Case notes Solutions/Answers for Innovation at Uber The Launch of Express POOL Assignment Comparing Commuting and Non-Commuting Hours By Chiara Farronato, Caleb Kwon

5-622-054
JANUARY 31, 2022




TEACHING NOTE


Innovation at Uber: The Launch of Express POOL
Assignment: Comparing Commuting and Non-
Commuting Hours
This document provides solutions to the assignment linked to the HBS case “Innovation at Uber:
The Launch of Express POOL” (HBS No. 619-003). The assignment asks students to use the spreadsheet
supplement (HBS No. 619-702) to analyze the effects of the switchback experiment described in the
case. In particular, the assignment evaluates the effects of extending wait times from 2 to 5 minutes on
customer experience and efficiency metrics. The focus of the assignment is on identifying differences
in the experimental treatment between different time periods during the day: commuting hours versus
non-commuting hours. Using the data supplement, please answer the following questions.

Whenever we refer to an outcome (e.g. Express trips), please consider the value per switchback period.

The R code used to answer the assignment questions can be found in TN Appendix A.

Throughout the solutions, when doing t-tests we assume unequal variances. For Problem 1 this
works because commuting and non-commuting hours are not allocated randomly across hours of the
day. For Problem 2 it is a conservative approach, although not strictly necessary because the treatment
and control samples are constructed to be from the same type of drivers and riders, same locations,
same hours of the day.


Problem 1: UberPOOL and Commuting versus Non-Commuting Hours
Compare commuting hours versus non-commuting hours in the control group (i.e., with 2-minute
wait times).

For this exercise we need to select a subsample of the dataset: the observations in the control group,
i.e., the observations where the variable “treat” = FALSE.

Then, within the “treat” = FALSE subsample, we can run t-tests between the observations with the
variable “commute” = TRUE and observations with the variable “commute” = FALSE.

,622-054 Teaching Note—The Launch of Express POOL Assignment: Comparing Commuting and Non-Commuting Hours




In this solution document we compute the difference in a particular metric of interest between
commuting and non-commuting hours as “average metric during commuting hours – average metric
during non-commuting hours.”

1. Do commuting hours experience a higher number of ridesharing (Express + POOL) trips
compared to non-commuting hours?

_X__ YES ___ NO

For this question we need to construct a variable for the total number of trips considering Pool
and Express, so: “trips_total” = “trips_pool” + “trips_express.”

The average number of ridesharing trips is calculated as the average “trips_total” for each
subsample (“commute” = TRUE and “commute” = FALSE).

When calculating this we get average number of trips equal to 5,046.00 for commuting hours,
and equal to 3,763.40 for non-commuting hours.

So commuting hours do experience a higher number of total ridesharing trips.

2. What is the difference in the number of ridesharing trips between commuting and non-
commuting hours?

___1,282.60____________

This is computed as the difference between 5,046.00, the average number of trips for commuting
hours, and 3,763.40, the average number of trips for non-commuting hours.

3. Is the difference statistically significant at the 5% confidence level?

_X__ YES ___ NO

We run a two-sample t-test between the commuting and non-commuting observations. The p-
value from the t-test is 9.6e-6 in R and 8.6e-6 in Excel, way below the threshold of 0.05. With a
p-value well below 0.05 we can confidently conclude that the difference in ridesharing trips
between commuting and non-commuting hours is significantly different from zero, so we can
reject the null hypothesis. Below is the output from the t-test in R and Excel.

Output from R:




2

, Teaching Note—The Launch of Express POOL Assignment: Comparing Commuting and Non-Commuting Hours 622-054




Output from Excel:




4. Do riders use Express at higher rates during commuting hours compared to non-commuting
hours?

__X_ YES ___ NO

We need to construct a new variable for the share of Express trips as follows: “share_express”
= “trips_express” / (“trips_express” + “trips_pool”).

The average share of Express trips is 0.70 for commuting hours. We calculate this by computing
the average of “share_express” for the sample with “commute” = TRUE.

Then we calculate the average of “share_express” for the subsample with “commute” = FALSE
and we get that the rate is 0.65 for non-commuting hours.

We conclude that riders use Express at a higher rate during commuting hours. We can speculate
as to why this occurs. One possibility is that we have more price sensitive users during
commuting hours. Another possibility has to do with product quality. During commuting
hours travel patterns may be more predictable (everybody goes to the offices downtown, or to
school) so there are enough people requesting rides that go in the same direction at the same
time and riders don’t have to wait too long before being matched under the Express product
nor experience detours while in the car to pick up other passengers.

5. What is the difference in the share of Express trips between commuting and non-commuting
hours?

_______0.05________

This is computed as the difference between 0.70 for commuting hours and 0.65 for non-
commuting hours.

6. Is the difference statistically significant at the 5% confidence level?

_X__ YES ___ NO


3