I think some of these examples are not real examples of “inadequate equilibria”. They are instead situations with real switching costs, or where there are conflicting beliefs or interests.
To illustrate, the authors example of moving from proprietary to open-source journals seems like a real example to me. But their example of using Bayesian rather than frequentist statistical methods does not seem like a real example.
Note that I’m a Bayesian (though not a rabid one), and that I’ve taught introductory statistics. There isn’t some easy way to just switch to Bayesian methods. First, students need to understand the scientific literature, and that includes the past scientific literature. So for a considerable period of time, students will need to understand frequentist statistics. This is a legacy compatibility problem, not a coordination problem. Second, there are many scientists who don’t know Bayesian statistics. This is a retraining problem (which is not cost-free), not a coordination problem. Third, not everyone agrees that Bayesian methods are better. This is a persuasion problem, not a coordination problem. Fourth, Bayesian methods aren’t always better—there really are problems where the correct Bayesian approach is much, much more difficult to carry out than a simple frequentist approach that is usually gives mostly-correct results. So it really is necessary for at least some people to understand frequentist statistics, though it would be good if the emphasis eventually changes in a Bayesian direction. There may be some coordination aspects to the current mess, but to a considerable extent the mess reflects real issues with real costs and benefits.
I think some of these examples are not real examples of “inadequate equilibria”. They are instead situations with real switching costs, or where there are conflicting beliefs or interests.
To illustrate, the authors example of moving from proprietary to open-source journals seems like a real example to me. But their example of using Bayesian rather than frequentist statistical methods does not seem like a real example.
Note that I’m a Bayesian (though not a rabid one), and that I’ve taught introductory statistics. There isn’t some easy way to just switch to Bayesian methods. First, students need to understand the scientific literature, and that includes the past scientific literature. So for a considerable period of time, students will need to understand frequentist statistics. This is a legacy compatibility problem, not a coordination problem. Second, there are many scientists who don’t know Bayesian statistics. This is a retraining problem (which is not cost-free), not a coordination problem. Third, not everyone agrees that Bayesian methods are better. This is a persuasion problem, not a coordination problem. Fourth, Bayesian methods aren’t always better—there really are problems where the correct Bayesian approach is much, much more difficult to carry out than a simple frequentist approach that is usually gives mostly-correct results. So it really is necessary for at least some people to understand frequentist statistics, though it would be good if the emphasis eventually changes in a Bayesian direction. There may be some coordination aspects to the current mess, but to a considerable extent the mess reflects real issues with real costs and benefits.