Thursday, July 3, 2008

Independent Contrasts Rule!

Despite the explosion of various types of phylogenetic comparative methods in recent years, Joe Felsenstein's independent contrasts still play a key role in the field. These contrasts can be thought of in a few ways: as a sort of mathematical trick to do phylogenetic generalized least squares (PGLS), as a transformation of evolutionary states to evolutionary rates, or as a method to "correct" for the phylogeny in a regression framework. In any case, they are conceptually simple yet statistically powerful.

Recently there has been an exciting development published in the American Naturalist: Felsenstein has extended his method so that one can now calculate contrasts both within and among species (the conceptual figure here is taken from this paper, Felsenstein 2008). This effectively accounts for error in the estimation of species means, which can cause bias in most applications of contrasts. But there's more to this paper than measurement error, and Felsenstein waits until the end of the paper to get into what is (to me) the best bit: this new method effectively uses contrasts to link micro- and macroevolution, unifying patterns within and among species. Sound familiar? I think this is the comparative method's version of the BEST approach.

Felsenstein, J. 2008. Comparative methods with sampling error and within-species variation: contrasts revisited and revised. Am. Nat. 171:713-725.


Dan said...

I agree that this is a very exciting development. I've already been talking to Rich about some very interesting applications of this method, so we'll probably take a stab at this soonish.

Tom Near said...

Luke, how will this new approach be implemented by end usesers? Will we see this incorporated to software packages that people are using for IC analyses (e.g. Mesquite, etc)?

Glor said...

The approach is currently implemented in PHYLIP, but hopefully it won't be long before it's included in other packages (e.g., ape or geiger for R).

Joe Felsenstein said...

Note also the 2007 paper by Ives, Midford and Garland (in Evolution) which allows for sampling variation but assumes that the within-species covariances are known, instead of estimating them. My method was actually first made available in PHYLIP's program CONTRAST in 2001, but was hardly used by anyone -- which is fortunate since I had the formulas wrong there until last summer!