@@sighsha3657 At that point we are testing how well we can predict "size" with and without specific variables in the model. So we see how well we can predict "size" with and without "weight" and we see how well we can predict "size" with and without "tail length". These tests help us asses how useful it is to use "weight" or "tail length" to predict "size". A small p-value suggests that a variable is useful.
Yeah, I'm struggling with this bit too. It feels like the coefficient line for 'weight' should be comparing the full model against a model with just weight as a predictor, but the explanation suggests the full model is being compared to a model with only tail length as a predictor.
I'm sorry if I'm asking a stupid question, why p-value of weight can tell us using weight and tail isn't significantly better than only tail. Thank you so much
First you need to understand linear regression: th-cam.com/video/nk2CQITm_eo/w-d-xo.html and then you can find the answer to your question in this video that describes the theory of multiple regression: th-cam.com/video/zITIFTsivN8/w-d-xo.html
hello, very helpful video. Anyway I have question though - why inference (06:35) 'using weight and tail isnt significanlty better than using tail alone' is derived from line "weight" and not from "tail" line below? Does the line "weight" of regression output compares exactly 'if using weight and tail is better than using tail alone to predict size'? shouldnt it be 'if predictor WEIGHT alone is better at prediction of size than MODEL WITH BOTH WEIGHT AND TAIL' instead or im wrong? I had course about multiple regression at academics couple years ago and trying to remind everything, I have found your video , but still I actually wanted to know that. Best wishes edit: i remember that in stepwise regression we somehow exclude predictors which do not make a significant contribution and are therefore not statistically significant, so reading/watching about stepwise regression may do the thing for me?
For each variable that we test "weight" and "tail", we test the "full model" vs the model without that specific variable. So, for testing "weight" the full model is "weight + tail" and the model without that variable is just "tail". For testing "tail", the full model is "weight + tail" and the model with that that variable is just "weight." You can learn more details here: th-cam.com/video/zITIFTsivN8/w-d-xo.html
i like that you sing; me writer director producer who is also very statistically literate for my work in psychology with daughter who is statistics major
@@statquest hey josh thanks for responding, I have done a linear regression model that describes the data best by using test based and criterion based model selection. I have been asked to "conduct full inference using the best fit model". I am slightly confused as to what needs to be done for this step, is it just the explanation of f-statistics and hypothesis testing obtained from summery of the model?
Multidimensional Scaling is pretty different. To learn about it, first learn about PCA (only 5 minutes long: th-cam.com/video/HMOI_lkzW08/w-d-xo.html ) and then MDS: th-cam.com/video/GEn-_dAyYME/w-d-xo.html
A small p-value would cause us to reject the hypothesis that random noise generated the data. For details about p-values, see: th-cam.com/video/vemZtEM63GY/w-d-xo.html
Sorry if this is a stupid question, but is there a good way to format the results table? For instance, if I wanted the beta to be rounded to 3 digits, and the t and p values to be rounded to 4 digits?
Good question! The only way I can think of doing it is drawing it yourself using the original values (in this case, they are stored in the variable "multiple.regression") and running them through the round() function.
@@LuisSantiago-xo4fm When the relationship is non-linear, you can try regression trees: th-cam.com/video/_L39rN6gz7Y/w-d-xo.html and th-cam.com/video/g9c66TUylZ4/w-d-xo.html
It depends on what you mean by tests. However, usually multiple regression fits the model and then tests each variable as described. So this would be regression first, tests second.
@@statquest um i thought i should perform the multicollinearity and heteroscedasticity and stationarity and do any correction before proceeding to fitting data?!
Support StatQuest by buying my book The StatQuest Illustrated Guide to Machine Learning or a Study Guide or Merch!!! statquest.org/statquest-store/
I'm giving this a thumbs up...just on the intro !
bam! :)
Excelent work. Double sworded, swiftly slaying the stats serpent from planet r
Bam! :)
I love your channel and your way of explaining things!
Thank you
Thank you! :)
Thanks a lot. I'm looking forward to seeing more about multivariate adaptive regression splines!
I'll keep that topic in mind.
Fantastic video! Thank you
Thank you!
This is super as always thanks
Thanks!
The r^2, adjusted r^2, and p-value look good;
HOOOOOORAY!
bam!
great video! As usual, I should say :) what about diff-in-diff (with R example, possibly)?
I'll keep that in mind.
Thanks a lot Sir. 🤩💐
Thanks!
always amazing!
Thanks!
thank you very very much!!
BAM! :)
wow amazing video , thank you so much
Thanks!
why is the results of the tail predicting the linear regression of weight and vice versa?
What time point, minutes and seconds, are you asking about?
@@statquest starting at 6.13
@@sighsha3657 At that point we are testing how well we can predict "size" with and without specific variables in the model. So we see how well we can predict "size" with and without "weight" and we see how well we can predict "size" with and without "tail length". These tests help us asses how useful it is to use "weight" or "tail length" to predict "size". A small p-value suggests that a variable is useful.
Yeah, I'm struggling with this bit too. It feels like the coefficient line for 'weight' should be comparing the full model against a model with just weight as a predictor, but the explanation suggests the full model is being compared to a model with only tail length as a predictor.
Please can you do a video on multi nominal logistic regression in R?
I'll keep that in mind.
I'm sorry if I'm asking a stupid question, why p-value of weight can tell us using weight and tail isn't significantly better than only tail. Thank you so much
First you need to understand linear regression: th-cam.com/video/nk2CQITm_eo/w-d-xo.html and then you can find the answer to your question in this video that describes the theory of multiple regression: th-cam.com/video/zITIFTsivN8/w-d-xo.html
@@statquest thank you so much Sir, I appreciate it a lot
what is data mouse data i kinda confuse but your video really help me
Thanks!
hello, very helpful video. Anyway I have question though - why inference (06:35) 'using weight and tail isnt significanlty better than using tail alone' is derived from line "weight" and not from "tail" line below? Does the line "weight" of regression output compares exactly 'if using weight and tail is better than using tail alone to predict size'? shouldnt it be 'if predictor WEIGHT alone is better at prediction of size than MODEL WITH BOTH WEIGHT AND TAIL' instead or im wrong? I had course about multiple regression at academics couple years ago and trying to remind everything, I have found your video , but still I actually wanted to know that. Best wishes
edit: i remember that in stepwise regression we somehow exclude predictors which do not make a significant contribution and are therefore not statistically significant, so reading/watching about stepwise regression may do the thing for me?
For each variable that we test "weight" and "tail", we test the "full model" vs the model without that specific variable. So, for testing "weight" the full model is "weight + tail" and the model without that variable is just "tail". For testing "tail", the full model is "weight + tail" and the model with that that variable is just "weight." You can learn more details here: th-cam.com/video/zITIFTsivN8/w-d-xo.html
@@statquest thanks
i like that you sing; me writer director producer who is also very statistically literate for my work in psychology with daughter who is statistics major
Thank you!
Hey, thank you for your videos, it is really helpful, how do we conduct full inference for multiple linear regression model?
I'm not sure I understand your question. Can you elaborate on it?
@@statquest hey josh thanks for responding, I have done a linear regression model that describes the data best by using test based and criterion based model selection. I have been asked to "conduct full inference using the best fit model". I am slightly confused as to what needs to be done for this step, is it just the explanation of f-statistics and hypothesis testing obtained from summery of the model?
@@hrk201 That would be my guess, but it's just a guess.
So when do you use MLR versus multidimensional scaling?
Multidimensional Scaling is pretty different. To learn about it, first learn about PCA (only 5 minutes long: th-cam.com/video/HMOI_lkzW08/w-d-xo.html ) and then MDS: th-cam.com/video/GEn-_dAyYME/w-d-xo.html
StatQuest is TOTES CRAY CRAY🤣
Totes! :)
How do you know that the relationship between tail and weight? Is there any decision rules? I don't get how to conclude using r square and p values
A small p-value would cause us to reject the hypothesis that random noise generated the data. For details about p-values, see: th-cam.com/video/vemZtEM63GY/w-d-xo.html
Sorry if this is a stupid question, but is there a good way to format the results table? For instance, if I wanted the beta to be rounded to 3 digits, and the t and p values to be rounded to 4 digits?
Good question! The only way I can think of doing it is drawing it yourself using the original values (in this case, they are stored in the variable "multiple.regression") and running them through the round() function.
@@statquest Thanks. I bought your machine learning book, but have not had a single minute to sit down and read any of it. Maybe in a couple months...
@@jeffrisher6965 Thank you for your support! I hope you enjoy the book when you have time to read it. :)
What if the relationship between Y and one of the Xs is not linear?
Then you might need to use a different method.
Is there any video of yours on that? This is actually a matter that gets me a bit confused 😅
@@LuisSantiago-xo4fm When the relationship is non-linear, you can try regression trees: th-cam.com/video/_L39rN6gz7Y/w-d-xo.html and th-cam.com/video/g9c66TUylZ4/w-d-xo.html
how to transform large data to be like the smaller values in teh video?
What time point, minutes and seconds, are you asking about?
hey! should i perform any tests beforehand ? or not ?
is it better to perform this model with R or python ? and is it okay to have 20 observation per variable ?
It depends on what you mean by tests. However, usually multiple regression fits the model and then tests each variable as described. So this would be regression first, tests second.
20 observations per variable is find. And it's up to you if you want to use R or Python.
@@statquest um i thought i should perform the multicollinearity and heteroscedasticity and stationarity and do any correction before proceeding to fitting data?!
@@statquest can I contact you please on a more practical platform I have some confusions ://
the best
Thanks!
Matlab?
Maybe one day!
Not really 'step by step'
What parts would you like more details for?