Simple Linear Regression - Discussion of the Normality Assumption

[@elias043011]

You have done in 12 minutes what my master's level professor couldn't do in a month, thank you!

[@romgossel7971]

Amazing job! Great video!

[@elviswanasunia878]

Well explained. Finally I have understood the importance of the normality assumption. Thanks.

[@paria4587]

Your explanation helped me a lot!

[@onurvarol1040]

Wondeful. You are the one.

[@alichehrazi3518]

Great Explanation! Most of the videos go over the assumption but don't explain the importance of them!

[@SzTz100]

Thank you, at last these error iid's are explained properly.

[@YukiYan-go6ov]

Thanks for the video. In practice, how do we determine the variance?

[@flauschigeswoelkchen]

Didn‘t watch this video yet but a lot of the other ones. As I am studying statistics I am so very thankful for your videos. They really do help a lot. Unfortunately I can’t find videos for asymptotic (convergence and such) but I do recommend your videos to all my classmates. Again, thank you!

[@gauranshagrawal4648]

After finding Expected value and variance of yi (with the assumption of normality of errors) , on what basis did you conclude that yi is also normally distributed ?

[@ChiomaIbe-z4r]

Thanks a lot Now i understand better

[@rajarshibhattacharya5192]

very helpful. Thank you!

[@jesskl6907]

Great job! Thank you :))

[@sukursukur3617]

Why normality tests? Why dont we implement poissonity test? What makes normal distribution privileged among other distributions?

[@Stats4Everyone]

The normal distribution is symmetric. The poison distribution is skewed. If we think about the purpose of regression, we aim to explain the variability in Y. If we end up having skewed error terms then we did not do a good job explaining the variability in Y. The error terms ideally are just randomly scattered above and below zero.... Theoretically, constructing a model with poison distributed error terms may be interesting.... we would lose the F tests that we run later on to determine if the model is significant.. also, we would lose the t and z tests that we use to determine if particular predictors are useful in predicting Y... Though there may be other ways you can run these tests.. some more math would be needed to look into that.