This piece tells a fascinating story: a farm in Kansas that has had very bad things happening to it for years, all because someone wrote an app that doesn’t report variability estimates when it reports point estimates.
In other words, this mapping app has a hard time figuring out where things are, sometimes–suicidal people, shady businesses, criminals, etc.–and so it reports its best estimate, but doesn’t mention that it is an estimate, and certainly does not report the level of uncertainty.
In other other words: when all it knows is that something is in the US, it tells the user that the thing is right at the center of the US. Reasonable, right? It’s a stats thing: use the center when you don’t know the value. But app users don’t know it’s an estimate, so they show up at a farm in Kansas and make the residents’ lives a living hell.
Point estimates need variability estimates.
I knew the game of Go was complex (despite simple rules, etc.), but I had no idea it was this complex. Last month, John Tromp and friends reported finally calculating all the legal configurations on a 19×19 Go board. The number is 171 digits long. And it took months to calculate on a supercomputer. SUPER. COMPUTER.
In a nice, thorough blog post with plenty of simulation goodness, Daniel Lakens (an experimental psychologist in The Netherlands) demonstrates pretty convincingly that recent reports of the death of p-values have been greatly exaggerated. I find it telling that a Bayesian statistician is coming to the aid of p-values. I don’t even know what to compare this to, but it’s cool.
I occasionally write functions in R to help me with some aspect of teaching, usually teaching undergrad stats. I always intend to share them with others, but I don’t actually know many other people who use R for teaching. So why not share in blog posts? I’ll start now.
Disclaimer: This functions is untested by anyone but me, it might not work, and I provide no guarantee of accuracy. I mean, I try, but you know…
An R function for visualizing single-sample z-tests, a function I named zvis, produces graphs as below, using (usually) pretty simple options:
zvis(x=104, mu0=100, sigma=15, n=30)
I won’t paste the code here because it’s kind of long, but anyone interested can have it from this link –> zvis. The whole set of arguments can be seen after the break. If, by chance, anyone is interested in this function and wants more explanation of these options, or has suggestions for improving the function, please let me know. Continue reading