An open letter to our frequentist friends

We have had to sit through a countless number of mind-numbing social science presentations and empirical papers during our academic career, most or all of which have relied on standard frequentist methods. We are writing today to request a favor. Why can’t we just admit that subjective priors are unavoidable in any field of inquiry? After all, we must necessarily begin with our priors when deciding what set of problems to solve and how to solve them. Why pretend otherwise? (By the way, all of you are smart people. Many of you have PhD’s and have many years of book learning and practical experience in your various fields. If any group of individuals is likely to have well-informed priors or good hunches about well-defined research problems, it’s you!)

Many of our colleagues, however, continue to reject Bayesian methods. You cling to an idealized conception of science. You equate “science” with standard frequentist methods, that is, with the ad hoc and easily manipulable statistical methods of Fisher, Pearson, and others. But this view of science is too narrow, too static. Science is ultimately about discovery, not about p values. Furthermore, all knowledge is contingent, including scientific or experimental knowledge. Nothing is certain. Thus the Bayesian notion of “degrees of belief” provides a realistic understanding of how our knowledge actually evolves over time. To discover, one must be willing to update one’s priors (whatever their source or level of subjectivity) in light of new evidence. Bayesian updating is an ongoing and never-ending process.

Where have we gone wrong?

Yours truly,

Prior Probability

This entry was posted in Bayesian Reasoning, Questions Rarely Asked, Science and tagged . Bookmark the permalink.

5 Responses to An open letter to our frequentist friends

  1. jecgenovese says:

    Reblogged this on peakmemory and commented:
    ” Science is ultimately about discovery, not about p values.”
    I agree with the thrust of this post. In my experience there are three big roadblocks to the adoption of Bayesian methods: 1. the lack of a set of agreed upon standard procedures, 2. journal editors and reviewers not being open to this approach, 3. stats courses in university still teaching the older methods.

    • enrique says:

      These roadblocks are indeed pretty big, but are they insurmountable?

    • James says:

      I’d add that it’ll help when it becomes easier to run a Bayesian analysis. Right now, and I’m mostly an R and SPSS guy, you have to know a fair amount of complex syntax, even for R. As much as I’m not a fan of GUI interfaces to do stats (looking at you SPSS) I’m not sure how “mainstream” Bayesian analysis will be till it becomes easier to conduct one.

  2. James says:

    Whether Baysian or Frequentist, it seems what needs to happen more, especially in the social sciences, is replication of previous work.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s