VTTS MIV
log(preference) −0.04 −0.11 −0.41∗∗∗
log(avail. money) −0.10 0.01 −0.21∗∗∗
log(avail. time) −0.18∗∗∗ 0.07 0.09
N 256 166 253
VTTS PT VTTS CS VTTS CP
log(preference) −0.23∗∗∗ −0.16∗∗ −0.36∗∗∗
log(avail. money) 0.00 0.05 −0.18∗∗
log(avail. time) 0.01 0.08 0.10
N 331 219 120
Significance levels: ∗∗∗:p<0.01,∗∗:p<0.05,∗:p<0.1
Appendix: Shopping attitudes
onl1: I often order products on the Internet onl2: Online shopping is associated with risks
onl3: Credit card fraud is one of the reasons why I don’t like online shopping onl4: The Internet has more cons than pros
onl5: Online shopping facilitates the comparison of prices and products
onl6: The risk of receiving a wrong product is one of the main reasons why I don’t like online shopping
onl7: I like to follow the new developments in the tech industry onl8: All I need, I find in the shops
onl9: Number of different IT gadgets in possession
ple1: I like to visit shops, even if I don’t want to buy something, just for looking around ple2: A disadvantage of online shopping is that I cannot physically examine the products ple3: Shopping usually is an annoying duty
Appendix: Correlation patterns and attitudes
PRO−ONLINE PLEASURE Male Age Income High education Car available HH members Store access.
Married Non−working Urban res. loc.
PLEASURE Male Age
Income Car available
Married Non−working
c > +0.3 +0.3 > c > +0.2 +0.1 > c > +0.2
−0.1 < c < −0.2
−0.2 < c < −0.3 c < −0.3
Appendix: Hybrid choice modeling framework
Appendix: HCM structural model I
The utility equations for shopping channeli ∈ {O,S}and individualn ∈ {1,2, ...,N}in choice scenariot ∈ {1,2, ...,Tn} with choice attributesXi,n,t and the latent variables LVz,nwith z ∈ {online shopping attitudes,pleasure of shopping} are given by
UO,n,t =XO,n,tβO +LVz,nµLVz +Zz,nΘz+f1,O,n,t+f3,n,t+ψO,n+O,n,t
US,n,t =XS,n,tβS+f1,S,n,t+f2,n,t+S,n,t where
f1,i,n,t = (βcost +Zonline,n∆online,cost +ψcost,n)·costi,n,t+ ϕLVonline,cost ·LVonline,n·costi,n,t
f2,n,t = (βtime,S +Zpleasure,n∆pleasure,time,S+ψtime,S,n)·timeS,n,t+ ϕLVpleasure,time·LVpleasure,n·timeS,n,t
f = (αM,L+α ·male +α ·age )·sizeM,L
Appendix: HCM structural model II and measurement model
The LV structural equations for latent variablez are linear functions of observed socioeconomic characteristicsZz,n for individualn:
LVz,n=Zz,nρz+ηLVz,n ηLVz,n ∼N(0, σLV2 z)
The latent variable measurement equations with responses to the attitudinal questions (items)Iw,n with w ∈ {onl1,onl2, ...,ple3}are given by
Iw,n=Iw +τIwLVz,n+νw,n
νw,n∼N(0, σI2w)
Appendix: Calculation of valuation indicators
First obtain the conditional estimates of the shopping cost coefficientλ^cost,n, which is then used to calculate the weighted average marginal disutility of costCn for each individual:
λrcost,n=βbcost+Zonline,n∆bonline,cost +ψcost,nr +ϕbLVonline,costLVronline,n
λ^cost,n= PR
r=1
hQ
i
QTn
t=1P(ci,n,t = 1|Xi,n,t,Zz,n,Ωb,Σb,Γr)ci,n,tλrcost,ni PR
r=1
Q
i
QTn
t=1P(ci,n,t = 1|Xi,n,t,Zz,n,Ωb,Σb,Γr)ci,n,t were Γr corresponds toLVonline,nr andψri,n. Finally,
Cn=
Tn
X
t=1
1 Tn
λ^cost,nP
icosti,n,t+βbdelivery costG,E delivery costG,En,t P
icosti,n,t+delivery costGn,t,E Travel and delivery time are fixed coefficients = =
Appendix: Daily scheduling data
• Average MIV costs: 0.17 CHF/km (base scenario) → 1.77 CHF/km (scenario 4)
• Average distance traveled: 60 km (base scenario) →11 km (scenario 4)
012345Average MIV km cost [CHF/km]
B S1 S2 S3 S4
050100150200250300Traveled distance by MIV [km]
B S1 S2 S3 S4
Appendix: Mobility tool ownership data
• Average MIV costs: 0.25 CHF/km (base scenario) → 1.14 CHF/km (scenario 4)
• Average distance traveled: 9’900 km (base scenario)→ 6’610 km (scenario 4)
0.511.522.5Average MIV km cost [CHF/km]
B S1 S2 S3 S4
010,00020,00030,00040,00050,00060,000Traveled distance by MIV [km]
B S1 S2 S3 S4
Appendix: Daily distance; by mode
020406080100Daily distance [in % relative to base scenario]
Base Scenario 1 Scenario 2 Scenario 3 Scenario 4
Appendix: Yearly distance; by mode
020406080100Yearly distance [in % relative to base scenario]
Base Scenario 1 Scenario 2 Scenario 3 Scenario 4
Small car Medium car Large car
Appendix: Adaptations in mobility tool ownership
020406080100a) Car fleet [% rel. to base scenario] Base Scenario 1 Scenario 2 Scenario 3 Scenario 4
Small car Medium car Large car
MPV Company car
020406080100b) Fuel type [% rel. to base scenario] Base Scenario 1 Scenario 2 Scenario 3 Scenario 4
Gasoline Diesel Gas
Electric Hybrid
20406080100 020406080100120d) Costs [% rel. to base scenario]
Base Scenario 1 Scenario 2 Scenario 3 Scenario 4
Appendix: Modeling framework adaptations in car usage
Exponential regression model accounting for panel structure (EMIX):
yn,t = expα+Znρ+ψα,n+βen·log(xn,t)+ζn,t
βen= −exp(βcost +Znκ+ψβ,n)·
distn,0
dist0 ωdist
A draw ofψis calculated as ψrα,n ψβ,nr
!
= σα 0
σα,β σβ
!
· ηαr ηrβ
!
From this follows that Var(ψα,n) =σα2, Var(ψβ,n) =σ2α,β+σβ2 and Cov(ψα,n, ψβ,n) =σα·σα,β
Appendix: Estimation results (adaptation in car usage)
BASE DIST USER EMIX
# estimated parameters 3 4 13 16
# respondents 163 163 163 163
# choice observations 741 741 741 741
# draws − − − 4000
R2 0.11 0.57 0.63 0.82
LLfinal −4010.39 −3747.20 −3696.99 −3573.68
AICc 8026.93 7502.65 7422.42 7183.08
# estimated parameters 3 4 9 12
# households 165 165 165 165
# choice observations 821 821 821 821
# draws − − − 4000
R2 0.02 0.46 0.54 0.95
LLfinal −2885.36 −2642.64 −2577.22 −2015.49
Appendix: Cost elasticities (adaptation in car usage)
• All cost elasticity main effects significant (p <0.01)
• EMIX: Substantial amount of unobserved heterogeneity
• Daily scheduling: Elasticities are larger in all model specifications
• Results strongly affected by the inclusion of covariates; decreasing elasticities for increasing distance traveled in the base scenario
• EMIX: Similar relative, but decreasing magnitudes compared to BASE model
BASE DIST USER EMIX N
Medianbn −0.47 −0.27 −0.26 −0.37 163
Meanbn −0.47 −0.34 −0.36 −0.42
IQR (0.0) (0.30) (0.35) (0.29)
Medianbn −0.16 −0.24 −0.23 −0.13 165
Meanbn −0.16 −0.29 −0.27 −0.19
IQR (0.0) (0.27) (0.25) (0.12)
Appendix: Further research (related to current dataset)
• Mobility tool ownership RP model
– Investigate preferences and substitution patterns
– Integrate in the mode/route choice model (self-selection) – Modification of availability conditions
• Savings rate in time-use model – Savings now part ofEf2
– Decrease in the median VoL by about 7%
• Interrelations between travel behavior and ICT usage – Based on diary data (RP)
– Comparison/validation of results obtained from the shopping choice model
• Inclusion of attitudes (now only in shopping choice model) – Better representation of the decision process?
Appendix: Further research (general)
• Finer grained distinction of activities and expenditures – Committed vs. freely chosen
– Secondary activities (during travel, at home, etc.)
• Better tailored research design to disentangle mode and user-type effects – Self-selection still present (at the individual and/or trip-level)?
– Suggestion: Additional route choice tasks for different modes/trip purposes
• Context-dependency of results
– Example: Travel vs. shopping vs. delivery cost sensitivity – Important for valuation studies
• More sophisticated research design to investigate shopping channel preferences – More relaxed assumptions (product categories, trip chaining, etc.)
Appendix: Main lessons learned
• Focus groups: Which attributes are important?
• Reduction in response burden, more specific research design
• Better address respondents who are less willing to participate (saliency effects)
• Inclusion of traditional modes (e.g. car) in SP experiments
• Within-subject designs (trip purpose, distance, shopping purpose, etc.)
• Avoid necessity to impute based on external datasets (e.g. fixed income)