Finding taxis on rainy days

A classic story on the play-list of many behavioural economics presentations is why you can’t find taxis on rainy days. The story is based on the idea that taxi drivers work to an income target. If driver wages are high due to high demand for taxis, such as when it rains, they will reach their income target earlier and go home for the day. The result is you can’t find a taxi when you need one most.

The story is such a favourite as it conflicts with conventional economic wisdom that people are maximisers who respond positively to incentives such as higher wages. Instead, drivers are satisficers who quit work for the day once have hit their target, even though the high wages would allow them to earn more than normal.

This story originates from a 1997 article by Colin Camerer and friends (I suggest following Camerer on twitter). They analysed taxi trips in New York and found that as wages went up, labour supply (taxis on the street) goes down. Their preferred explanation, based on what some drivers said, was that taxi drivers work to a daily income target. Their article did not include the reference to the rain, but it has become the way the story is traditionally told.

But, a new study suggests this negative relationship between wages and supply might not generally be the case for New York taxi drivers. Using a much bigger dataset of New York taxi driver activities, Henry Farber has found that, as standard economic theory would suggest, taxi drivers drive more when they can earn more. There was no evidence of income targeting in the data.

As another blow to the rainy day story, Farber also found that taxi drivers didn’t earn more when it was raining. As traffic was worse and they travelled less distance, their earnings didn’t increase despite the higher demand. There were less taxis on the street when it was raining, but this must be due to causes such as drivers preferring not to work when traffic is bad.

So how do we reconcile these conflicting findings? A starting point is in the original study. In a show of humility, Camerer and colleagues were open to the idea that their result might not be robust. They close with the following paragraph:

Because evidence of negative labor supply responses to transitory wage changes is so much at odds with conventional economic wisdom, these results should be treated with caution. Further analyses need to be conducted with other data sets (as in Mulligan [1995]) before reaching the conclusion that negative wage elasticities are more than an artifact of measurement or the special circumstances of cabdrivers. If replicated in further analyses, however, evidence of negative wage elasticities calls into question the validity of the life-cycle approach to labor supply.

To use the cliché, more research is required. And there has been a lot more research since Camerer and friends’ had their study published. While I’ve pitched the story as a new paper tearing up an almost 20-year old favourite, there has been a sequence of papers over the years with both supporting and conflicting results, including by Farber.

Farber’s explanation for the result in his latest paper is that he had access to a larger dataset - five million shifts compared to a few thousand in Camerer and friends’ or Farber’s earlier studies. Technological progress in recording taxi data also allowed Farber’s work to be at a much finer level of detail than was possible at the time of the original study. Other studies also had small datasets or used less reliable data such as from surveys (such as this one from Singapore), but there have also been at least one involving similarly large sets of taxi data that did find a negative relationship (such as a second from Singapore, although in that case the negative relationship seemed of too low a magnitude to support income targeting).

Another explanation might lie in the methodological battle about how you should measure the relationship between wages and supply for taxi drivers. Farber’s 2005 paper picked apart the original methodology, particularly around their assumptions on wages, and he chose a different approach based on drivers deciding whether to continue or not at the end of each ride. When I previously invested some time to understand it, I found Farber’s critique reasonably persuasive. However, I haven’t taken the time to understand the finer points of Farber’s new analysis and to what extent methodology determines the result, so it will be interesting to see some responses to this latest salvo.

Another potential distinction is that Camerer and friends’ original study was able to distinguish between owner-operators and employee drivers, each of which face different incentives. Farber wasn’t able to tease the two apart. However, Camerer and friends found a negative relationship for both groups, so at a minimum, Farber’s work suggests that the finding would not hold across both. Farber did consider whether there might be many different types of driver, which there may be. But if the satisficers do exist, there are not many of them.

On a brighter note, there is some hope that we will be better able to catch a taxi on rainy days in the future. With current taxi regulation and fixed pricing, the inconvenience of driving in bad traffic results in less taxis on the road. But with new entrants such as Uber able to charge more and adjust pricing at times of high demand, we might actually get more taxis or other vehicles on the road when we need them most. And we can have some comfort that when those taxis are needed most, there will be plenty of maximisers around to fill our need.