October 24, 2016

Does Uber equitably serve different types of neighborhoods?

Don MacKenzie

Don MacKenzie

As new mobility services occupy a growing niche in urban transportation, one important question is whether these services are providing equitable access for diverse communities. There are many ways that inequity or discrimination could arise in the provision of these services, and Ryan Hughes and I have recently published a paper in the Journal of Transport Geography that looks at one of these dimensions. Specifically, we find that UberX does provide an equitable quality of service – as measured by expected waiting time – to Seattle neighborhoods with different levels of household income and percentages of minority residents.

As with most of our publications, you can find both a link to the final published version and a non-paywalled PDF of the manuscript on our publications page. The details are all in the paper, so I won’t devote too much space in this post, but the most interesting results are captured by the following 4 charts, which deserve some explanation.

These four plots show how the relationship between the logarithm of waiting time and neighborhood characteristics varies throughout the day. Pink bands indicate 95% confidence limits.

These four plots show how the relationship between the logarithm of waiting time and neighborhood characteristics varies throughout the day. Pink bands indicate 95% confidence limits.

Basically, each of these plots is showing how the waiting time (technically, the logarithm of waiting time) for an UberX ride depends on residential density (people per square mile), employment density (jobs per square mile), average household income, and the fraction of neighborhood residents who are minorities. Perhaps more importantly, they show how waiting time depends on each of these variables, after also adjusting for the effects of the other variables, at different times of day. The black line in each plot indicates the best estimate of how much the log of waiting time will change for a one-unit change in the noted variable. So in panel (a), we can see that at midnight, a change in population density of 1,000 people per square mile is associated with a change of -0.000017  x 1,000 = -0.017 in log waiting time. In other words, an increase of 1,000 people per square mile is associated with roughly a 1.7% shorter waiting time, holding all else equal.

The pink bands around the black lines indicate 95% confidence limits – when these confidence limits do not overlap zero, we say that the effect is statistically significant. We can see that higher population density is associated with shorter waiting times, and this effect is statistically significant, at all times of day. However, the effect is strongest in the middle of the night, and weakest (closest to zero) around the time of the morning rush hour.

Similarly, we can see that higher employment density is associated with significantly shorter waiting times throughout most of the day, except during the evening rush hour. Between midnight and noon, an increase of 1,000 jobs per square mile is associated with about a 0.25% shorter waiting time. During the evening rush hour, this effect is not significantly different than zero (since the pink bands overlap zero).

Surprisingly, we found that higher average incomes in a neighborhood are associated with longer expected waits for an UberX, even when density is held constant. A given neighborhood can expect to wait about 2.5% longer for an UberX ride than a similar neighborhood with an average income $10,000 lower, although this varies somewhat throughout the day.

Also surprisingly, areas with more minorities see no significant difference in waiting times at many times of the day. They see significantly longer waiting times in the late night hours, but significantly shorter waiting times around the morning rush hour. An area with 10% more minorities can expect about a 2% longer waiting time late at night, but nearly a 4% shorter wait around 7:00 AM, compared with a similar area with fewer minorities.

To put these numbers in perspective, the following figure shows how average household income and the fraction of minority residents varies throughout the Seattle regions:

Average household income and fraction of minority residents at census block group level.

Average household income and fraction of minority residents at census block group level.

Our data don’t allow us to conclude why we see these patterns in waiting times, but we can speculate. The effects of higher residential and employment density make sense: more residents and jobs means more demand for trips, which attracts more drivers, which means that on average, the distance to the nearest driver will be shorter. The effect of residential density may be weakest in the morning rush hour because high demand for commute trips is reducing the pool of available drivers in residential areas. Similarly, high demand from commuters finishing their workdays may explain why the effect of employment density disappears around the evening rush hour.

Less clear is why we see the observed patterns with lower-income and more diverse neighborhoods. One possibility is that there is less demand for trips in these areas, so nearby cars remain available for longer. Another possibility is that UberX drivers themselves may be more likely to live in these areas, reducing waiting times since they are more likely to be in these areas at the start and end of shifts. Alternatively, the road network in these areas may allow for a faster response time by nearby drivers (due to higher connectivity, e.g. fewer cul de sacs, or higher speed limits). Finally, it is possible that some UberX drivers might avoid minority neighborhoods late at night, although the associated waiting time penalty is small in any case.

Regardless of the exact reason, the bottom line is the same: at least according to estimated waiting time, good quality of service is available from UberX in neighborhoods of Seattle beyond just those that are “white and wealthy.” Additional ongoing work can help to determine whether this translates into truly equitable access for all individuals.

This research was funded by the University of Washington’s Royalty Research Fund and the Pacific Northwest Transportation Consortium (PacTrans).

Share this:Share on Facebook