on sabermetrics

[Joust Williams]

I remember having to look at data collected over a couple of years to estimate the time it took to manufacture each of these tools. The method of collection changed at an indeterminate time (which I could eyeball and find a decent cut-off), the machines' efficiency were wildly different (and naturally this was not captured consistently), the data was simply wrong a lot. Each type of tool has different rules (there's like 8 variables per tool), none of which I actually concretely know. By the end of it, I had nothing left to work with.

So the only thing I could think of doing was to go to the source and figure out the time it takes to do each subprocess. I'm not Einstein. But I could have made something up with basic linear regression and ignored educating people on coefficients of determination and people would have been OK with that. And I am sure this happens all the time.

Sometimes, the linear fit was OK (well, it fit the data but who knows if it was right or not) and sometimes you could have nice things like "this part of the process takes about negative 4 minutes to do". Nice.