Might have been a good idea to actually explain how anova and regression can be seen as 2 sides of the same coin, i. e. how they can be applied to the same data set.
Oops wasn’t finished. We used those for RBD - block design. I’m still confused as they seem related. Regression looks at if x is correlated to y. Anova asksxare all the means equal or not. Where is the relationship between them? When do you start with a regression and end up using an anova table? Thank you.
But nominal regression models also exist. Logistic regression is another example where the outcome is categorical as well. Is there a difference between these and ANOVA?
I'm not a fan of this explanation. Framing it in terms of how they are different I think undermines the truth of how one is a special case of the other. You can answer the ANOVA question with regression because x doesn't have to be numerical. And I would never say regression models cause and effect, only correlation. If it modeled cause and effect, you would be able to make accurate predictions over infinite time. With the advent of cheap/free computing, the way we teach statistics is very outdated. I would never do a t-test these days to check a null hypothesis of no difference in the means. I would always run a permutation. It's like 2 more lines of code. Similarly, I think it's more helpful to just teach the regression method of ANOVA. Just my 2 cents. Take it or leave it.
This video and the source article in the book do not say that Regression can prove cause and effect. The script for the second-to-last slide (#18) says, "The purpose of Regression is to provide a Cause and Effect Model ... Subsequent to that, data can be collected in Designed Experiments to prove or disprove the validity of the Model." There is a 3-part series of articles in the book on Design of Experiments, which show how a Regression model can be validated in Designed Experiments to show cause and effect (or show the model cannot be used to show cause and effect). There will be 3 videos on Design of Experiments at some time in the future.
Hi. I am curious as I’m learning about this right now. We are learning simple linear regression, multiple linear regression and randomized controlled blocks, and we are using anova. I’m kind of confused. I understand that testing B1=0 vs not equal 0 tells us if there is a correlation between x and y. If there isn’t we stop and if there is we find out what it is. For multiple we do the same but then use anova to find out which slopes are different? I’m confused that regression and anova are not related as I’m certain we are finishing regression questions with anova tables? Wondering where my confusion is. We’ve only used SSM (model), SSE (error) and SST so far. So maybe this is more advanced? We have also used SSG (groups) and SSTrt (treatment) for the
Very apt, there can be no more explanation than this. Am happy I came across this
This was so helpful. I always wondered what the heck ANOVA has to do with regression
Might have been a good idea to actually explain how anova and regression can be seen as 2 sides of the same coin, i. e. how they can be applied to the same data set.
Great video
Thanks much.
Oops wasn’t finished. We used those for RBD - block design.
I’m still confused as they seem related.
Regression looks at if x is correlated to y. Anova asksxare all the means equal or not. Where is the relationship between them? When do you start with a regression and end up using an anova table? Thank you.
Excellent video! Thank you!
You're welcome!
But nominal regression models also exist. Logistic regression is another example where the outcome is categorical as well. Is there a difference between these and ANOVA?
Great video! like it!!
really helpful video, thanks a lot!
Thank you
I'm not a fan of this explanation. Framing it in terms of how they are different I think undermines the truth of how one is a special case of the other. You can answer the ANOVA question with regression because x doesn't have to be numerical. And I would never say regression models cause and effect, only correlation. If it modeled cause and effect, you would be able to make accurate predictions over infinite time.
With the advent of cheap/free computing, the way we teach statistics is very outdated. I would never do a t-test these days to check a null hypothesis of no difference in the means. I would always run a permutation. It's like 2 more lines of code. Similarly, I think it's more helpful to just teach the regression method of ANOVA.
Just my 2 cents. Take it or leave it.
Hi. Why wouldn't you run a t-test and run a permutation test instead?
This video and the source article in the book do not say that Regression can prove cause and effect. The script for the second-to-last slide (#18) says,
"The purpose of Regression is to provide a Cause and Effect Model ... Subsequent to that, data can be collected in Designed Experiments to prove or disprove the validity of the Model." There is a 3-part series of articles in the book on Design of Experiments, which show how a Regression model can be validated in Designed Experiments to show cause and effect (or show the model cannot be used to show cause and effect). There will be 3 videos on Design of Experiments at some time in the future.
Hi. I am curious as I’m learning about this right now. We are learning simple linear regression, multiple linear regression and randomized controlled blocks, and we are using anova. I’m kind of confused. I understand that testing B1=0 vs not equal 0 tells us if there is a correlation between x and y. If there isn’t we stop and if there is we find out what it is. For multiple we do the same but then use anova to find out which slopes are different? I’m confused that regression and anova are not related as I’m certain we are finishing regression questions with anova tables? Wondering where my confusion is. We’ve only used SSM (model), SSE (error) and SST so far. So maybe this is more advanced? We have also used SSG (groups) and SSTrt (treatment) for the
1:20
1:00