Weekend Long Reads: Deconstructing Productivity

by Kevin Schofield

This weekend's "long read" is an article published by Noah Smith, looking at the productivity of construction workers and how it's measured.

The conventional wisdom is that for decades construction productivity has been in a slow decline, in sharp contrast to other intensive industries such as agriculture and manufacturing. This graph from the Economist sums up the picture well:

Smith points out that there are two problems with the conventional wisdom. First, it lumps all construction together; once you disaggregate it, there are some significant differences in productivity for different types of construction. For example, industrial construction productivity has increased, while transportation construction productivity has declined.

Similarly, multi-family home construction has become much more productive than single-family homes, suggesting that some kinds of buildings lend themselves more to the kinds of innovation and technological improvements that make workers more productive.

Smith argues that part of the reason why construction productivity as a whole has languished for so long is that over time there has been a shift in the mix of construction projects in the U.S. toward more single-family homes and transportation projects, some of the least productive types of construction. But Smith also raises a second problem: He casts doubt on the way we measure productivity in construction, or more specifically the accuracy of the underlying data. He rightly points out that worker productivity doesn't suddenly spike over the course of a few years and then regress just as quickly, so the jagged edges in the graphs above are strong arguments that the underlying data is unreliable.

Smith's article is a good reminder that it's usually a good idea to reserve a fair amount of skepticism when confronted by the conventional wisdom — and that not all data is created equal.

What happened to construction productivity?

Featured image is attributed to J J (under a Creative Commons, CC BY-NC-ND 2.0 license).