3 ms·

> I can’t imagine that this calculation is meaningful in any way. The errors don’t average out here; the curve we’re modeling is exponential, tiny errors in ass

by boulos 7y ago

> I can’t imagine that this calculation is meaningful in any way. The errors don’t average out here; the curve we’re modeling is exponential, tiny errors in assumptions can lead to roughly 100% error in the output. Isn’t there a word for when a lot of math and complexity and specific numbers with names imbue an article with some extra sense of being true or having authority? The official term for the issue is False Precision [1], though I usually just say “false sense of precision”. If someone does actually carry through and calculate margins of error due to their varied assumptions, it lets the reader decide. Unfortunately, I’ve rarely seen folks treat errors as anything other than independent, unless they’re actually building a model that they evaluate with a Monte Carlo method (e.g., system temperature and cpu busy-ness are correlated when machine failure rates). [1] https://en.m.wikipedia.org/wiki/False_precision https://en.m.wikipedia.org/wiki/False_precision

dahart 7y ago
That’s pretty good, and does apply here I think. But FWIW I was thinking about something slightly closer to Occam’s Razor and/or the Law of Parsimony, but specific to how adding more data or assumptions or ‘evidence’ can sometimes make something seem more true, even though it’s the opposite.
- boulos 7y ago
  Yeah, that’s how you ended, but I think the root cause in the piece you linked isn’t “deception” but false precision. The false precision is what gives the sense that it seems true. Perhaps this is a variant of curse of dimensionality for errors? That is, if each piece of evidence is only 80% “likely”, 80% to the Nth drops like a stone. Perhaps Dave’s Law? :).
  - dahart 7y ago
    Ooh, I like the sound of Dave’s Law. :P More seriously, I love the idea of the curse of dimensionality for errors. Random tangent, but phrasing it that way totally reminds me of the phrase ‘flaw of averages’ and the story about US jet fighters. I wasn’t thinking the author was being intentionally deceptive. It can feel like accounting for more things makes something more accurate. Sometimes it’s actually true, sometimes failing to account for all the relevant inputs is a real problem. It’s legitimately difficult to know when you’ve accounted for too little or too much, right? I guess this is just one of those times when each piece being accounted for is making the result less accurate rather than more accurate, because the data truly does have a false precision, like you said. Didn’t Google go on work-from-home a week ago? I did mid-week last week and I’m feeling stir-crazy already, but starting to wonder if the most responsible thing to do wouldn’t be to catch the virus right now and quarantine myself; guarantee I don’t spread it around.

by boulos 7y ago

dahart 7y ago

That’s pretty good, and does apply here I think. But FWIW I was thinking about something slightly closer to Occam’s Razor and/or the Law of Parsimony, but specific to how adding more data or assumptions or ‘evidence’ can sometimes make something seem more true, even though it’s the opposite.

boulos 7y ago
Yeah, that’s how you ended, but I think the root cause in the piece you linked isn’t “deception” but false precision. The false precision is what gives the sense that it seems true. Perhaps this is a variant of curse of dimensionality for errors? That is, if each piece of evidence is only 80% “likely”, 80% to the Nth drops like a stone. Perhaps Dave’s Law? :).
- dahart 7y ago
  Ooh, I like the sound of Dave’s Law. :P More seriously, I love the idea of the curse of dimensionality for errors. Random tangent, but phrasing it that way totally reminds me of the phrase ‘flaw of averages’ and the story about US jet fighters. I wasn’t thinking the author was being intentionally deceptive. It can feel like accounting for more things makes something more accurate. Sometimes it’s actually true, sometimes failing to account for all the relevant inputs is a real problem. It’s legitimately difficult to know when you’ve accounted for too little or too much, right? I guess this is just one of those times when each piece being accounted for is making the result less accurate rather than more accurate, because the data truly does have a false precision, like you said. Didn’t Google go on work-from-home a week ago? I did mid-week last week and I’m feeling stir-crazy already, but starting to wonder if the most responsible thing to do wouldn’t be to catch the virus right now and quarantine myself; guarantee I don’t spread it around.