Previous thread >>16256996This is one of the board's newest generals. Fairly high activity due to edge lords trying to be funny but instead spreading facts about the absolute state of our world.Intro stats is fairly easy, intermediate stats come the programming and we have already have several battles about what language is the best in the thread. Nobody uses SAS funnily enough, SPSS has had some people trying to joust the edgelords who are into R and C++, while the stata children are silent as usual.Come one, come all. State your dumb questions, /pol/tardy or not. Some fairly useful and funny math is showcased in this thread.
https://en.wikipedia.org/wiki/Central_limit_theorem Neat.
>>16287506>Fairly high activity due to edge lords trying to be funny>thread dies prematurely again even though I bumped it a couple of time.lol, lmao even.
>>16287507The discussion about when it applies or not applies in the last thread was actually very interesting. I wish they would discuss it more or source some interesting litterature on the subject pro and against.
Reminds me of a Twitter thread couple of years ago where someone posted a study done on gradschool of some uni about what they'd like to learn more in-depth to be able to make them better research. The respondents were primarily PhD and postdocs. Guess what the majority answer was, yes statistics. Why are majority of the PhD retards so dumb lmao
>>16287817That "research" probably oversampled some dumb humanity, psychology, biology or medicine... trash majors who never had to study statistics in school and get crapped out when most of sciences is just statistics in one form or another. The majority of those trash majors cant even understand what a random variable is. Pretty much of the field I work involves statistical modelling itself. Do I wish to know more about probablity and statistics? Maybe, but I dont think I can any use from a grad course or whatever. The majority of what I study are self read on demand from papers and other books.
>>16287727rule of thumb:does the value of one datapoint depends on the values of other datapoints?>yesclt doesn't apply>noclt should be ok
>>16287844No, the respondents were all STEM including physics. Physicists really don't have a clue on statistics
>>16288371you sound like a dumb humanity or psychology major who doesn't know how to read research properly.
>>16288391>humanityhumanities*
What's a good book to read to learn probability and statistics? I'm a physicist who doesn't know shit about statistics.
>>16288391I'm a physicist myself tho, none of my peers are well acquainted with stats. We just do what everyone does in the field.
bump, I am reading this course: https://mathweb.ucsd.edu/~mleok/courses/math273b/
>>16287901Though the large deviations principle and Cramer's theorem still apply in most cases even in the case of conditional dependencies between samples. If the process generating the samples is stationary (either strict or WSS) and the MGF exists and is finite for all real values then the LDP will apply even with conditional dependencies.
>>16288563In general people who believe strongly in physical/causal determinism don't tend to spend a lot of time researching probability. This is unfortunate because it leads them to not understanding how their measurements and equipment functions in real experiments and then coming to dumb conclusions sometimes.
>>16288882that's why I said it was just a rule of thumb, among other things it hedges for the tendency of people to call any limiting distribution a 'central limit theorem'that said, I can't think of any real-world examples that are (weakly) stationary and yet 'essentially non-independent' (i.e. not just a transformation of some underlying independent-increment series), so I don't see what LDP offers *in practice* over the basic LLN, CLT etc.(also not a fan of the assumption that all moments are finite)but maybe you'd know better
https://ocw.mit.edu/courses/18-s997-high-dimensional-statistics-spring-2015/
bump
I have list of 4 numbers eg (4,2,1,3). Now I have a new observation say 5. Is there any test to see if this new observation 5 is significantly different? I made these numbers up, irl these are sensor readings of some sort, but the sample size is really low.
>>16290896With 4 or 5 observations you cant do anything and will just have to assume the sensor is correct. If your trying to detect errors in the sensor itself you could try increase polling rate so you have more data to work with.30 seems to be a minimum number for stats to start working but it's still extremely low and not conclusive at all. Sample size is a common issue.In general terms you want to make an initial assumption about the distribution of the data and then compare them using a relevant test to see if it's significantly different.
>>16290924Alright thanks. I'm gonna have to wait a couple of weeks to get more data.
>>16290985Get somewhere in the range of 75-200 at the very least. I understand that some sensors are literally checking for few events that can be very far apart (Poisson) but you have to get at least 30 to be able to say something like the other anon said. You should aim for at least double and preferebly over 100.
>>16288502Do you know a bit of programming? Depending on what language you know, there'll probably be books. Some of those books are even free on GitHub pages. https://scipy-lectures.org/packages/statistics/index.htmlis a good start because I'm a pythonfag
>6 day old thread>25 postsjust one cringey samefag endlessly self bumping it's lame vanity thread that nobody cares about every time the thread reaches page 10. >>16290341>>16292339>>16294008lol even OP isn't interested enough in this thread to make meaningful replies
>>16294021This is kind of a flare up thread. Sometimes it has extremely insightful communication and thoughts. Sometimes it is just not there. I would say I have learnt a lot from this thread. I am working on an infographic for the thread but now I am looking for answers to some questions I think must be answered in it.
>>16294021Not op, I bumped it because I like probability and statistics but I find it hard. This thread could be a good learning point for me, place to ask questions etc
>>16294417It actually is. Just ask away.
>>16294028post the draft
>>16294028Also we should make a getting started guide or something and link at rentry.org. I can volunteer but I'm a python fag and I don't know anything on R, which statisticians love.
Everyone know that a confidence interval tells you that, for example, 95% of intervals will contain the parameter. So if you have a collection of 95% intervals, and select one at random, isn't the probability that the interval contains the parameter .95? Why is it not true that a randomly selected interval contains the parameter with probability .95?
How can it help me in real life?
>>16295219probability that a randomly selected interval contains the parameter is not necessarily 0.95. It's either 0 (if the interval doesn't contain the parameter) or 1 (if it does), but we don't know which one it is.Look into confidence level coverage and confidence distribution.
>>16295235In many cases to make a wise decision for yourself, critically evaluate all the hot garbage in the news. Simplest example is say a shitty news poll says 95% of the voters says vaccine had no side effect on them. Would you trust it? If you're statistically aware, you'll immediately realise you don't trust the poll yet because they didn't mention the sample size and other characteristics (eg were respondents healthy, young old etc).Of course if you're in sciences, you'll need it to test your results rigourously.
>>16294546Not now brother. I will in due time. I once started this thread because I love the subject. I named it, I have had very thoughtful and delightful conversation in this thread. Probably one of the last places I will visit forever on 4chan. I love you guys. I will help you gain some knowledge.
>>16295379Granted.
>>16287506Hello!What is the use case of hypothesis testing if we can just use known estimation methods such as maximum likelihood? Why I have to learn useless notations and techniques while I could use well established methods that can be also used for simulations?I am a beginner, just was wondering.
