I’ve written before about mercury emissions, mercury as a commodity, and mercury use in artisanal mining. But the reason we pay so much attention to mercury is because of its human health impacts, and these are primarily caused by eating contaminated seafood.
Different types of seafood have different amounts of mercury. Because mercury is bioaccumulative, organisms that are higher on the food chain tend to have greater mercury concentrations. Of course, the particular environment where the organism lives also plays a big part.
Scientists have been interested in the mercury content of seafood for decades. Recently, a group of researchers undertook the herculean task of aggregating data from almost 300 studies. The result is the Seafood Hg Database (and an accompanying paper). The database contains the mean mercury concentrations measured in each study for one or more of 62 seafood categories. Overall, the database represents over 62,000 individual measurements from around the world.
It’s a great dataset to play around with and experiment with visualizations. In the graphic below, I plot mercury concentrations for a subset of common seafood types. Each circle represents the mean concentration measured in one study, and the size of the circle is proportional to the number of samples in that study. I’ve overlaid box plots for each seafood category that show the median of all the means, as well as first and third quartiles (whiskers go to 1.5x the IQR).
I think this is much more instructive than simply plotting the grand mean (average of all the study averages) for each seafood category. For one thing, you lose a lot of information on how much mercury concentration varies within a category. Take tilefish, for example. This is one of the species that EPA and FDA advise pregnant women not to eat. But there are relatively few studies of tilefish, and the mean mercury concentrations they measured vary by an order of magnitude.
Click on the image below to bring up the full interactive Tableau Public visualization:
Click on the image to see full version
geovisualist 
This seems to suggest that the data is less certain than I expected. Or is that unfair? Is a very large sample essential to making a conclusion about the safety of a species?
I don’t think the issue is sample size. There is just a lot of variability within these categories. Part of that variability is a function of geography. Fish from some areas have more mercury than others. There is also variability in the species that comprise these categories. Some species of mackerel, for example, have higher mercury levels than others. All this makes it difficult for the consumer, who probably does not know where the fish was caught and what exact species or subspecies it belongs to.
One interesting question, which I may explore in a future post, is whether there has been a change over time in any of these species.