There’s a lot of press happening right now about an “unusual mortality event” underway in the northern Gulf of Mexico involving dead dolphins, particularly young ones (see for examples here, here and here). Much of the drama that drives these stories to the front page arises out of potential links to the Deepwater Horizon oil spill last year. It’s important to critically evaluate these sorts of reports, however, and not to allow traditional press/media to create causal relationships for us, based on an editor’s impression of a perceived correlation. So, lets have a look at the data then. Here’s the official word from NOAA:
What I first see when I look at this graph is an annual cycle of mortality in ceteaceans in the northern GoM, with a peak in March (blue columns), then a bad year in 2010, when 3 times as many animals were found as the 02-09 average. For 2011, I see the peak occurring earlier than usual (Feb) but the number of animals per month actually similar to 2010.
What does this mean? Well, to me it means dolphin deaths are not that rare to start with and that this is the time of year for it (i.e. it’s not actually “unusual”, it’s “seasonally-appropriate”). Secondly, it means that it’s not much worse in magnitude than last year, albeit peaking a little earlier. This is important because the Deepwater Horizon disaster didn’t occur until late April, so could not have been responsible for the high numbers in March and April of 2010. To put it another way, 2011 is unfolding much like pre-spill 2010. Thirdly, it suggests to me that NOAA folks could do better with graphing. I say that because to properly interpret this chart, the blue columns representing the average number of deaths 02-09 really, REALLY need error bars! An average on its own is somewhat useless without a measure of spread or variability to go with it. As it stands, the graph makes it look like 2010 and 2011 were way worse than the 8 preceding years, but actaully there might have been equally bad years in that series too, but we can’t tell because the interannual variation is canceled out by years of lower mortality. In other words, important information was lost in the averaging process. I might also ask what qualifies as “unusual mortality event”? NOAA says
“a stranding that is unexpected; involves a significant die-off of any marine mammal population; and demands immediate response.”
All three elements of that definition involve humans! The first – unexpectedness – involves our expectations and therefore a certain subjectivity. The second – significance – is similarly bound up with expectations, unless they mean statistically significant, in which case see my earlier comment about error bars. The third – requiring immediate response – is a real can-of-worms. There can be all sorts of reasons to respond, including public pressure from media hype. Point is, this is not a very objective definition of a UME. I would rather see a scientific definition, something like a monthly mortality value >2 standard deviations above the long term average, or similar.
When I look at what I’ve just written, I can easily see why the science is so often neatly excised from the discussion of these sorts of things in the media; it seems pedantic, and sciencey. It’s so much easier to just syllogize: oil was spilled, dolphins died, therefore oil killed dolphins. But in these cases, the devil IS so often in the details. I am not saying oil wasn’t involved; I’m just saying we can’t infer causality from correlation without a closer examination of the data.