Friday, October 25, 2013

Chris Sims on Bayesianism

I just love this piece by Chris Sims: "Bayesian Methods in Applied Econometrics, or, Why Econometrics Should Always and Everywhere Be Bayesian", from 2007.

In addition to the solid content, there are some great take-away snippets, such as:

  • "Bayesian inference is hard in the sense that thinking is hard."
  • "(People) want to characterize uncertainty about parameter values, given the sample that has actually been observed."
  • "Good frequentist practice has a Bayesian interpretation."

  • And Sims' conclusion: "Lose your inhibitions: Put probabilities on parameters without embarrassment."

    I can live with that!

    © 2013, David E. Giles

    1 comment:

    1. "A 95% confidence interval contains the true parameter value with probability .95 only before one has seen the data. After the data has been seen, the probability is zero or one."

      I am a bit puzzled by the language here, but is this just another way of stating that 95 out of a 100 times that a confidence interval is constructed it will contain the true parameter? If so, I don't find this distinction between pre and post sample probability to be particularly helpful. Mostly because I can only construct a confidence interval or calculate a p-value after I have seen the data, but perhaps also because I do not have particularly strong background in probability theory and statistics.