Also this week: compile ALL the p-values, grad students vs. abuses of power, Michael Phelps vs. pumpkinseed sunfish, p-hacking with covariates, one year’s worth of data = two year’s worth of data, a canonical R gotcha, and more. Including a contest to win 1000 Internet Points, and links from Brian!
William Playfair: inventor of the line graph, the bar graph, and the pie chart. Click through for cool pictures of some of the world’s oldest graphs. (ht Marginal Revolution)
Interesting potted history of algebra, calculus, and their effect on physics. Calculus as the “killer app” of algebra. Also the point that progress in physics often consists of reinterpretations of the same equations. (ht @noahpinion)
Ecology grad students often feel a strong urge to collect a second year’s worth of data, or data from a second site, or a second species, or etc. Stephen Heard argues that that’s usually pointless–or it would be except for the fact that reviewers often like to see it.
Econometrician Marc Bellemare comments on recent work on p-hacking with covariates in economics and political science. Equally relevant to ecology, and easily accessible to ecologists. If you’re ever going to run a general(ized) linear model in which some of your predictor variables are covariates intended as “control variables”, you should read this post and have a look at the papers discussed therein. (ht Economist’s View) Relatedly: Andrew Gelman points out that, in a hierarchical model, adding a predictor can increase the residual variance. Keep that in mind if you’re trying to use proportion of variance explained by a variable as a measure of its “importance”. Gelman also reminds us that, in a hierarchical model of observational data, the effects of an individual-level predictor and its group-level mean cannot be straightforwardly interpreted as “direct” and “contextual” effects, respectively, even if the data are a random sample from the population of interest.
Speaking of p-hacking…Simply Statistics graphs the distributions of published p-values in each of many scientific and social scientific fields. In every field except one, the distribution is strongly bimodal, with one peak at very low p-values and a second peak centered just below 0.05. In the exceptional field, published p-values have a unimodal distribution centered on very low p-values. +1000 Internet Points to the first commenter who can name the exceptional field without peeking. 🙂 No it’s not physics. And no, it’s not “ecology, evolution, and earth sciences” either, though in that field the second peak around 0.05 is relatively small compared to its size in most other fields. You may also be
interested depressed horrified interdepressified to learn that the distribution of published p-values in “complementary and alternative medicine” looks like the distributions for most other fields.
Terry McGlynn shares some data showing that preprints might be taking off in biology (specifically, on bioArxiv). Terry explains why he’s not jumping on the preprint train, at least not yet, while also saying that he doesn’t think his reasons should be anyone else’s. I’d say the same as Terry. Do what works for you. And here’s a counterpoint to Terry’s post. UPDATE: And here’s Manu Saunders’ counterpoint to the counterpoint. Includes the completely unsurprising observation that the vast majority of ecology preprints on PeerJ and bioArxiv had received even a single peer comment, none had more than two, and the few comments there were were not nearly as substantial as typical peer reviews at decent journals. It’s totally fine if you want to post your preprints in the hopes of getting substantial feedback. But you’re very unlikely to get it. And if you do get it, that makes you very unusual. More data here reinforcing Manu’s observation.
A canonical R gotcha: the colon operator takes precedence over arithmetical operators. I just ran afoul of this one and resorted to emailing Ben Bolker in desperation when I couldn’t figure out why the hell my code wasn’t working.
Australia: the only country in history to lose a war to birds. (ht @noahpinion)
It’s David Shiffman’s favorite week of the year! For some value of “favorite”. 🙂 Ok, the question of whether Michael Phelps can beat a great white shark in a swimming race is silly clickbait. Especially since you have to simulate the race, because the actual race would look like this. Here’s a more interesting and fun question: what’s the biggest fish Michael Phelps could beat in a 100 m swimming race? Obviously he can’t beat a great white shark, but equally obviously he could beat a dwarf goby. So where’s the crossover point? I’m guessing something smaller than a trout? I would totally watch Michael Phelps race, like, a pumpkinseed sunfish. Bonus: Phelps and the sunfish could actually race safely; it wouldn’t have to be a bogus simulated race. C’mon, let’s have #pumpkinseedsunfishweek! In the comments, please share your suggestions on the best fish for Michael Phelps to race, along with your rationale. Thread winner gets +1000 Internet Points. 🙂
Here is a very thoughtful group of graduate students taking on all forms of abuse of the strong academic hierarchy (starting from a sexual harassment incident) in a very wise and constructive fashion starting with department-wide conversations. Check out the resource tab on the above linked web page to see what they implemented at their home university. Seriously, if you want to do something about these problems check out what they’re doing. If you are interested in learning more about how you could extend this to your university they have a workshop Tuesday evening at ESA.
Mike Kaspari is boldly listing the 10 fundamental principles of ecology to organize his teaching. Read it for the science to see if you agree with his list, or read it to think about how this approach can help organize your teaching (Meghan conducted a similar exercise for intro bio).