Crash course in ABC

| No Comments
On July 3rd and 4th, the Australian Statistical Conference put on a free satellite workshop at the University of New South Wales, which introduced Approximate Bayesian Computation (ABC) to anyone interested. Having some familiarity with Bayesian statistics and a strong curiosity for ABC, I decided it was worth my 4th of July to attend. 

The first day of the workshop provided a crash course in ABC methods, while the second day consisted of lectures on ABC applications to research. It turns out that approximate Bayesian analysis is useful when the likelihood function is computationally intractable and when likelihood-based inference models are unavailable. ABC is a non-parametric Bayesian method where pseudo data is generated by a candidate \theta parameter. The summary statistics from the two data sets are compared and if they are similar, then \theta is acceptable for the real data. We use this procedure to estimate the posterior distribution of the parameters of a model. This gets more complicated with dimensionality, but I'll leave that to the Bayesian experts for now. Wikipedia has a decent explanation on this for those who are interested. 

ABC is a fairly new statistical frontier that dates back to only the 1980s. It started as a niche idea in population genetics and is now gaining momentum in mainstream statistics. Although ABC still feels a bit weird and awkward to me, my gut intuition is that these methods will become useful when modeling other natural systems too. 

Leave a comment

About this Entry

This page contains a single entry by Hillary Scannell published on July 5, 2014 12:13 PM.

G'Day from Sydney was the previous entry in this blog.

Invasion of the jellyfish! is the next entry in this blog.

Find recent content on the main index or look in the archives to find all content.