Adaptive Medication

I, like an increasing number of 40+ humans, am taking medication to lower my blood lipids in the hopes this reduces my risk of heart disease. For quite a while now I've been twisting my GP's arm (gently), trying different medications to see if there is one that will fit my personal genotype better. The motivation for this was the recognition that the usual "best" medication - according to randomized studies of large populations, wasn't doing the job for me. It occurred to me that genetic variation among people means that some people will respond better than others (duh.), and hence, the only way to work out what works for me, is, well, Adaptive Medication. Try something new. Test. Try something else. Test. Initially I tried to work out a model of cholesterol synthesis to generate some hypotheses but gave that up. If you think climate scientists are having trouble coming up with good models you should see physiologists. I knew there was a reason I never took biochemistry as an undergrad.

Thus I was intrigued by Pascale Hammond's discussion of how a clinical health care provider works and why this is science. Frankly I couldn't agree more - but the AMed spin raises an interesting point about what happens with all the data generated by those diagnostic tests of differential diagnoses. Unless I give my GP explicit permission, that information is not usable to improve medical understanding. Millions, nay, Billions or Trillions of dollars spent every year testing hypotheses and none of it used to advance the underlying science. I understand that there are privacy issues, but ... but ... this is my potential future longevity that is suffering! Maybe there would be a way for individual patients to "volunteer" to have their diagnoses and test results entered into a global double-blind database in much the same way that gene sequences are.

The black box of risk

I've been interested in the nature of risk and uncertainty for awhile now - this is one of those areas where ecologists are coming late to the game, so I'm playing alot of catchup. I just came across a really nice editorial on the perils of risk modelling in the financial world by Roger Pielke Jr. from earlier this year. Here's a nice quote that I like (for reasons that should be obvious):

The general lesson to take from such experiences is that it is rarely the models that are at fault; it is instead the use of those models in ways that are inappropriate and can lead to flawed decisions, sometimes with very large consequences. Too often the models are treated like black boxes and their use is overlooked as being the domain of technical experts. To make better use of risk models in business decisions, we need to open up the black box and better understand the role of models in decision making. Because of the potential for conflicts of interest, it is important to have independent eyes looking at the models and their use, a role that too often goes overlooked.

Although Roger is talking about hedging investment funds, I think the same could be said for models of risk to endangered species. Building the models is easy - even incorporating risk, error and uncertainty is not difficult. But understanding the outputs of those models demands a certain level of knowledge. And worst of all, we have not worked hard to understand or develop decision making in applied ecology, let alone understand the use of models in decision making. The politicization of climate change science should stand as a warning to us - as the stakes rise the challenges of honestly providing scientific advice and maintaining credibility become much greater.

What do we do about it?

1. Ecological managers at all levels must recognize that they are making decisions.
   
2. Making decisions implies some effort, however implicit, to forecast what will happen in the future after the decision is made.
   
3. We have excellent theoretical tools for forecasting the future of populations and ecosystems (not communities, unfortunately).
4. Ecological managers should be trained in using those tools while making decisions.

As an educator I can try to fix 4 for future managers. The rest of you are on your own!

Staying supple

I spent the morning in the company of a great crew that works on implementing AM for the Central Platte River in Nebraska - the Platte River Recovery Implementation Program. One of the things that is great about the program is the well worked out and detailed governance procedures - although the governing committee gives lots of folks a hard time, it also provides a clear "chain of command", and a place for stakeholders to have their voices heard.

One of the things that's not so good is the lack of flexibility created by those same detailed procedures. Two things in particular that changed after the agreements were signed 1) Phragmites australis spread throughout the central Platte, and 2) Ironoquia plattensis, a new caddisfly species was discovered. Surprise! The future is not like the past. This is exactly the situation that AM should be able to deal with - new information that changes how management actions will play out. Phragmites is problematic because it dramatically changes how the vegetation community will respond to short duration high flow events - it is very resistant to scouring and drowning. It seems as though the reality of Phragmites is settling in, but it took a tremendously long time, and caused much angst. So the key problem is how to design a program that is sufficiently well tied down that people will sign on, but remains flexible enough to deal with new circumstances far outside the scope of the original concept.

Maybe too much to ask, especially when the stakes are high.

Postmortem AM

One of the things that continually amazes me is the extent to which collecting simple data during management activities can be used to refine the management actions. It is one case in which the perfect is sometimes the enemy of the good - when an understandable desire to have the science "done well" prevents the collection of simple data that might be "good enough". In these days of GPS units in phones, it would be great to have the time and location of management actions, at a minimum.

That said, it is always nice when data collected during a management action is well analyzed. There's a nice example of that from the Outer Hebrides, where Thomas Bodey and colleagues used stable isotope analysis on whiskers collected from culled American Minks. The carcasses are in the hand - why not extract information from them, if it will help make the eradication campaign more efficient. The nice thing about an eradication campaign in an archipelago is the clear iterated nature of the decision making - distribution of effort in habitats on new islands provides a nice opportunity to apply lessons learned from previous campaigns.

And in other news, check out Methods.Blog, where Rob Freckleton is collecting samples of new methods in ecology and evolution from across a wide range of journals. He is also the editor of a new journal from Wiley-Blackwell devoted to methods in ecology and evolution. Looks interesting!

Obama does AM!

From a Washington post article:

Obama discussed his professorial leadership style in a recent interview with U.S. News & World Report. He said he is not afraid of doubt and is comfortable with uncertainty: "Because these are tough questions, you are always dealing to some degree with probabilities. You're never 100 percent certain that the course of action you're choosing is going to work. What you can have confidence in is that the probability of it working is higher than the other options available to you. But that still leaves some uncertainty, which I think can be stressful, and that's part of the reason why it's so important to be willing to constantly reevaluate decisions based on new information."

Nice. But clearly he hasn't read Frank Knight's book "Risk, Uncertainty and Profit" because he describes uncertainty as probability. Environmental managers in the public service take heart, because your boss is comfortable with uncertainty. Grapple with it, please.

Credibility eliminates uncertainty

or rather - honestly incorporating uncertainty into forecasts or projections can undermine the credibility of the forecast. And worse:

It is known that perceived usefulness (which presumes credibility) in the eyes of policy makers often depends on whether the forecast (or other expert input) corresponds to the potential user’s preconceived beliefs, ...

This from the 2005 report of the National Academy of Sciences on decision making for the environment.

I knew it! Expressing prior beliefs as prior distributions is important. Now, what to do about undermining your own credibility by demonstrating the effects of users beliefs on your forecasts .. .

Predictions, Forecasts, projections and scenarios

OK, so looks like the theme for the week is why I'm not making predictions. Or rather, that I am making predictions but really I ought to be building scenarios. Audrey Coreau and colleagues have a nice paper in Ecology Letters (2009 12: 1277–1286) on "The rise of research on futures in ecology". Their essential thesis is that we need to spend more time talking about what hasn't happened yet, i.e. the future, because it will help strengthen the basics of our science. Oh, and incidentally it will help with decision making. They build on the distinctions MacCraken made between predictions, forecasts, projections and scenarios primarily emphasizing the value of qualitative scenarios for developing possible futures. They do see a role for predictive models within those scenarios, so phew, I'm not out of a job yet.

I find their definition of a projection curious: "a statement about what would happen, based on the extrapolation of past and current trends (e.g. population projections)." The way they use the term projection it is based on observed data extrapolated out into the future - assuming no change in what is presently going on. This is consistent with how Hal Caswell describes matrix projection models of populations, although those are quite different from simple extrapolation based on trend data. What makes this definition curious is that it is different from how MacCraken defines a projection: "a projection is a probabilistic statement that it is possible that something will happen in the future if certain conditions develop. " Specifically - what happens if conditions ARE NOT the same as they are now, and in particular, if conditions are affected by management options between now and then.

Hmm, so I guess I'm making projections sensu MacCracken.