3 ms·
If it's your thing, I highly recommend NNT's YouTube channel [1]. He uploads short videos (that he apparently films with a potato), where he explains stat conce
by narush 5y ago
If it's your thing, I highly recommend NNT's YouTube channel [1]. He uploads short videos (that he apparently films with a potato), where he explains stat concepts "for beginners." Of course, he also spends lots of time _roasting_ random researchers/statistics users/noobz.
Actually though, I've actually a bunch of his videos super educational and well made - if you can get past the terrible camera / sound quality.
I watched his explanation of the central limit theorem [2] the other day and found it super helpful. I'm not great at math, but I took multiple stat classes at uni (including two where we went through proofs for the CLT at some level). But this single 10 minute video helped me "get it" more than any class. The big observation that I was missing is that the sample mean is pretty much the same as the sample sum, and the sample sum can sum to "things in the middle" in more ways that things not in the middle (e.g. if you have two dice, rolling a 7 is more likely than rolling a 12).
I'm sure I'm butchering it mostly, but that little bit of intuition was just *nice* - especially after 2 stat classes of rote proof following w/ little to no intuition building.
[1] https://www.youtube.com/channel/UC8uY6yLP9BS4BUc9BSc0Jww https://www.youtube.com/channel/UC8uY6yLP9BS4BUc9BSc0Jww
[2] https://www.youtube.com/watch?v=bfM9efdStN8 https://www.youtube.com/watch?v=bfM9efdStN8
- jdmoreira 5y agoWow! I didn’t know about this, thanks for the recommendation. I love this guy, it’s a guilty pleasure for sure because of his personality but I find it somehow refreshing.
- shannifin 5y agoThanks for the links! Wish he would write a full-length book of "stats for beginners" so I could better understand his critiques of its misuse, but this looks like a good place to start anyway.
- hogFeast 5y agoIf you know a programming language, stuff like Wolfram (or doing it in Python/R) can really help. Stats suffers quite heavily from having to introduce some ideas, most importantly distributions, which can be tricky to intuit at first. Running through some examples with Wolfram (there is a free version which you can in Jupyter) or using the stats package from scipy in Python was very helpful to me personally.
- jplr8922 5y agoI agree. I once spoke with a statistics professor in charge of the undergraduate program at my university. I asked him tons of questions about the cursus, and swiftly replied "Yes the program is far from perfect. However understand that statistics is still a new field compared to calculus. I do not think that we know how to teach all of this correctly...". I wish Taleb could write a book about it!
- pram 5y agoDoes that stuff work in Wolfram Alpha?
- fareesh 5y agoWhat is the cloth around his neck?
- halikular 5y agoI think it's his shirt.
- subroutine 5y agoI know of scarcely anything so apt to impress the imagination as the wonderful form of cosmic order expressed by the law of frequency of error. The law would have been personified by the Greeks if they had known of it. It reigns with serenity and complete self-effacement amidst the wildest confusion. The larger the mob, the greater the apparent anarchy, the more perfect is its sway. It is the supreme law of unreason. - Sir Francis Galton, on the Central Limit Theorem Two interactive demos... http://mfviz.com/central-limit/ http://mfviz.com/central-limit/ http://www.ltcconline.net/greenl/java/Statistics/clt/cltsimulation.html http://www.ltcconline.net/greenl/java/Statistics/clt/cltsimu...
- fny 5y agoI have very mixed feelings about these videos. Some don't quite build enough intuition, others are extremely handwavy. Also, Taleb often disses certain methods without providing alternatives or entertaining the idea that alternatives already exist. My recent favorite is his rant about correlation where he shows that linear regression can yield a slope of zero on certain piecewise examples. He just makes fun of regression and moves on without acknowledging piecewise and kernel regression exist to address those cases. Instead he'll make some comment about information theory being the solution without any further elaboration. Sure correlation can be abused, but so can mutual information and any other information theoretic metric you choose.