I have been watching videos on this channel for months and I refer back to the Effect frequently and I just now realized that you are the same person when you introduced yourself at the beginning of the video 🤯
Thanks for this insightful comparison. Really helpful. I have been struggling on where to focus on, especially because I am mostly interested in doing causal econometrics and impact evaluation in the coming years. I decided to spend more time and improve on Stata and R only. Maybe in the future I may look at Python and Julia, but for now I think Stata and R would serve my econometric needs. And this video has reinforced my decision. Looking forward to more insightful videos on best practices in 21st century econometrics😁
Hey Nick, I recently graduated from my economics major and your videos did a tremendous job in helping me nail my econometrics and causal inference classes in the last year of my studies. Now I'm a data scientist doing demand forecasting, so I lean towards Python for versatility and scalability. Though I have been guilty of using Eviews to validate statistical test results. Thanks again for your work on TH-cam!
It's basically just R and Python in my view. But obviously it completely depends on what you're doing. Academia most likely R, anything else more likely Python and in-house solutions
Dear the most genius sir :) , can you help me with two things please. I am analyzing panel data (twoways within fixed effect). I have around 900 observations. First, my residuals are not normally distributed. I read in many websites that in panel data the assumption of normality of residuals is ignored, however, I could not find any scientific resource to cite this information in my paper. Is it true that the normality assumption is ignored in panel data and my paper would not be rejected due to non-normality and is there any resource to cite from to support the ignorance of residuals normality assumption? I have been trying for months and changing the specification of the model either for the sample, time period, and the variables with no result. Second, Do I have to run the Ramsey's RESET test for (twoways within fixed effect model)? If yes, is this test appropriate or there is a specific one for the mentioned mode?
Normality of the error terms isn't very important. Normality is only important for ensuring that the sampling distribution is also normal (so you get correct standard errors) but even in that case, unless your errors are truly wild, like power-law distributed or something, your sampling distribution is pretty close to normal anyway and your standard errors will only be barely off. If you search Google scholar for "violations of normality assumptions regression" you'll find a whole buncha papers to this effect, for example Hubbard 1980 or Schmidt & Fenan 2018. As for RESET, run it if you suspect your model is misspecified. In general you don't want to run a robustness test unless you have reason to believe the assumption it's testing will fail. See my article on robustness tests nickchk.substack.com/p/robustness-tests-what-why-and-how
While I prefer R's plotting, 1-indexing + python's vectorization and modelling still feels tagged on, I cannot say those have been worth coming back (for 'real' work) for me.
I've written a few packages using Rcall and found it to be pretty fiddly. Usually if I'm working across languages in the same project I'll just run separate scripts for separate languages, saving objects to file instead to pass them back and forth. But I pretty much only do dual language when it's data cleaning in language A -> export -> everything else in B anyway so that's a workable pipeline. I think I'd only find interlinking functions useful if I were, say, running actual models in both languages, and that's not really something I do. That said, maybe the official Stata -> Python interlinkage works better than Rcall. I suspect the R Python pipeline is better too.
I know they use it in computational finance sometimes. Wouldn't you have to program everything yourself from scratch though for most econometric applications? If you're willing to do that then c# would be fine but also so would any general purpose programming language.
Hi Dr.Nick I graduated from my mathematical economics degree from UWaterloo, are going to purse a master in machine learning. If I still want to do a PHD in economics will I still need to do a coursework master degree? btw, I had taken Advanced Micro, and Advanced macro economics plus 11 statics courses.. Any help is appreciated
I don't know about canada, but in the US it's not expected that you get a masters before doing a phd. Going straight from undergrad is probably more common, especially if you already have the necessary math background which it sounds like you do.
@@NickHuntingtonKlein Aw thanks for your quick reply, also if you have any projects in econometrics pls let me know, I more than willing to gain an ra experience : )
Totally true on Python. It clearly is not made for stats! I prefer R/Rstudio. They were made for stats. Sorry, but Stata is dated--way dated. It is also incredibly expensive and, thus, anti-poor people here and in developing countries!
![[LIVE] : ONE ลุมพินี 89 | คู่เอก "ยอดไอคิว vs คิริลล์"](http://i.ytimg.com/vi/Z3vYkHJ_5Ig/mqdefault.jpg)
One note: Python is actually by far most common in the private sector (and actually Stata for econ consulting), but other than that good points!
I have been watching videos on this channel for months and I refer back to the Effect frequently and I just now realized that you are the same person when you introduced yourself at the beginning of the video 🤯
I learned most of these languages and in the following order:
Stata - Python - R - Julia
And by far R and Julia are the best! (At least for me)
Thanks for this insightful comparison. Really helpful. I have been struggling on where to focus on, especially because I am mostly interested in doing causal econometrics and impact evaluation in the coming years. I decided to spend more time and improve on Stata and R only. Maybe in the future I may look at Python and Julia, but for now I think Stata and R would serve my econometric needs. And this video has reinforced my decision. Looking forward to more insightful videos on best practices in 21st century econometrics😁
Well done, thank you!
Hey Nick, I recently graduated from my economics major and your videos did a tremendous job in helping me nail my econometrics and causal inference classes in the last year of my studies. Now I'm a data scientist doing demand forecasting, so I lean towards Python for versatility and scalability. Though I have been guilty of using Eviews to validate statistical test results.
Thanks again for your work on TH-cam!
I myself have a bias skew towards R. It is very versatile and was used to teach my econ program
It's basically just R and Python in my view. But obviously it completely depends on what you're doing. Academia most likely R, anything else more likely Python and in-house solutions
*Stata is EXPENSIVE (you didn't mention)
*MATLAB is EXPENSIVE (you mentioned)
*R, Python, Julia, are FREE
Excel....nevermind 🙂
I did mention that Stata is expensive when I mentioned that Matlab was
@@NickHuntingtonKlein true , true
Dear the most genius sir :) , can you help me with two things please. I am analyzing panel data (twoways within fixed effect). I have around 900 observations.
First, my residuals are not normally distributed. I read in many websites that in panel data the assumption of normality of residuals is ignored, however, I could not find any scientific resource to cite this information in my paper. Is it true that the normality assumption is ignored in panel data and my paper would not be rejected due to non-normality and is there any resource to cite from to support the ignorance of residuals normality assumption?
I have been trying for months and changing the specification of the model either for the sample, time period, and the variables with no result.
Second, Do I have to run the Ramsey's RESET test for (twoways within fixed effect model)? If yes, is this test appropriate or there is a specific one for the mentioned mode?
Normality of the error terms isn't very important. Normality is only important for ensuring that the sampling distribution is also normal (so you get correct standard errors) but even in that case, unless your errors are truly wild, like power-law distributed or something, your sampling distribution is pretty close to normal anyway and your standard errors will only be barely off. If you search Google scholar for "violations of normality assumptions regression" you'll find a whole buncha papers to this effect, for example Hubbard 1980 or Schmidt & Fenan 2018.
As for RESET, run it if you suspect your model is misspecified. In general you don't want to run a robustness test unless you have reason to believe the assumption it's testing will fail. See my article on robustness tests nickchk.substack.com/p/robustness-tests-what-why-and-how
While I prefer R's plotting, 1-indexing + python's vectorization and modelling still feels tagged on, I cannot say those have been worth coming back (for 'real' work) for me.
Have you got any advice on interlinking languages? E.g. RStata or Rcall for interlinking Stata & R?
I've written a few packages using Rcall and found it to be pretty fiddly. Usually if I'm working across languages in the same project I'll just run separate scripts for separate languages, saving objects to file instead to pass them back and forth. But I pretty much only do dual language when it's data cleaning in language A -> export -> everything else in B anyway so that's a workable pipeline. I think I'd only find interlinking functions useful if I were, say, running actual models in both languages, and that's not really something I do. That said, maybe the official Stata -> Python interlinkage works better than Rcall. I suspect the R Python pipeline is better too.
You are simply the bets
What do you think about c#?
I know they use it in computational finance sometimes. Wouldn't you have to program everything yourself from scratch though for most econometric applications? If you're willing to do that then c# would be fine but also so would any general purpose programming language.
Hi Dr.Nick I graduated from my mathematical economics degree from UWaterloo, are going to purse a master in machine learning. If I still want to do a PHD in economics will I still need to do a coursework master degree? btw, I had taken Advanced Micro, and Advanced macro economics plus 11 statics courses.. Any help is appreciated
I don't know about canada, but in the US it's not expected that you get a masters before doing a phd. Going straight from undergrad is probably more common, especially if you already have the necessary math background which it sounds like you do.
@@NickHuntingtonKlein Aw thanks for your quick reply, also if you have any projects in econometrics pls let me know, I more than willing to gain an ra experience : )
Ken - they didn't mention SHAZAM, but they do mention Excel !
R will always be the best, at least for me 😂
Totally true on Python. It clearly is not made for stats! I prefer R/Rstudio. They were made for stats. Sorry, but Stata is dated--way dated. It is also incredibly expensive and, thus, anti-poor people here and in developing countries!