Uber vs Taxi Usage: the Toronto VfH Case Study Gozde Ozonder, MSc - PowerPoint PPT Presentation

Uber vs Taxi Usage: the Toronto VfH Case Study Gozde Ozonder, MSc TMG Workshop Eric J. Miller, PhD Mobility Tools & Services Department of Civil & Mineral Engineering February 5, 2020 University of Toronto

The Telegraph, 2015 NL Times, 2019 The Local, 2019 Global News, 2017 The Telegraph, 2014 Pacific Standard, 2019 2

Agenda Credit: https://canadianobligation.weebly.com/blog/uber- protesting-toronto-taxi-drivers-block-traffic-downtown Study Area Data Findings Analysis & & & Results Objectives Methods Caveats 3

Study Area & Objectives (1) Validating the representativeness of the most recent regional travel survey’s Uber trips (2) Greater Toronto and Tracking the changes or stabilities in taxi usage between 1996 & 2016 Hamilton Area (GTHA) (3) Identifying profiles for Uber and taxi users 4

Data Credit: https://www.cnet.com/news/uber-will-reportedly- Credit: http://tts2016.ca/en/index.php let-riders-drivers-record-audio-of-trips-for-safety/ Credit: INRO, 2019 Emme4 + Network Assignment • Travel times Transportation Tomorrow Survey Uber Trip Data (TTS) through the City of Toronto • ~ 5% of the GTHA • Pick-up/drop-off times & dates • Household-based • Distance traveled • Daily travel diary • Service type 1996 2001 2011 2016 2006 TTS TTS TTS TTS TTS 5

Analyses Fall 2016 Uber & 2016 TTS • Trip pattern comparison analysis: Survey vs Actual • Spatiotemporal Uber Trips • Trip length ( & ) TTS between 1996 & 2016 • Taxi time-series analysis: Taxi Users & Trips • Attributes of taxi users [1996-2016] • Trip patterns 2016 TTS • Subsample comparison analysis: • Survey population / trip-making population / Uber users / taxi users • Persona analysis: User Profiles • Decision trees 6

Methods Homogeneity predicted class label • Determined based on purity/impurity predicted probability measures for “Uber” ▪ e.g., entropy, Gini % of observations in index, Chi square, that node etc. Decision Trees Variable Importance • Supervised machine • Evaluating the impact of Feature selection learning algorithms each variable on the outcome • Use “recursive Avoid overfitting! partitioning” heuristic • Random forests ▪ Permutation importance 7

Survey vs Actual Uber Trips 8

Survey vs Actual Uber Trips (1/2) 9

Survey vs Actual Uber Trips (2/2) 12 am – 6 am 6 am – 9 am 9 am – 3 pm 3 pm – 7 pm 7 pm – 12 am 10

Taxi Users & Trips [1996-2016] 11

Taxi Users & Trips [1996-2016] (1/5) 12

Taxi Users & Trips [1996-2016] (2/5) 13

Taxi Users & Trips [1996-2016] (3/5) 14

Taxi Users & Trips [1996-2016] (4/5) 15

Taxi Users & Trips [1996-2016] (5/5) 0.437 % 0.384 % 0.421 % 0.424 % 0.387 % 2016 Uber Mode Share: 0.23% 16

User Profiles 17

Younger people User Profiles (1/5) App-based service? Tech-savvy? Smartphones? Uber: Taxi: Mode = 30 Mode = 40 Median = 33 Median = 51 Mean = 35 Mean = 52 Older people Poor vision? Less tech-savvy? Reluctant to change habits? 18

User Profiles (2/5) • Both: Home & “Other” • Uber > Taxi: Work • Taxi > Uber: Shopping TAXI UBER Home Home Other Other • Taxi > Uber when Shop Shop 0 ≤ Income < $60K Work Work • Uber > Taxi when $60K ≤ Income 19

User Profiles (3/5) {11, 12, …, 98} {Downtown Toronto (PD1); Neighboring PDs (PD2-6); Rest of Toronto (PD7-16); Mississauga (PD36); Others} {“Yes”, “No”} {General Office, Clerical; Manufacturing, Construction, Trades; Professional, Management, Technical; Retail Sales and Service; Unemployed} 20

User Profiles (4/5) UBER USERS TAXI USERS 21

User Profiles (5/5) Leonard: 35 & Emily: 54 & younger, lives older, taxi outside the core user. of the City (outside PD1-6), but not in Mississauga, owns a driver’s license, is either unemployed or Nick: 35 & employed in older, lives sector “P” or “M”, outside the City Uber user. of Toronto but not in Kyle: 35 & older, lives in the Lorelai: 35 & younger, lives Mississauga, City or in Mississauga, is either in downtown Toronto, its taxi user. unemployed or works in a neighboring PDs, or in sector other than “M”, taxi user. Mississauga, Uber user. 22

Findings • Representative Uber trips by the latest regional travel survey • Strong stability in taxi user profile & trip patterns (20-year period) • Different profiles of Uber & taxi users • Pivotal Attributes: age & residential location In general in the GTHA: • 54 → taxi • 35 , residing in downtown Toronto/surrounding areas/Mississauga → Uber 23

• Relatively early days Caveats of Uber in Toronto [ ] • Weekend trip-making (missing!!!) Fall 2016 2018 M T 2020 W • Behavior of the R visitors to the region (not captured!!!) S F D Credit: https://leadworklife.com/illustrations/diffusion-of-innovation/ 24

Thank You! Questions/Comments gozde.ozonder@mail.utoronto.ca & miller@ecf.utoronto.ca 25 Credit: https://www.reddit.com/r/BeAmazed/comments/6041sn/aerial_view_of_Toronto