This is awesome. So intuitive and interesting. Why did we ever use null hypothesis testing? With the computational power we have now, this should be the norm.
that's true in a certain sense that he should have written proportional instead of equal by avoiding the use of marginal distribution indication for scaling
Your teaching style is very effective. Explanation and pacing is very good and your voice maintains attention very well. Thank you for making this video, it was quite informative.
One question I would have on this, is how can you be sure you are not biasing your result using these informative priors? I believe the most conservative approach is indeed the uniform (equivalent to I don't know anything so everything is equally possible for me), but when I start getting "clever", choosing appropriate priors, I can't make a real hypothesis test with that because I already tell the coin to be 50:50 (while someone could have potentially given me a magic coin of 10:90).
I believe the point of the prior is to introduce bias responsibly. That is, they should probably only be used if the prior was decided on from previous experience and expertise, and creating a posterior distribution could be helpful in cases that you believe will generate similar results from previous experiments but only have a limited sample size.
Isn't there an error at 5:18 Shouldn't the beta distribution's a and b be 86 and 84 NOT 106 and 114 ???? as the mean of 86 and 84 gives the mean on the screen (0.506) ...... Whereas the mean of the beta(106,114) is 0.481
Thank you for making this video. I took statistics class before, but my knowledge is limited. Please add descriptive details so I can understand your video.
How can we specify belief power of prior? In this example alfa, beta=30. And we can assign 250 for both. There is no boundary for us to prevent assigning 250 instead of 30. In a real life data, if you assign powerful prior, this means you have a bias and you may have implemented pressure to information coming from data; otherwise you have come close to non-prior case.
I shouldn't be saying that loud but dunno about you, I find this prior distribution & Ledoit-Wolf shrinkage method for accrued efficiency very difficult to picture and don't get me started on these affecting eigenvalues instead of eigenvectors... it's a mess in my head right now... I really need to pull myself together
The Beta function evaluated in (1.0, 1.0) is the Uniform distribution. He says that he will asume not having any information about the probability of getting heads or tails. And for that he will use a prior with an uniform distribution: Beta (1.0, 1.0) = Uniform; so the probability of getting heads or tails has a uniform probability from 0 to 1.
@@Magnuomoliticus but how do you know how to accurately increase the parameters of the prior distribution ? The only thing I don't understand here is how he decided that beta(30,30) was a more accurate depiction of what he knows about the coin. why 30? And thanks for your previous answer.
@@12rooted Well that's a great question that I don't know the answer of. My first guess is that it's arbitrary which distribution you use. But let's wait if someone else can clarify that!'
Your animation were based on binomial likelihood and in Stata you choose Bernoulli likelihood are they the same if we remove the binome factor (choose (N,X)
No they are not the same, but a single stochastic variable with a binomial distribution can be described by several stochastic variables with Bernoulli distributions.
I have the same version of Stata as yours. However, my Bayesmh window doesn't have the "univariate distribution" option. What could be the reason? Can you give me a hint?
If the coin is held with heads facing up, what is the likelihood it will yield heads when it is tossed? If the con is held with heads facing up, what is the likelihood it will yield tails when it is tossed? If the coin is held with tails facing up, what is the likelihood it will yield tails when it is tossed? If the coin is held with tails facing up, what is the likelihood it will yield heads when it is tossed?
Hi, thanks for the video. What I wonder is, what are " default priors" when it comes to bayesian inference? As I understand, the priors are specific to each hypothesis or data, so how come some packages include these defaults? What do these priors entail?
What i dont understand is how is multiplying liklihood and prior distribution going to give us what we call the posterior distribution. If anything the product just seems like a random function
Maybe the video creator intended to explain Bayesian statistics, but did not. The concepts start to be explained, then there is a stepwise jump into mentioning prior and posterior probability, with the introduction of on screen equations but no further explanations - it's like it was read out of a technical manual that only 'insiders' know about. This then quickly turns into how to use the software/which buttons to press, which seems applicable to those who already know about Bayes and want to use the software - and not for those who want an introduction. So I'm sorry to say this video was not useful to introduce Bayesian statistics and I would recommend giving it a miss.
BAYESIAN STATISTICS IS AN EXTENSION OF THE CLASSICAL APPROACH. VARIOUS DECISION RULES ARE ESTABLISHED. THEY ALSO USE SAMPLING DATA. I LEARNED ABOUT THIS WHEN I WAS STILL IN HIGHSCHOOL IN ATENEO DE ZAMBOANGA UNIVERSITY, MY GRADES IN ALGEBRA ARE HIGH.
Haha, “I’m going give a relatively non-technical explanation…” then proceeds to speak entirely in words that have definitions specific to statistics. Most people who remember the definitions of all the words used probably also remember what Bayesian is. People who don’t remember or never did know the vocabulary used have no hope of learning here what Bayesian is.
And this is an explanation for who? For a bunch of statisticians? Certainly not for the 'unintroduced'! A few seconds into the 'introduction' and you are already using highly technical jargon! Perhaps you need a course in pedagogy first! 😅
This is a very bad introduction. You jumped from the absolute basics to straight up prior and posterior. I'm really tired of these videos that area dvanced videos as "beginner videos" in disguise. They really spam all of RUclips but don't provide any value. Please explain it more simply next time and please elaborate what each concept means that you introduce within a few seconds. Sorry for being this critical but I'm not here to learn and not to waste my time.
That was excellent explanation of the interaction between the parameters, thank a lot for putting the time and effort to do the animations
Hello how are you?
This is the best introduction to this that I've found online! Thanks!
Wow, my understanding acquired from this video is more than from dozen of hours on classes.
Same
It was the most comprehensive video with the amazing explanations about prior, likelihood, and posterior. Thank you so much for this wonderful video.
Hello how are you?
This is awesome. So intuitive and interesting. Why did we ever use null hypothesis testing? With the computational power we have now, this should be the norm.
excellent explanation. I had been surfing internet, for clarity
What to say, an excellent explanation of Bayesian updating, long life to Stata and its People!
Thank you Sir, the best explanation I found on youtube..
Posterior is proportional to the MLE x prior , not equal =
that's true in a certain sense that he should have written proportional instead of equal by avoiding the use of marginal distribution indication for scaling
Your teaching style is very effective. Explanation and pacing is very good and your voice maintains attention very well. Thank you for making this video, it was quite informative.
@4:30 what's the difference between credible interval and confidence interval? After reading about it made me even more confused...
One question I would have on this, is how can you be sure you are not biasing your result using these informative priors? I believe the most conservative approach is indeed the uniform (equivalent to I don't know anything so everything is equally possible for me), but when I start getting "clever", choosing appropriate priors, I can't make a real hypothesis test with that because I already tell the coin to be 50:50 (while someone could have potentially given me a magic coin of 10:90).
I believe the point of the prior is to introduce bias responsibly. That is, they should probably only be used if the prior was decided on from previous experience and expertise, and creating a posterior distribution could be helpful in cases that you believe will generate similar results from previous experiments but only have a limited sample size.
Hello how are you? I need some help
Thank you very much for the explanations of non-informative prior and informative prior. Very helpful for my research.
At 1:40, shouldn't the area under the graph be equal to 1? What does the y-axis represent?
.75x speed
2x
0.25x
Thanks
He was going too fast
he might sound like a regular human at .825 speed
Finally I understand this thing. Thank you.
Isn't there an error at 5:18
Shouldn't the beta distribution's a and b be 86 and 84 NOT 106 and 114 ???? as the mean of 86 and 84 gives the mean on the screen (0.506) ......
Whereas the mean of the beta(106,114) is 0.481
great vid! so informative
excellent explanation sir.....
Thank you for making this video. I took statistics class before, but my knowledge is limited. Please add descriptive details so I can understand your video.
How is it that you are able to neglect the probability of y for the posterior distribution function, which is normally on the denominator?
why is the posterior narrower at 5:15?
Awesome, thank you! Animations are really helpful.
What informs the choice of a beta?
How can we specify belief power of prior? In this example alfa, beta=30. And we can assign 250 for both. There is no boundary for us to prevent assigning 250 instead of 30. In a real life data, if you assign powerful prior, this means you have a bias and you may have implemented pressure to information coming from data; otherwise you have come close to non-prior case.
Thank you. The first video that makes me understand this reasoning in one go.
I shouldn't be saying that loud but dunno about you, I find this prior distribution & Ledoit-Wolf shrinkage method for accrued efficiency very difficult to picture and don't get me started on these affecting eigenvalues instead of eigenvectors... it's a mess in my head right now... I really need to pull myself together
1:25 Why does this mean? Prior = Beta (1.0, 1.0)
The Beta function evaluated in (1.0, 1.0) is the Uniform distribution. He says that he will asume not having any information about the probability of getting heads or tails. And for that he will use a prior with an uniform distribution: Beta (1.0, 1.0) = Uniform; so the probability of getting heads or tails has a uniform probability from 0 to 1.
@@Magnuomoliticus but how do you know how to accurately increase the parameters of the prior distribution ? The only thing I don't understand here is how he decided that beta(30,30) was a more accurate depiction of what he knows about the coin. why 30? And thanks for your previous answer.
@@12rooted Well that's a great question that I don't know the answer of. My first guess is that it's arbitrary which distribution you use. But let's wait if someone else can clarify that!'
Your animation were based on binomial likelihood and in Stata you choose Bernoulli likelihood
are they the same if we remove the binome factor (choose (N,X)
No they are not the same, but a single stochastic variable with a binomial distribution can be described by several stochastic variables with Bernoulli distributions.
I have the same version of Stata as yours. However, my Bayesmh window doesn't have the "univariate distribution" option. What could be the reason? Can you give me a hint?
Amazing! Thank you so so much! :)
great explanation
If the coin is held with heads facing up, what is the likelihood it will yield heads when it is tossed?
If the con is held with heads facing up, what is the likelihood it will yield tails when it is tossed?
If the coin is held with tails facing up, what is the likelihood it will yield tails when it is tossed?
If the coin is held with tails facing up, what is the likelihood it will yield heads when it is tossed?
would someone please tell me what is he saying at 0:28 ? thank you
I think he says: "Many of us were trained using a frequentist approach to statistics..."
Hi, thanks for the video. What I wonder is, what are " default priors" when it comes to bayesian inference? As I understand, the priors are specific to each hypothesis or data, so how come some packages include these defaults? What do these priors entail?
Vgl-AAa
@@jamiilax4163 ??
Thanks. Perhaps you do another video to call it part 0 as the building blocks for this part 1. Introduction that is :)
Chuck the new stata 17.1 has different command structure. Can you please redo the video for version 17.1.
Thank you. That was very clear and helpful.
What i dont understand is how is multiplying liklihood and prior distribution going to give us what we call the posterior distribution. If anything the product just seems like a random function
Thank you for this video its clear to me
Thanks . I love statistic.
that was so so helpful. thank you.
Maybe the video creator intended to explain Bayesian statistics, but did not.
The concepts start to be explained, then there is a stepwise jump into mentioning prior and posterior probability, with the introduction of on screen equations but no further explanations - it's like it was read out of a technical manual that only 'insiders' know about. This then quickly turns into how to use the software/which buttons to press, which seems applicable to those who already know about Bayes and want to use the software - and not for those who want an introduction.
So I'm sorry to say this video was not useful to introduce Bayesian statistics and I would recommend giving it a miss.
It was a really bad video if you’re actually trying to understand bayesian statistics
Brilliant video thank you a lot
Thank you for your kind help.
"Non technical"
3:07
Right.
Hi can someone explain why this form of probability is important ?
excelent video
how to calculate odd ratio in bayesian ordered logistic plz tell me
Ok so how has the Bayesian model been tested and demonstrated superior to other statistical methods. I'm always skeptical without hard evidence.
Hi,
On what depends the type of likelihood distribution?
Thanks,
amazing! thanks!
Please, could you send us the video transcript?
excellent sir
i understand nothing
Please could you indicate some friendly material about bayesian inference?
It doesn’t exist. This stuff is taught horrendously everywhere
@@bigfishartwire4696 100% Agree
DURING HIGHSCHOOL DAYS, MY CLOSEST FRIENDS ARE THE NICE ONES.
the coin could land on its edge, neither heads or tails. Forgot about that potential event didn't you.
this is Advance basic concept.
There's no information about what the Y in the graph is/refers to. This is unacceptable
BAYESIAN STATISTICS IS AN EXTENSION OF THE CLASSICAL APPROACH. VARIOUS DECISION RULES ARE ESTABLISHED. THEY ALSO USE SAMPLING DATA. I LEARNED ABOUT THIS WHEN I WAS STILL IN HIGHSCHOOL IN ATENEO DE ZAMBOANGA UNIVERSITY, MY GRADES IN ALGEBRA ARE HIGH.
Haha, “I’m going give a relatively non-technical explanation…” then proceeds to speak entirely in words that have definitions specific to statistics. Most people who remember the definitions of all the words used probably also remember what Bayesian is. People who don’t remember or never did know the vocabulary used have no hope of learning here what Bayesian is.
Proving the non-existence of God was harder than I thought.
too many basic errors: "distribution closer to .5" such a claim is not even formally defined
"With a mean closer to 0.5".
And this is an explanation for who? For a bunch of statisticians? Certainly not for the 'unintroduced'! A few seconds into the 'introduction' and you are already using highly technical jargon! Perhaps you need a course in pedagogy first! 😅
This is a very bad introduction. You jumped from the absolute basics to straight up prior and posterior.
I'm really tired of these videos that area dvanced videos as "beginner videos" in disguise. They really spam all of RUclips but don't provide any value.
Please explain it more simply next time and please elaborate what each concept means that you introduce within a few seconds. Sorry for being this critical but I'm not here to learn and not to waste my time.
what tHE BLEEP did he just say?
Woo
Really bad video for a newbie trying to learn Bayesian statistics
so fucking fast..