🧵 Untitled Thread
Anonymous at Wed, 6 Nov 2024 22:48:08 UTC No. 16465434
I still don't get statistics. Most of it just seems bullshit guess-work to me. Just because you have a sample of the population, there is no way to really quantify how good that sample is. Your sample can legit be wrong. I get that you're going to try and account for, say in polling, demographics and popular interest categories and ensure they are accounted for first so it is more accurate. But what is it about having a 1000 sized sample in a population of 350m that allows you to have any confidence whatsoever?
I also don't get distributions. Yes things in nature follow a distribution, but some things won't. Like preferences of food. Obviously it's logical to say most will like the most popular and there are outliers, but the idea it would fit perfectly in a distribution seems odd to me. Or is it more that, by necessity, for some reason I don't get mathematically, things must follow a distribution?
I also don't get why a standard deviation is useful, or even accurate, or why a chi-square is accurate for accepting data? Why does standard deviation matter? OK, so you know the variation and volatility of data which I guess is helpful, but it will almost always be a sample. So is this good for making greater estimates of larger populations if you know your sample is more volatilte? And no this isn't post-election cope, I just don't get why people put stake in pollers, and I have been calling Alan Lichtman a retard for years.
I also know this is googlable, I just can't get my head around even the explanations for retards.
I want to work with stats so I want to understand this very abstractly, almost theologically, so I'm not just someone following a formula and regurgitating shit.
Anonymous at Wed, 6 Nov 2024 22:49:33 UTC No. 16465437
im a nigger
Anonymous at Wed, 6 Nov 2024 22:53:24 UTC No. 16465441
Also is a Z score just a way to tell you how much you fit into a normal distribution?
Can a theologian explain these to me? Why they are useful in understanding God's work.
Anonymous at Wed, 6 Nov 2024 23:55:45 UTC No. 16465487
why is bessel's correction n-1 and not n-(1-df)
Anonymous at Wed, 6 Nov 2024 23:56:47 UTC No. 16465488
>>16465487
sorry, n-(1+df)
Anonymous at Fri, 8 Nov 2024 22:32:16 UTC No. 16467648
>>16465434
Actual real life statistician here.
The thing is that if you can guarantee that the people being surveyed were picked independently and you have a large enough sample, then the chances of being off become smaller and smaller. Notice that the hard part here is being able to pick respondents independently and not really the sample size. For most applications, a sample size of anywhere from 30 to 100 can actually pretty accurately reflect a true population on millions. The more people you poll, the more confident you can be of your prediction and the smaller the confidence interval is going to be.
When you talk about election polls, two issues are that the actual number are so close to 50% that you need a really small confidence interval to really make an accurate prediction. If you poll a candidate at 51% give or take 3% then your prediction isn't all that useful. But if the same candidate is polling at 70% give or take 3%, then the chances of being off by 20% and the candidate actually losing the election are negligible.
The main problem is that you can't really sample independently, so you have to manually bias your sample to correct for that.
When you think of a distribution, you're likely thinking of a histogram. Distributions aren't the same thing as histograms. And to be fair, yes, most things in nature don't really follow a distribution nicely because statisticians can't even agree what a probability really means. Either way, there are plenty of distributions that support frequent outliers (leptokurtotic, or "fat-tailed" distributions). The most accepted definition of a probability distribution is that if you could sample an infinite number of points, or perform an experiment an infinite number of times, then your histogram would actually become the real underlying probability distribution. If you are a based onionsboy then you would say that probability is your subjective degree of confidence.
Anonymous at Fri, 8 Nov 2024 22:40:29 UTC No. 16467652
>>16467648
Anyways, I got tired of answering your questions so I'll just leave you with a series of books for you to read if you want to understand statistics. In order:
A First Course in Probability - Sheldon Ross
Mathematical Statistics and Data Analysis - Rice
Applied Linear Regression - Weisberg
Probability and Random Processes - Grimmett
Statistical Inference - Casella and Berger
Probability and Measure - Patrick Billingsley
Theoretical Statistics, Topics for a Core Course - Keener