Friday, 26 April 2013

Trevor Hastie: Statistical Models for Presence-Only Data (Friday 3rd May 2-3)

Trevor Hastie is a world-renowned statistician with major contributions to applied statistics, including two classic textbooks. His current research focuses on applied problems in biology and genomics, medicine and industry.

We have selected a paper co-authored with William Fithian that is sure to be a citation giant for ecologists and statisticians alike, addressing many key challenges in analyzing presence-only data: "Statistical models for presence-only data: Finite-sample equivalence and addressing observer bias".

This is a heavy-hitter, so we will focus mostly on Sections 1, 2, and 5.

Feel free to post any questions and/or comments.


  1. we thought this idea of changing the weights in logistic regression to make it do a point process model was pretty cool. But we noticed that you can already do a ppm using GLM with a Poisson link, so the situations when this would be of benefit might not be as broad as the authors suggest? e.g. you can do a GAM with family=poisson easily enough...

  2. We've been talking a bit about presence-only records and how reliable they are - cleaning them to take into account spatial accuracy, when the data were collected, replicate entries... (important stuff but generic to the whole pres-only literature rather than being specific to this paper)

  3. The idea of combining data from multiple sources (section 5) and how you can add pres-abs to a pres-only analysis in order to turn a relative measure of occurrence into an absolute measure came across well...

  4. Can/how incorporate geographical barriers and dispersal into this sort of model... (especially relevant to range-shifting like what we talked about a couple of weeks ago)

  5. While we talked more about other sections of this paper today, perhaps the bit of most relevance in preparing for the Symposium is the equivalence results between MAXENT and other methods. For a gentler introduction try the MAXENT post on the Methods Blog.