the only good straight foward, video on bootstrapping out there. No book-canned stratified answer, as it is so often common in statistics. Thank you, this video is a piece of art.
Thanks, nice video of a very useful series. Just a doubt : at the end you say that a way to correct the biased estimation of the variance is to add a quantity to each value. But this does not change the variance ... Could you elaborate on the last part of the video about balanced bootstrap?
Great presentation. One thing that’s bothering me is that the 95% CI is constructed so that the CIs 95% of the time contain the true parameter value. As said on one slide. The next slide shows 95% of sample means not of CIs. I imagine this holds true but it is not addressed. Would be good to get confirmation.
First off, excellent vid. My question is - and I hope I state it clearly: Is balancing the bootstrap necessary? Can't it be assumed that an obvious outlier in a small data set is an anomaly, and the fact that the resampling doesn't pick it up as often means that it is "correcting" the data?
It hurts me to start with it depends, but it depends. Maybe you're thinking of outliers in a normal distribution, like the one in the video, but that's not what always happens. If you check your data and you see that the bootstrapped standard deviation is the same as the one in the original data without considering outliers (which you know are data points that were incorrectly measured FOR SURE, for example) you can think of it as correcting the data. But you could just have data where some data points are more prone to be picked up than others like height for male and female, in a dataset with more males. There is a chance you'd have even more males, which means bigger values in a higher frequency, and that would bias your dispersion metrics.
about the balancing part: we compute the bootstrap mean, then we subtract the difference between bootstrap mean and sample mean and get... sample mean. why not use sample mean from the beginning?
I believe bootstrap method is primarily used to understand the spread or confidence interval of the data. Based on my limited experience, most data when you bootstrap it, the mean will eventually converge to the sample mean. So when it doesn't, it implies that our initial sample might be inherently biased, or we probably need to repeat the bootstrapping procedures more until the result stabilize. Either case, the presenter offers us one simple way to possibly correct for the bias.
re adjusing a BS parameter to counter bias , a question arises. Why BS if you are going to end up with same adjusted parameter value as the observed value by adding back the difference between the obs sample's paraemter g variance eg say var_obs =0.15 and the bs parameter eg variance var_bs=0.1. Adding back the difference will simply adjust the bs value to the sample parameter value.
the previous slide showed an example in which the bootstrapped estimator for variance is biased. Balanced bootstrapping removes or at least decreases that bias.
the only good straight foward, video on bootstrapping out there.
No book-canned stratified answer, as it is so often common in statistics.
Thank you, this video is a piece of art.
There's a very nice video which has come out recently regarding bootstrapping which clearly explains it.
czcams.com/video/isEcgoCmlO0/video.html
I learned more in this 10 minute video than I did in my 3 hour lecture.
Matthew, this is very nice video with clear elucidation of bootstrapping. Thanks you for sharing.
You do a good job at explaining this. I never thought of plotting the sample means from 1to 10000 or more in R.
You're a hero. This video taught me more about bootstrapping than several hours of lectures.
Great and concise explanation, thank you! Just what I needed to understand what my prof. wanted me to do and why!
Well explained in a simple way. Thank you!
Thank you so much for this video.
this is so helpful, thank you !
wow you are a lifesaver
Thank you for the explanation!
Thanks, nice video of a very useful series. Just a doubt : at the end you say that a way to correct the biased estimation of the variance is to add a quantity to each value. But this does not change the variance ... Could you elaborate on the last part of the video about balanced bootstrap?
Great presentation. I thought you were going to construct 95% CI for R2.
THANK YOU!!
Please, some material about gibbs sampling? I need it so much.
Great presentation. One thing that’s bothering me is that the 95% CI is constructed so that the CIs 95% of the time contain the true parameter value. As said on one slide. The next slide shows 95% of sample means not of CIs. I imagine this holds true but it is not addressed. Would be good to get confirmation.
100,000th viewer! Thank you
if we want the resampling mean value to be greater than then how to proceed
ty pham
First off, excellent vid. My question is - and I hope I state it clearly: Is balancing the bootstrap necessary? Can't it be assumed that an obvious outlier in a small data set is an anomaly, and the fact that the resampling doesn't pick it up as often means that it is "correcting" the data?
It hurts me to start with it depends, but it depends. Maybe you're thinking of outliers in a normal distribution, like the one in the video, but that's not what always happens. If you check your data and you see that the bootstrapped standard deviation is the same as the one in the original data without considering outliers (which you know are data points that were incorrectly measured FOR SURE, for example) you can think of it as correcting the data. But you could just have data where some data points are more prone to be picked up than others like height for male and female, in a dataset with more males. There is a chance you'd have even more males, which means bigger values in a higher frequency, and that would bias your dispersion metrics.
about the balancing part: we compute the bootstrap mean, then we subtract the difference between bootstrap mean and sample mean and get... sample mean. why not use sample mean from the beginning?
I believe bootstrap method is primarily used to understand the spread or confidence interval of the data. Based on my limited experience, most data when you bootstrap it, the mean will eventually converge to the sample mean. So when it doesn't, it implies that our initial sample might be inherently biased, or we probably need to repeat the bootstrapping procedures more until the result stabilize. Either case, the presenter offers us one simple way to possibly correct for the bias.
how to do bootstrapping with gretl please?
what does resampling the data with replacement means??
re adjusing a BS parameter to counter bias , a question arises. Why BS if you are going to end up with same adjusted parameter value as the observed value by adding back the difference between the obs sample's paraemter g variance eg say var_obs =0.15 and the bs parameter eg variance var_bs=0.1. Adding back the difference will simply adjust the bs value to the sample parameter value.
the added value is the sample parameter value (i.e. var_obs) + MEAN of var_bs. Mean of var_bs is not equal to var_bs
6:57 I think R² has a standard formula for 95% CI
Here you can play with the topic more visual
seeing-theory.brown.edu/frequentist-inference/es.html#section3
كيف اترجم الفديو للعربية؟
You never added why you would want to do balanced bootstrapping. It is to get better performance statistics.
the previous slide showed an example in which the bootstrapped estimator for variance is biased. Balanced bootstrapping removes or at least decreases that bias.
honoured to be the 1000 one click like
Sound quality is bad!!