Kremer’s O-ring theory of economic development


Jason Collins


January 11, 2013

The latest issue of the Journal of Economic Behavior and Organization has a new paper by Garett Jones (ungated version here) on the  O-ring theory of economic development. Its been floating around as a working paper for a few years, so its nice to see it get a home. But before I post about that paper, I thought I’d revisit Michael Kremer’s classic 1993 paper on which Jones builds.

The name of the theory comes from the cause of the Challenger space shuttle explosion. In that case, a highly complex machine with thousands of components failed because one minor part, an O-ring, failed in the cold conditions. This is despite the remaining components being in order.

Kremer saw a similarity between what occurred in the Challenger case and what may happen in production in the economy. A company with a great product and service may fail due to bad marketing. An otherwise functional good may sell at a much reduced price due to a single defect. Kremer asked, if processes of this nature are the norm, what does this imply for economic development?

Kremer pictured firms that engage in production involving a series of tasks. Workers have different skill levels, which is represented by a probability that they properly perform the task. Even if workers have relatively high completion rates, small differences are costly. A firm employing a production process with 10 tasks and workers who complete their step with 95 per cent probability produces only 60 percent (0.9510) of the output of a firm with perfectly competent workers. It is also disastrous to have a weak link in the chain, as 9 fully competent workers paired with someone who messes up half the time will see their total output halved compared to a firm with the perfect ten.

This setup has some interesting implications. First, people will sort by skill as firms will find it worthwhile to employ people of the same competency. Those firms with the best workers will then attract the most capital. This will lead to large wage differentials, with the high-skilled workers paired together being much more productive than the low-skilled. Large differences in wages might also be observed across borders if there are differences in skill between countries.

If we assume that the tasks in a process are sequential, there are high costs to messing up at later stages of the production process as the work of everyone before is wasted. Kremer uses the example of Rembrandt, who finished off the face and hands of portraits after his assistants had done the easier work. The result of this assumption is that high skilled workers will be employed later in the production chain. If there is a lack of high-skilled workers, firms will focus on producing goods with short and easy production processes. Kremer suggests this may explain the higher share of primary production in poor countries, while rich countries will specialise in complex products.

Finally, Kremer asks what happens if firms cannot perfectly assess a worker’s skill. In that case, there will be imperfect matching between firms and workers, and firms will be less efficient. But if workers use education to both signal and increase skill, there are benefits from the improved matching and the significant output increase as mistakes drop. Kremer states that this provides a strong argument for subsidising education, as small increases in education and skill increase the returns to education and skill, causing a virtuous circle.

Kremer’s model also provides an argument for boosting IQ. Any measure that can systematically increase worker quality will have multiplicative effects. This matches the observation that boosting a person’s IQ increases their income, but boost the population’s IQ and the wealth gains are many times higher.

I tend to see Kremer’s model as a natural counterpart of another story about the benefits of high-quality workers. Modern growth is not primarily generated by people not messing up, but by people coming up with great ideas. If you place a bunch of innovative people together, there can be huge network effects. The cost of the low-skilled does not multiply in the same way as for Kremer’s model, but alternative assumptions about the strength of network effects of innovation for those high-quality people can generate similar disparities in wages. Firms pay the low skilled less simply because they have zero marginal product in those innovative processes.

Jones’s new paper has some relevance to that last point. I’ll post on his paper in the next few days. (That post on Jones’s paper can now be found here.)