An unintended consequence of Uber

Among other things, Uber has made it far easier for party-goers to get home safely. A study published in 2017 found that after Uber’s arrival in Portland, Oregon, alcohol-related car crashes declined by 62%. But at the same time, the spread of ride-hailing apps may have tempted people to drink to excess, knowing that they won’t be at the wheel. A study published in November 2019 by three economists - Jacob Burgdorf and Conor Lennon of the University of Louisville, and Keith Teltser of Georgia State University - found that the widespread availability of ride-sharing apps had indeed made it easier for the late-night crowd to binge.

By matching data on Uber’s availability with health surveys from America’s Centres for Disease Control and Prevention, the authors found that on average alcohol consumption rose by 3%, binge drinking (in which a person downs four or five drinks in two hours) increased by 8%, and heavy drinking (defined as three or more instances of binge drinking in a month) surged by 9% within a couple of years of the ride-hailing company coming to town. Increases were even higher in cities without public transport, where the presence of Uber led average drinking to rise by 5% and instances of binge drinking to go up by around 20%. (Heavy drinking still rose by 9%.) Remarkably, excessive drinking had actually been in decline before Uber’s appearance, giving further evidence that the firm’s arrival affected behavior.

The launch succeeds: more people can leave the car at home after drinking. Months later, another dataset associates the service’s arrival with increased alcohol consumption. Both outcomes can coexist because reducing one consequence of risky behaviour changes the calculation around the behaviour itself.

Making safe transport harder would be idiotic. Fewer alcohol-related crashes is a substantial benefit. The studies also have limits: cities change, drinking is self-reported and service availability follows patterns of its own. The useful point is that “safer” at one stage can increase activity earlier in the sequence. Helmets, refunds, insurance, moderation and easy recovery all alter what people feel able to risk. That response may be acceptable, or even desirable. It still belongs in the evaluation.

A launch report usually measures the intended benefit first. Trips completed. Drivers reached. Incidents prevented. The behaviour that adapts is slower to appear and may sit in health, support or enforcement data controlled elsewhere. By the time it is visible, senior leaders may already have approved the launch using the successful metric.

A serious evaluation therefore needs outcomes from outside the service itself and enough time for people to adjust. Compare similar places, inspect pre-existing trends and state what the study cannot establish. Otherwise one correlation becomes a triumphant case study or a moral panic according to whoever chose the headline. Both reactions waste the useful uncertainty.

Map the full sequence before celebrating. What becomes cheaper, safer or more reversible? Who might do more of it as a result? Which harm moves to another stage or group? Then measure the net outcome without pretending every consequence can be reduced to one number. Late at night, one person leaves the car at home and another orders an extra drink. The service records two journeys.