Been following your videos for a while and really like your explanations. Some video ideas: 1.) Data Transformations predictors/responses 2.) When is an outlier an outlier? 3.) Distributions (Playlist; Gamma, Beta, Multivariate Norm, etc) 4.) Models (Playlist; SLR, ANOVA, TS, etc) 5.) Dubious Statistics (Playlist; maybe talk about common pitfalls, notorious misuses, even review of popular papers) Anyways, keep up the good work-you have a great style
Thanks! Actually, I was thinking of ANOVA as the next explainer video, so your suggestion is perfect! I’ve also logged the ideas you’ve suggested too, I think they’d be great topics for the channel
Keep up the good work man ! As an econ major who takes a lot of stats and econometrics classes your videos made me better understand some basic concepts I was taught in class and also introduced me to some pretty neat concepts as well. I think im gonna pursue a stats/data related masters after I get my degree. Thanks !
Yeah I can see that. Being paired has a specific definition for hypothesis tests, the “groups” are just two observations from the same people at different times.
Love the content, your way of explaining these concepts is amazing (at least to me). How do you determine if you have a lot of outliers or not enough data?
Thanks! Unfortunately there’s no hard rule on judging those two questions, it really does depend on what context you’re working in. There are actually statistical tests for detecting outliers though, but not a lot of people are aware of them
@very-normal In a lot areas, outliers are rarely talked or conveniently glossed over. Like you mentioned there are a bunch of different methods/test to identify outliers, but they aren’t common knowledge. Since there aren’t any hard and fast rules, it would be interesting to hear your thoughts on best practices and intuitions. For example, when I was first exposed to research - for the lab I was working in the “right way” to remove outliers was to run whatever model, plot the standardized residuals and cooks D, remove anything outside of +/- 2 and anything greater than 4/n…. Not necessarily the best approach 😅
Admittedly I’m not sure if there are Bayesian analogues to the nonparametric or rank tests off the top of my head. I’m sure there are, but it’s definitely not a part of any stat curriculum I’ve seen
There’s another argument in the wilcox.test function that lets you specify if you want a confidence interval to be made (conf.int, by default is False). That should give you what you need in the resulting output
i would love to see how the alpha error and the power of the test were calculated using MW and t-test when data are normal and non-normal. 🙂 but nevertheless, coming from an Applied Stat graduate, i love ur content. No hardcore math. But the foundations in a sense are not neglected.❤
9:57 this is a good example, but what happens when the effect size is minimized? These stats, and the difference between them, impacts research more the smaller the effect size since all stat tests should be able to detect big differencesa anyways I spent all day in R and Im done with my workday lol
Been following your videos for a while and really like your explanations.
Some video ideas:
1.) Data Transformations predictors/responses
2.) When is an outlier an outlier?
3.) Distributions (Playlist; Gamma, Beta, Multivariate Norm, etc)
4.) Models (Playlist; SLR, ANOVA, TS, etc)
5.) Dubious Statistics (Playlist; maybe talk about common pitfalls, notorious misuses, even review of popular papers)
Anyways, keep up the good work-you have a great style
Thanks! Actually, I was thinking of ANOVA as the next explainer video, so your suggestion is perfect! I’ve also logged the ideas you’ve suggested too, I think they’d be great topics for the channel
Keep up the good work man ! As an econ major who takes a lot of stats and econometrics classes your videos made me better understand some basic concepts I was taught in class and also introduced me to some pretty neat concepts as well. I think im gonna pursue a stats/data related masters after I get my degree. Thanks !
This is by far the best video on the topic for curious minds. Keep it up❤
Ooh yes I have been waiting for this video since I watched part 1!
Thanks for the clear explanations and practical angle as always!
This channel is gold
Just graduated Uni, and now have time to binge very normal along with some stats textbooks
Congrats on graduation!
Thank you so much. Would love to see a video on the different distribution models.
Specifying that they're unpaired by setting unpaired to false is mad confusing
Yeah I can see that. Being paired has a specific definition for hypothesis tests, the “groups” are just two observations from the same people at different times.
Based on the help docs for `wilcox.test`, I think that should have said "paired=F".
(Overall, I love the channel and your videos!)
Excellent video! Please next explain how goodness of fit tests are non parametric test
Love the content, your way of explaining these concepts is amazing (at least to me). How do you determine if you have a lot of outliers or not enough data?
Thanks! Unfortunately there’s no hard rule on judging those two questions, it really does depend on what context you’re working in. There are actually statistical tests for detecting outliers though, but not a lot of people are aware of them
@very-normal
In a lot areas, outliers are rarely talked or conveniently glossed over. Like you mentioned there are a bunch of different methods/test to identify outliers, but they aren’t common knowledge.
Since there aren’t any hard and fast rules, it would be interesting to hear your thoughts on best practices and intuitions.
For example, when I was first exposed to research - for the lab I was working in the “right way” to remove outliers was to run whatever model, plot the standardized residuals and cooks D, remove anything outside of +/- 2 and anything greater than 4/n….
Not necessarily the best approach 😅
Beautiful video. Thanks for making it :)
Can't help wonder if there's a Bayesian alternative for this two-sample test ...
Admittedly I’m not sure if there are Bayesian analogues to the nonparametric or rank tests off the top of my head. I’m sure there are, but it’s definitely not a part of any stat curriculum I’ve seen
@@very-normal formal curriculum often overlook Beyesian methods indeed
How should you calculate confidence intervals, if for instance you want to know an upper and lower bound for your location shift?
There’s another argument in the wilcox.test function that lets you specify if you want a confidence interval to be made (conf.int, by default is False). That should give you what you need in the resulting output
i would love to see how the alpha error and the power of the test were calculated using MW and t-test when data are normal and non-normal. 🙂
but nevertheless, coming from an Applied Stat graduate, i love ur content. No hardcore math. But the foundations in a sense are not neglected.❤
9:57 this is a good example, but what happens when the effect size is minimized? These stats, and the difference between them, impacts research more the smaller the effect size since all stat tests should be able to detect big differencesa anyways
I spent all day in R and Im done with my workday lol
Its preferable to just go with 10 employees and measure the before-after difference. That way, you eliminate employee-level heterogeneity.
This series is a bit heavy for me, Any other sources with detailed explanations would be appreciated.
I’ve always found the material at Penn State to be good. If you look up “penn state nonparametric” on Google then it should come up
normal?
very
First?
yesssir welcome