This video provides an introduction to the technique of bootstrap resampling, which is a computational method of measuring the error in a statistic's estimator.
Thanks very much, I'm studying statistics at the moment, and asked a lecturer about bootstrapping today. This was a lot more clear and lucid than his explanation. I'm really impressed by this technique; I had figured as it is evidently powerful, it would be difficult to implement, but I can already think of the code. Thanks again
Thanks for the vid, the very last slide was quite helpful. I find it quite amazing that you can infer something about the original population by resampling from an already small sample!
Thanks for your video!! I have a question. If I want to estimate the t-statistic of one parameter in, for example, an OLS estimation with a small sample, the procedure is the same?
Not sure it's proper to shade under and below the cdf curve - I think the boundaries would be marked at the values of x where cdf(x) = 0.025 and cdf(x) = 0.975. Graphically, this is a dotted lines from 0.025 and 0.975 on Y-axis running horizontal to the function, then straight down to the corresponding X values on the X-Axis
Great Video! I'm writing an essay on this topic and have a few questions. Firstly, I've been reading about forming an empirical distribution from the sample data, would this just be done in the step when sampling with replacement from the sample data is undertaken? Also say if N=10 for instance and then you let B=10,000 then surely your bootstrap resamples will repeat? Is this allowed or is there a limit on B dependant upon N? If you could help it would be appreciated :)
please help me to solve this question Bootstrap resampling is a way to get the most out of a limited amount of data for model estimation or validation. Given a random walk model x1 = ε1, xt = xt−1 + εt, t > 1, where xt is the observation and εt is the noise at time point t, and data sample 6, 1, 3, 7, 12, 6, 19, 29, 5, −6, −9, −29, −11, −13, −5, generated from the model, describe how to generate more data from this sample using bootstrap resampling.
That is why it is named the bootstrap. Literally. It makes sense if you understand the central limit theorem. The term bootstrapping is derived from the phrase pulling oneself up by their bootstraps, or, doing a seemingly impossible thing without outside help.
2X looks fine too.. and people who are watching this should be thanking him.. he didn't made it for people who already know it.. its more for those who is just trying to gain insight.. saying slowly takes time and allows people who are listening to gain a better insight .. so please just increase the speed if you people don't like the speed of the video. bottom line : Either post constructive comments or move on. its for learning not commercial..
Thanks for the comment. Yes, you are right that you form the empirical distribution just by sampling with replacement. And in the case of N=10, B=10000, the bootstrap samples will definitely start to repeat. This just means your resampled statistics should converge once B becomes large enough. So, the error in the bootstrap resampled statistic will be limited just by the small N size.
Hmm.. but doesn't the central limit theorem apply to the mean statistic? If I take N resamples and calculate the mean I should get a normal distribution. But if I take N resamples and calculate the mode, I might not get a normal distribution, right?
The bootstrap method is to estimate a certain parameter of your data. When you generate a bunch of new samples what you're doing is calculating that parameter on each sample...so in your case you may generate 1000 bootstrap samples and calculate the mode for each of them, which would give you 1000 modes. At that point CLT comes in to play because when you have enough samples the distribution of your parameter estimates--the mode--will itself be normally distributed. When you take the mean of the 1000 modes you can treat it as normal.
Infinitely many given that revisions are ok. More realistically, as many as your computer can handle and even more realistically as many that provide the best estimates in a given amount of time and bootstrap samples.
hello great video. i will be using this technique in my study . i hope you can help me with more references and videos that i can refer to in order to understand this test more~
Hey man, you did explained the concept very clearly. However, you gotta put more energy and enthusiast. You sounded like if you don't fully understand the concept and was talking with extreme care not to say something innacurate, which is fine, except for the part that it sounded slow and kind of boring. Gotta put more energy!!!