Multicollinearity (in Regression Analysis)

[@jackg8217]

Thank you for taking the time to do these videos, they have been a huge help with my studies

[@perfectmoments3876]

Hi, how do you come up with the critical value of 0.1 respectively 10? Is there a source for that?

[@MichelleCox-l4w]

This is a great video, thank you!

[@Dataverse3211]

Thank you for this wonderful lecture 👍🤝

[@datatab]

Thanks for your nice Feedback!

[@paparokauli]

God bless you woman!

[@jasbirmanhas3355]

Can you make video on coefficient of determination

[@datatab]

Thanks for your message! Yes sure! I can try!

[@kennyjohndelacruz3435]

Can you make a video about two tailed test for next pls.

[@datatab]

two tailed t-Test? or in general two tailed?

[@juanwang3705]

hi, this video is really helpful. You mentioned in the next video, you will tell how to test the multicollinearity of dummy variables. but I can't find that video. could you send me a link?

[@datatab]

Oh sorry, we have just a video on dummy variables! But for dummy variables it is the same, so you test it in the same way. kzbin.info/www/bejne/mJ_Nga1-hpVnd8U

[@juanwang3705]

@@datatab ok! Thanks.

[@RichardGoza]

I am under the Linear regression tab and i do not see the subtab check condition. All i see under Linear Regression is Test assumptions, Effect size and Summary in words. Please help!!! I have a subscription on your website.

[@allexclusive007]

The same problem here. Why doesn't she clarify it?

[@konstantinosbanos]

Independent and correlated? How is this possible?

[@wayneisk320]

The variables are independent measurements. Like weight and height or weight and body fat percentage. But these values correlate to a degree with each other. As correlation increases the value of these variables become more interchangeable.

[@konstantinosbanos]

@@wayneisk320 It’s not possible two variables be independent and correlated simultaneously. At the example you mentioned, the variables are not independent.

[@wayneisk320]

@@konstantinosbanos Technically, yes. In practice you do not know the level of independence until you measure them. Commonly all variables for regression are considered "independent" until testing. However, the partial correlations can have new information and will still be used as independent variables. Particularly in large datasets with 1 million+ samples you will find most features may have weak correlations. Excellent example of this is gene SNIPs and medication interactions. There will always be correlations when working with this type of data caused by external co-related factors that are not measured. Independence is accepted as part of feature selection.

[@wayneisk320]

@@konstantinosbanos There is a famous quote from a statistician that "no variable is independent." This was in reference that with enough features all variables will have mutual information. Feature selection and feature abstraction/crafting create independence. But only after statistical analysis. Also mutual information can, with some data, be "removed" in preprocessing. Not technically "independent", but independent enough.

[@jasbirmanhas3355]

Kindly send video on correlation

[@datatab]

Hello Jasbir, just search at youtube "correlation DATAtab" we have greate Videos about correlation!

[@hasanatayoub6512]

Can you teach multicolineality by using this series😢 .what is nature of mulicoliniality .is multicoliniality really problem? .what are its practical consequence? .how do one detectit? .what remedial measure?

[@mikefieselman]

The nature of multcolinarity is the similarity on impact one independant variable has with another independant variable. it is really a problem because we're looking for a model that best fits, and why have a model that shows 4 independant variables's influence on a dependant variable, when 3 gives same result. Ask yourself is 3 better than 4? yes it is Practical consequence is that you are not recieving the best model to explain the impact on variables to the independant variable. you detect it with VIF formula which is (1-(1-R^2) where r^2 is correlation squared or you can get R^2 through running a regression analysis on excel remedial measure... you remove one of the dependant variables that shows multicollinarity either from a correlation matrix chart or VIF diagnosis, or looking at the P-value regardless of multicollinarity. You're welcome :)

[@mikefieselman]

You can also detect multicollinarit by performing regression model with only the dependant variables, where you have 1 of the dependant variables as y and the rest as x.

[@mikefieselman]

this is called auxillary regression

[@MonsieurSchue]

So apparently the service is no longer free.:(

[@wanjadouglas3058]

Good job

[@datatab]

Thanks : )

[@scottmacnevin3555]

Superb, thank you

[@datatab]

Many thanks for your Feedback! Regards Hannah

[@mahammadodj]

Thank you so much!

[@datatab]

Glad it helped!

[@XavierGarciaMartínez-h7q]

Thanks!!!

[@Romeo-sf7tw]

Genius!

[@datatab]

Many thanks! Regards Hannah