The intuition behind quantile regression
Музыка
This video provides an intuitive idea of the potentially complicated topic of quantile regression.
The video starts by discussing the case of OLS (Ordinary Least Squares) regression as the conditional mean, before discussing the cases of the conditional median (i.e. median regression) and conditional quantiles (quantile regression). The video explains why quantile regression cannot be estimated at the 0% and 100% quantiles, and explains why quantile regression is more robust to outliers on the dependent variable than OLS.
The video shows figures and graphs as well as some code in R software, although viewers can follow the video without any knowledge of R software.
The R code and slides that accompany this video are freely available from my github page:
github.com/alexcoad/Econometrics
Пікірлер: 21
This is unquestionably the best explanation I have seen on the intuition of Quantile regression. Concise, well organised and superbly presented. Bravo!
Thanks so much! Insightful and right to the point.
extremely clear explanation, no messy lengthy equations, just concise and simple explanation.
This is probably the best explaination on youtube
The most clrar explanation about Quantile Regression, Really looking forward to see more video!
Great!!
Extraordinary presentation. Could you provide a more technical explanation of QR with respect to the location and scale functions ?
Thank you
thank you so much for this video! this is the best one i've seen so far and it has truly helped me comprehend quantile regressions
what a nice explanation! well organized, and extremely clear!
Many thanks to you sir for this helpful video Very clear explanation Please, keep going.
Thanks so much for the wonderful explanation
Great video Dr. Coad.
Well explained ❤️ highly appreciated 👍🏻😊
Thank you, Alex!
Awesome mesmerising
Excellent video, thank you. Just one tiny quibble... at about 7:00 it is stated that with median regression, 50% of observations must be above the line of best fit and 50% below, but that's not quite right is it? E.g. the example graph shows 3 observations above the fitted line and 8 below.
@alexcoad1
3 ай бұрын
Dear Gerygone, thanks, it is true that with median regression (similar to the case of calculating a median) 50% of observations should be above the line of best fit and 50% below. In the video at about 7:00, the graph is not drawn with 50% of observations above/below the line of best fit, as you pointed out, therefore at about 7:00 the graph's best-fit line does not accurately correspond to the case of median regression.
@michaelbedward
3 ай бұрын
Many thanks for your reply. That's cleared up some confusion I had about the influence large outliers could have when minimizing absolute residuals.
Hi, thank you so much for this explanation. I just wanted to know if we need to check for stationarity before doing a quantile regression. If yes, if the variables have unit root at level, do we take the first difference and then perform the quantile regression? Thank you for your consideration, hoping to seek some guidance.
@alexcoad1
3 ай бұрын
Thanks Eshanair, stationarity is a time-series concept, while quantile regression was initially designed for cross-sectional data. Quantile regression can also be applied to panel data (the case of "panel quantile regression"), and time series data, if you are willing to make some extra assumptions. With panel data, if there is a unit root, then taking the first difference could indeed help.