Posted by: Jeremy Fox | January 30, 2012

Statisticians, meet ecologists

Interesting discussion thread over at statistician Andrew Gelman’s blog, about time series analysis of the lynx-hare cycle. Standard phenomenological statistical models (autoregressive moving average models) don’t fit or predict these data all that well. Andrew links approvingly to a recent statistical paper which does much better by fitting a simple mechanistic model–indeed, a laughably simple model, the original Lotka-Volterra predator-prey model! Ecologists Eric Pedersen, Ben Bolker, and myself then showed up in the comments to point out that ecologists have been mechanistically modeling the lynx-hare cycle and other ecological time series for 20 years now. I think this work–by Royama, Turchin, Wood, McCauley, Ellner, Harrison, Rees, Kendall, Bjornstad, Grenfell, King, Keeling, Ben himself, and numerous others*, is some of the best and most important ecological work of my generation. This work has always involved collaboration between ecologists, applied statisticians, and people who wear both hats, but apparently it’s not widely known among statisticians. Hopefully the linked thread will help to change that, as Andrew’s blog is very widely-read by statisticians.

Which does raise the question, are there other ways to increase cross-talk between statisticians and ecologists? NIMBioS is doing its part, and IIRC the Canadian statistical society had a special session on mechanistic time series models at its annual meeting a couple of years ago. What else could be done? Certainly sounds like there’s an opportunity for somebody (not me) to write some sort of perspectives-type review paper for a stats journal, highlighting the success of mechanistic time series modeling in population and disease ecology.

And while we’re at it, how about more cross-talk between the population and disease ecologists who’ve been doing most of this work, and ecologists working in other areas? Ecologists as a whole are increasingly statistically sophisticated. But often that sophistication is at the service of fitting and testing purely phenomenological statistical models. This is problematic, because translating the predictions of mechanistic ecological models into a form that can be evaluated by phenomenological statistical models often is tricky. Unfortunately, this translation process often seems to be based on nothing more than intuition and arm-waving, leading to an unfortunate trend towards using rigorous, sophisticated methods to test shaky, unsophisticated hypotheses. Why not try to follow the lead of the best population and disease ecology and fit mechanistic models directly to our data? And don’t say “we can’t do that because we don’t have enough data”, because one thing ecologists should not be in this LTERNCEASNutNetNEONCIEE era is data-limited.

*My own efforts along these lines are so far quite modest and not without problems. But I hope to do better in future. A long-term ambition of mine is to apply this approach to many-species communities.


  1. I left a comment on Andrew’s blog, but I’ll say here that I think the interaction between stats & ecologists is pretty good (even if not perfect). I think the lynx-bare analyses are one good example, as you show, but there’s a lot of work going on on developing and fitting more mechanistic models. Come to ISEC in Norway this summer, and you’ll see.
    I would dispute your suggestion that we have enough data. We might do for a few systems, but when I was looking for data on long term community dynamics it wasn’t easy (not a lot of data, and sometimes difficult to extract – this was 4 or 5 years ago though).

    • Thanks Bob. Re: having enough data, I didn’t mean to suggest that we necessarily do. But if we don’t, we often should be able to go out and get it (at least from model systems). NutNet is an example, and microcosm experiments are another. It’s true we don’t have a lot of data on long-term community dynamics, except from lake plankton. But there are other ways to get the data needed to parameterize whole-community models. Indeed, a strength of recent population and disease ecology is their ability to make effective use of other lines of evidence to estimate parameter values, rather than relying solely on the time series data.

  2. They (Reilly and Zeringue p. 298) state: “The problem with this approach is that there is no data on the hare population for this period (meaning, the period from which the lynx pelt data is taken, i.e. 1821 to 1934); hence we will need to impute the hare population…”. But this is flatly wrong, as Elton’s data include those from snowshoe hares from (at least) 1845 to 1910, maybe longer. They should have known this, and seemingly would have, had they read the paper they cite (Tong, 1990) that they claim reviewed “the dozens of papers” that had analyzed those data by that time.

    Maybe I should withhold final judgement until I’ve read that paper again, and more thoroughly, but it’s kind of hard given the strange things that they say in there and the amount of biology overlooked or glossed over. Looks to me like a classic case of statisticians not understanding the underlying subject matter.

  3. On second read, that paper is completely idiotic.

  4. […] like predicting future pest outbreaks, have come from fundamental work in population ecology (see here for […]

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 )

Facebook photo

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

Connecting to %s


%d bloggers like this: