[17:15:05] If I'm reading the schedule correctly, office hours here began 45 minutes ago? [17:44:35] rfarrand: ^ [18:25:45] Hello [18:27:54] hello [18:29:30] we will be starting this tech talk just a few min late [18:29:54] Slides for today: https://docs.google.com/presentation/d/1IyM2Rtyquhjz2hKKPMh69pVps1p9gYykKi55nHCxRxk/edit#slide=id.g10a5755072_0_181 [18:30:15] If you are watching the tech talk live, please don't read ahead on the slides, because it starts with a quiz :) [18:31:11] so we should be prepared to write/type some answers [18:31:56] Yes, get your pencils out [18:34:11] ok, we are live [18:34:19] people watching remotely will be able to see shortly [18:34:39] * greg-g waves [18:34:54] * greg-g is watching/listening on the youtube stream [18:34:57] we have 6 people watching remotely and 1 (+ joel) in SF [18:35:04] 7 people watching remotely [18:37:04] any questions about how the quiz will work? [18:37:20] how much time will we have for each? [18:37:29] 1 min for each question [18:37:35] thx [18:38:26] I assume I shouldn't just put: 0% - 100% :P [18:38:39] that would not be 90% of the time [18:38:48] 0-90% [18:39:04] (I know that's wrong :P ) [18:39:46] That's a great question I won't answer, greg-g [18:43:10] wow .. these are interesting :) [18:47:21] yup [18:47:52] guess how many questions you got right [18:48:07] 3 - 5 [18:48:15] 5 [18:48:16] With large ranges, 6-8 [18:48:17] 4-7 [18:48:25] 5 [18:48:28] 2-5 correct [18:48:29] 2-8 [18:48:58] heh csteipp :) [18:49:06] 1) 7.3% [18:49:12] 2) 1 million [18:49:22] 3) 290 [18:49:33] 4) 265,518 [18:49:45] 5) -182.9 C [18:49:57] 6) 10 million [18:50:17] 7) 9575 km [18:50:31] 8) 4.5 years, men older [18:50:45] 9) 118,406 [18:50:59] 10) 223 m tall [18:51:20] 4/10 ain't bad, right? :) [18:51:26] oops [18:51:57] counting 4.5 as 5, I got 6 [18:52:08] this is my second time taking the quiz, and I got 8/10 :P [18:52:13] 6 [18:52:16] meeple27: :) [18:52:23] 0/10, but only because i joined this talk 20 minute late :( [18:52:25] if i coult my 3 near misses .. that would be 5 out of 10 ;-) [18:52:29] me (4) [18:52:29] 4.5 year age diff = 5 years [18:52:41] question 8 is 4.5 [18:52:48] ;-) [18:52:53] yes, rounding off age diff. [18:52:54] 5 to 20 [18:53:01] i picked 5 - 15. [18:53:07] so, that is a near miss for me. [18:53:19] ok. 5 then [18:53:43] 2 here [18:53:44] first time i did this I got 6/10 [18:53:45] 1 [18:54:09] 2 exact. [18:55:02] 2 exact; 3 very close (like 10K for 9575); and 5 which are off by factors of 4x / 5x consistently (which is interesting). [18:55:30] it was only re that one question :P [18:59:20] the oxygen was the easiest .. because if it was anyway close to temperatues reached on earth .. oxygen there would become liquid ;-) .. so, i picked a range lower than known cold temps on earth. :) [19:00:23] anyone else want to explain their answers? [19:01:01] for the great circle distance i used airline speeds, approx # of hours .. to estimate and came pretty close. [19:01:09] The 2 answers I got wrong this time...I got both right the first time! Doh! [19:01:26] 500 miles / hour and about 10-15 hours is what i figured. [19:01:44] but, i thought that sounded too small and bumped it up from 8K to 10K on the lower end. :) [19:01:48] Outside software, I recommend "How Many Licks?: Or, How to Estimate Damn Near Anything" by Aaron Santos. [19:03:20] 10K - 15K [19:03:38] i had initialy picked 8K - 14K. [19:03:38] but, bumped it to a bigger number ;-) [19:03:39] I had 10K too [19:04:14] but, the interesting part is that for 5 of the answers, my lower bound was 5x higher than the correct answer. [19:04:16] I did 8-50k, because I thought "I have no desire to try to figure out how big the earth is right now in km) [19:04:22] s/)/"/ [19:04:35] any questions? [19:04:48] anything to bring up before we end? [19:04:49] greg-g: That was the only one I got right. [19:04:59] greg-g: the equator's length is 40k km ;) [19:05:04] greg-g: Because I'm constantly estimating airline mile accumulation. [19:05:21] i'll have to ponder that ;-) [19:05:24] mobrovac: good to know :) [19:05:46] dapatrick: are you a mile collector? [19:05:49] i tihnk i kept bumping up a lot of times based on my initial assumptions. [19:05:58] i.e. assumed that my first guess was too low. [19:06:20] greg-g: I was in a previous life. And the habit stuck. [19:06:43] I tend to do a rough estimation, then divide by ~3 and multiply by ~3 to get a ~10x range. [19:06:44] How many people correctly estimated how many they correctly estimated? [19:06:53] dapatrick: :) [19:07:10] I said 6-8, and I got 5.5, so not quite. [19:07:12] jaufrecht_: I estimated 2-5 and got 4 [19:08:03] The problem with trying to be 90% accurate on things you know little about is that you need really big ranges, but really big ranges are embarrassing. [19:08:37] as with software development :) But yeah, it was weird to estimate things you had no idea about [19:08:39] I've gotten comfortable with a 10x range as a "close enough" answer in many cases, but I don't like going outside that. [19:08:57] it would be like me (who has never touched mediawiki) trying to estimate how long it would take to re-implement flow from scratch [19:10:03] I think there is also room for bias based on whether one believes the C-section and car crash numbers were accurately and completely reported. [19:10:15] The joke rule for estimation has a kernel of truth in it: double your number and bump up the time unit to the next larger. [19:10:31] (in software) [19:11:09] jaufrecht_: thanks for the interesting exercise and fun discussion! [19:13:04] Trey314159, what specifically makes it embarrasing? [19:14:12] I think it's admitting that you know so little [19:14:12] I also feel some sort of tug to keep the ranges smaller, although it wouldn't have occured to me to label it embarassment. I wonder what benefit we get from having a System 1 bias toward false precision. [19:14:20] Perhaps embarrassing isn't quite the right word. But an answer like "0 to 5B people" is of such a large scope that it is (or feels) worthless. [19:14:46] also it almost feels like you're trying to be ridiculous [19:15:01] under what circumstances is a precise and wrong answer better than an accurate but utterly unprecise answer? [19:15:28] a precise and slightly wrong answer is pretty valuable [19:15:30] That feels like the wrong question. [19:15:40] a precise and wildly wrong answer is worse than useless [19:16:03] it seems tied directly to what you will do with the information. [19:16:05] When is your answer so imprecise that it's the same as "I don't know" and is IDK an acceptable answer? [19:16:42] Sometimes IDK is not acceptable (e.g., estimating software), so "1-5 years" is not acceptable, even if it is true. [19:17:06] what if IDK isn't acceptable? What makes it better to be knowingly probably wrong? In the example of public works, it generally accepted by policy makers that big projects will always end up costing more than the estimate [19:17:39] but that this is simply how our (US, at least, and I think most industrial countries') public planning process functions, and it's not "lying" so much as working within the system. [19:18:25] i.e., if the public was exposed to more accurate estimates, people would not support large infrastructure projects because they in turn have cognitive biases that discount the long-term benefits. [19:19:13] How does that apply with customers, users, donors, managers? How much agency and power are we negotiating over when we make estimates? [19:20:05] and more tactically, if you are embarassed to provide an IDK answer, what about an "I don't know, and I would have to do X and Y in order to have a more precise and accurate estimate." ? [19:21:12] Too many questions that require longer answers than IRC makes plausible. I try to improve my own thinking and estimating (learning about System 1 and 2, reading How to Estimate Damn Near Anything, learning to say "I don't know"—Freakonomics has a nice podcast on that), and give the best estimates I can when asked. [19:23:01] I'm not embarrassed to say IDK (and I'd rather do that than give a uselessly large range) but I think large ranges feeling useless trumps realistic estimating of 90% interval for answers waaaaay outside your area of expertise. [19:24:23] We all knew ranges that would guarantee results ("0-100%") but didn't use them because they are useless / seem to defy the spirit of the exercise / feel like not making a real guess. [19:29:32] the exercise would have been easy if the goal was 100% instead of 90% :P [19:42:15] Coding Horror has a post (from 2006!) about estimating (software and random facts), and the comments hit all the high points of the difficulties of doing the exercise: http://blog.codinghorror.com/how-good-an-estimator-are-you-part-ii/ [20:13:08] jaufrecht_: last thought—are you familiar with calibrated probability assessment? There are online quizes. You answer and say how likely you are to be right, then see if you actually are right, for example, 70% of the time that you said you'd be right 70% of the time. [20:22:40] Trey314159: sounds cool.