←back to thread

307 points MBCook | 1 comments | | HN request time: 0s | source
Show context
bunderbunder ◴[] No.42151125[source]
I'd love to see some sort of multiple regression or ANOVA on this, instead of singling out a single variable. Is car brand really the best independent predictor? Or is it specific design decisions you tend to see in certain brands?

(Like, say, maximizing driver distraction by consolidating a bunch of essential controls and information displays into a touchscreen display that's really difficult to operate when it's sunny outside. Just to pick something at random, of course.)

Somewhat related, I was recently shopping for refrigerators, and fell down a data rabbit hole. If you just look at the overall style of fridge, French doors look like a terrible option from a reliability perspective. But then, digging in a bit more, it turns out that's kind of a spurious correlation. Actually it's the presence of bells and whistles like through-door ice dispensers that kill a refrigerator's reliability. And then perhaps on top of that the amount of extra Rube Goldberg machine you need to make a chest height ice dispenser work in a bottom-freezer French door refrigerator creates even more moving parts to break. But a those problems don't apply to a model that doesn't have that feature.

replies(10): >>42151205 #>>42151278 #>>42151294 #>>42151316 #>>42151413 #>>42151582 #>>42151717 #>>42151753 #>>42162767 #>>42176234 #
doctorpangloss ◴[] No.42151205[source]
It's an interesting perspective. I was recently shopping for shoes, and a fully closed shoe had more places where it could break compared to my flip flops. That's why whenever you are doing a dangerous activity, flip flops are recommended.
replies(2): >>42151248 #>>42151259 #
1. bunderbunder ◴[] No.42151259[source]
I'm not entirely sure an anecdote about the dangers of singling out just one variable is a great counterpoint to a criticism of the practice of singling out just one variable.