“Big Data” and Policy Learning

In early February 2014, during an industrial dispute with management about extending the London Tube’s hours of service, many of the system’s train drivers went on strike. Millions of passengers had to make other arrangements. Many switched their journey patterns to avoid their normal lines and stations, which were strike-hit, and to use those routes still running a service. Three economists downloaded all the data for the periods before and after the strike period from London’s pre-pay electronic travel card system (called the Oystercard), covering millions of journey patterns and linking each journey to a particular cardholder (Larcom et al, 2015). They found that one in 20 passengers changed their journey – an interesting ‘flexibility’ statistic on its own.
However, after the strike, they also found that a high proportion of these people also stayed with their new journey pattern when the service returned to normal, strongly suggesting that their new route was better for them than their old one had been. They considered two possible explanations of why people could have been using the ‘wrong’ Tube lines in the first place. One is that they were trying to maximise their welfare all along but had limited their initial search behaviour because of high search costs, so failing to optimise. The other possibility is that Tube travellers only ‘satisfice’: they had not set out to maximise their welfare in the first place, but were just going with the first acceptable travel solution that they found. The scale of savings made by the strike-hit changers was so high, however, that only the second, ‘satisficers’ explanation makes empirical sense.