When a city raises its minimum wage, economists want to know what happens next, and the honest answer is that this is unusually hard to find out. The trouble is not a shortage of data. It is that a city which raises its wage is rarely typical. Cities that pass such laws tend to be growing, expensive and politically distinctive, so a change in employment afterward might be caused by the new law, or by the same conditions that produced the law in the first place.

Researchers get around this with comparison. The best-known approach is to find a place that resembles the city as closely as possible but did not change its wage, and then watch the two together. In one influential design, economists compared fast-food restaurants on either side of a state border. The restaurants sold similar food to similar customers in the same regional economy, and the obvious difference between them was which side of the line they stood on. Whatever happened to employment on one side and not the other could reasonably be attributed to the wage.

The results of such studies have not settled the question, and it is worth being precise about why. Most find small employment effects, close enough to zero that measurement error matters. Some find modest losses of hours rather than of jobs, meaning the same number of workers scheduled for less time. Others find that turnover falls, which saves employers money they had been spending on hiring and training. These are not contradictory findings so much as different questions: jobs, hours and turnover are three separate things, and a study that measures one is silent about the others.

What a careful reader should take from this is not a verdict but a method. When a headline announces that a wage increase did or did not cost jobs, the useful questions are narrow ones. Which city, and compared with what? Over how long? Jobs, or hours, or both? A study covering six months may catch a hiring freeze and still miss an adjustment that takes three years.

Economists disagree about the size of the effect. They disagree much less about the difficulty of measuring it, and that second kind of agreement is the more useful thing for a reader to carry away.