Yup. The only I would like is confidence intervals and the ability to do multiple overlay data plots. Maybe the developers will add this in the future. :-)
I believe the binned scatterplot you show is more akin to k-nearest neighbor regression, since each bin has a variable width and contains a fixed number of observations. For kernel regression the bins should have a fixed width with a variable number of observations.
Good point about nearest KNN, I had not thought of it like that. I guess the KNN encourages you to think of the estimator as a function you can evaluate anywhere you want to, whereas the idea behind the binned scatterplot is that you only ever evaluate it at the centroids of the bins. In that sense, you could think of the binscatter as a special case of the KNN.
I'm not familiar with the filter but from a quick wiki look they look similar, but with a kernel that puts zero weights on data points further away than some distance. There's some relation to N-W with an Epanechnikov (triangle) kernel.
Is it possible to use Kernel regression to the multivariate framework jointly? From what I can see in R (rather than stata) it only presents the bivariate case.
Hi Ravi, I'm a bit overburdened right now but maybe I can do the spline video in the future. In the meantime, I recommend looking in Cameron & Trivedi's book. I don't think I understand what you mean by non-linear data? The non-parametric regression is appropriate when the true "regression function" (or, conditional mean function) is non-linear.
Watching this in 2022......brilliant explanation!
Awesome Possum! I love the tip about binscatter in Stata. I was using twoway scatter with lpoly. This appears much better.
Great to hear that! Once you go binscatter, it's kind of hard to go back... such a powerful yet simple tool (means in bins - what's not to love?)
Yup. The only I would like is confidence intervals and the ability to do multiple overlay data plots. Maybe the developers will add this in the future. :-)
@@dmaslach I agree completely! If it becomes more widely used, eventually they probably will...
Excellent video sir, help me understand this topic easily
Beauty of simplicity!! Very well Explained. Thanks.
I believe the binned scatterplot you show is more akin to k-nearest neighbor regression, since each bin has a variable width and contains a fixed number of observations. For kernel regression the bins should have a fixed width with a variable number of observations.
Good point about nearest KNN, I had not thought of it like that. I guess the KNN encourages you to think of the estimator as a function you can evaluate anywhere you want to, whereas the idea behind the binned scatterplot is that you only ever evaluate it at the centroids of the bins. In that sense, you could think of the binscatter as a special case of the KNN.
Can you please publish the data you used? I like to try it out. Thanks for excellent explanation.
This is a 2016 video powered by 1440p it's great!
I have a smooth brain. What are some good keywords for me to get started learning in a direction to learn this?
This is very similar to savitzky golay filter, no?
I'm not familiar with the filter but from a quick wiki look they look similar, but with a kernel that puts zero weights on data points further away than some distance. There's some relation to N-W with an Epanechnikov (triangle) kernel.
excellent video
Hey. I knew this guy before he got famous!
famous and arrogant
Great content.
Is it possible to use Kernel regression to the multivariate framework jointly? From what I can see in R (rather than stata) it only presents the bivariate case.
Thank you so much Mr Anders. Your explanation for binned scatterplot really useful for me. Could you make availability of your Matlab code?
Shoot me an email. Some of the code will be used in a course, so I'd prefer to not put it online as is.
Hi Mr Anders, I really like your teaching. May I have your email for the Matlab/R code?
If you do a quick google, you should be able to find my homepage where my email is (sorry, I've gotten a lot of spam so I'm a little paranoid :)
Is there a unique slope in kernel regression?
Thank you Sir Anders! I wonder if you have an available Rcode for NAdaraya Watson Estimator.
Unfortunately not, just in Matlab for this video.
Thank you!
Hi Anders, Can we fit non-parametric regression for a non-linear data? And also can you please come up with videos on use of splines?
Hi Ravi, I'm a bit overburdened right now but maybe I can do the spline video in the future. In the meantime, I recommend looking in Cameron & Trivedi's book.
I don't think I understand what you mean by non-linear data? The non-parametric regression is appropriate when the true "regression function" (or, conditional mean function) is non-linear.
Thank you so much
Thank you so much Mr Anders, could you help me to proof Kernel is density's function in mathemetic method?
Rahmat Hidayat, i dont understand your question, I’m afraid.
ECE 449 UofA
Tak !
pika pika
why are you typing so bad. I got headache.