Not the most glaring issue with #JurassicPark, but this plot from Michael Crichton's original 1990 novel looks pretty darn weird. /1 
In context: This is the height distribution of the Procompsognathid dinosaurs ("compys") in the park. The distribution looks Gaussian (bell curved), and we would expect the height distribution to look approximately Gaussian if the dinosaurs were a natural population. /2
However, the dinosaurs in Jurassic Park weren't supposed to be a "natural population" --- their breeding was controlled. So this bell curve was used as a clue that the dinosaurs were mating. /3
Some minor issues: The plot above indicates 67 compys, but a few pages earlier we were told that there were 49. /4
Also, the plot seems to indicate fractions of dinosaurs at 27, 28, 29, 38, 39, and 40 centimeters. So I rounded these numbers and reproduced the plot in R. #rstats /5
It still looks weird. It looks too perfect. This is the big issue. Even if the dinosaur heights did look somewhat like a Gaussian distribution, we are just looking at a small number of dinosaurs and so we wouldn't expect it to look *exactly* bell curved. /6
I simulated some dino-heights from a real Gaussian distribution and rounded these heights to the nearest centimeter. I did this a bunch of different times so you can get a feel of what height distributions we would expect to see. /7
(technical detail: I simulated from a Guassian with the same mean and standard deviation as the sample mean and sample standard deviation, respectively, from the book's data) /8
Not so bell-curvy, right? We should be suspicious when things look too perfect. /9
Though, I've been assuming that Crichton's plot is supposed to be a histogram (where we have the counts of animals on the y-axis against the heights on the x-axis). But it's too perfect to be from real data, and he doesn't *explicitly* say that it's a histogram. /10
So what could it be? My guess is that it's some sort of density estimate, plotted only at centimeter intervals. A density is like a "smoothed version" of a histogram, where you try to remove all that noise that I noticed wasn't in Crichton's plot. /11
A density usually has an area of 1 under its curve, but Crichton's plot clearly has an area much greater than 1. So I think it's a density plot multiplied by 67 --- the mistaken number of Compys (I'm not going to explain away the 49/67 discrepancy). /12
The problem with this is that the density estimate still looks too clean. I made a few density plots in R (using the default settings) from data generated from a Gaussian. /13
Ok, all those ups and downs are mostly removed, but it still looks more spiky and more skewy than Crichton's plot. So I upped the smoothness and got this. /14
Some of those look pretty close! But I *really* had to up that smoothing parameter. That could make it harder to see non-normality. Like, suppose this was the true density: /15
That's clearly not normal, but using too smooth of a density estimate gets things like this, which still kind of look like Crichton's plot. /16
Whereas with better choices of the smoothing parameter, we can more clearly see that big peak and the tiny tails. /17
So to conclude: In addition to playing God with nature and blah blah blah, the folks over at Jurassic Park smooth their density estimates too much. /End
I think my undergrads were more impressed with this thread's like count than anything I've taught them this semester 😂
