I just watched all 17 videos in this playlist in 1 day. At the start I understood nothing about multiple regression, I didn't even know what the numbers and letters in the equation meant. Now I feel ready to pass my statistics test in 2 days. Thank you very much Brandon
I have watched all the videos on linear regression and multiple regression. It was detailed explaination. Now I can see regression from few different angles.
I am new to statistics and i'm reading a research where i need to understand those variables.. idk.. i mean its a medicine research and they're using advanced statistics ???? u know science these days .. insane stuff anyways i understood the fact that lower AIC and BIC values are better.. feels like anyone that wants to conduct a research in any topic needs to be an experienced statistician first so that he/she can analyze data and understands metrics like what we have here
thank you so much for all these inspirational lectures. I found that such knowldge and understanding of statistics cannot be seized or achieved through books. Your explanations has made the whole process both efficient and enjoyable.
I watched another video showing that k should include the intercept and the error term. So k should be 3 + 2 in this case. If we substitute 5 into k in the AICc calculation at 10:53, we'll get the exact AICc value as in the tool
Because they used 4 term in the model, we then have 5 parameters. If you change number of parameter (k) to be k=5 then the AICc will be 1024.28, probability.
At 10:40 or so you assume that the difference between what the package got for AICc and what you got is due to your use of the approximate formula based on SSE values. And, so, you go on to say that the difference isn't important, as long as you use the same approximator for all other models, so that the relative values of AICc show the same pattern. But what is the error in the approximator depends on any of the inputs? What if it's higher, for example, for models with more parameters? That could cause the approximate method to favor a different model from the exact method. Unless and until you can tell us upon what the error of the approximator depends, anything based on the approximate value for log-likelihood is highly suspect. And the same goes for the "trick" of getting an approximate Bayes Factor from the SSE values (not that anyone should actually use Bayes Factors for anything other than impressing the Great Unwashed ... at least until they deal with Lindley's Paradox).
Heey!! I would like to use the information conatined in this video in my final economic project if that's fine. How could I mention it in the bibliography?? Thanks in advance!!
I didn't understand one thing: at the point when my output (I use Jamovi software) gives me R square, R square adjusted, AIC, BIC and RMSE, what is the order in which I look at these outputs? For example: Model 1 (one variable): R2 = 0.365, R2adj = 0.347, AIC = 201, BIC = 206, RMSE = 3.14 Model 2 (two variables): R2 = 0.378, R2adj = 0.343, AIC = 202, BIC = 208, RMSE = 3.10 Model 3 (three variables): R2 = 0.378, R2adj = 0.324, AIC = 204, BIC = 212, RMSE = 3.10 Which of the three models do I choose? I would say from what I understand: - for AIC I would say model 1 is the best - for the BIC I would say that the model 1 is the best one - for the RMSE I would say that the model 2 and the model 3 are equal So... should I choose model 1?
I was seeing all your videos with multiple regression playlist , can you tell me about what is correlations , like multiple correlation coefficient, partial correlation coefficient etc
I just watched all 17 videos in this playlist in 1 day. At the start I understood nothing about multiple regression, I didn't even know what the numbers and letters in the equation meant. Now I feel ready to pass my statistics test in 2 days.
Thank you very much Brandon
I have watched all the videos on linear regression and multiple regression. It was detailed explaination. Now I can see regression from few different angles.
As always, very well done. I appreciate the way you explain the concepts so those new to statistics can understand them!
I am new to statistics and i'm reading a research where i need to understand those variables.. idk.. i mean its a medicine research and they're using advanced statistics ???? u know science these days .. insane stuff anyways i understood the fact that lower AIC and BIC values are better.. feels like anyone that wants to conduct a research in any topic needs to be an experienced statistician first so that he/she can analyze data and understands metrics like what we have here
thank you so much for all these inspirational lectures. I found that such knowldge and understanding of statistics cannot be seized or achieved through books. Your explanations has made the whole process both efficient and enjoyable.
I was hoping you'd put this up. I was looking forward to hearing your verson of this. Thank you!
I comment each videos. Good job Brandon
Thank you so much. This is very helpful.
I watched another video showing that k should include the intercept and the error term. So k should be 3 + 2 in this case. If we substitute 5 into k in the AICc calculation at 10:53, we'll get the exact AICc value as in the tool
THANK YOU
Because they used 4 term in the model, we then have 5 parameters. If you change number of parameter (k) to be k=5 then the AICc will be 1024.28, probability.
At 10:40 or so you assume that the difference between what the package got for AICc and what you got is due to your use of the approximate formula based on SSE values. And, so, you go on to say that the difference isn't important, as long as you use the same approximator for all other models, so that the relative values of AICc show the same pattern.
But what is the error in the approximator depends on any of the inputs? What if it's higher, for example, for models with more parameters? That could cause the approximate method to favor a different model from the exact method.
Unless and until you can tell us upon what the error of the approximator depends, anything based on the approximate value for log-likelihood is highly suspect. And the same goes for the "trick" of getting an approximate Bayes Factor from the SSE values (not that anyone should actually use Bayes Factors for anything other than impressing the Great Unwashed ... at least until they deal with Lindley's Paradox).
perfect timing
Heey!! I would like to use the information conatined in this video in my final economic project if that's fine. How could I mention it in the bibliography?? Thanks in advance!!
Yes of course! Depending on the citation system you are using there is a way to cite RUclips videos and other online content.
I didn't understand one thing: at the point when my output (I use Jamovi software) gives me R square, R square adjusted, AIC, BIC and RMSE, what is the order in which I look at these outputs?
For example:
Model 1 (one variable): R2 = 0.365, R2adj = 0.347, AIC = 201, BIC = 206, RMSE = 3.14
Model 2 (two variables): R2 = 0.378, R2adj = 0.343, AIC = 202, BIC = 208, RMSE = 3.10
Model 3 (three variables): R2 = 0.378, R2adj = 0.324, AIC = 204, BIC = 212, RMSE = 3.10
Which of the three models do I choose? I would say from what I understand:
- for AIC I would say model 1 is the best
- for the BIC I would say that the model 1 is the best one
- for the RMSE I would say that the model 2 and the model 3 are equal
So... should I choose model 1?
Thanks for this great video.
I was seeing all your videos with multiple regression playlist , can you tell me about what is correlations , like multiple correlation coefficient, partial correlation coefficient etc
Namaste Brendon
Can we select neural network models using AIC?
Technically, yes! But in practice is where cross validation is the preferred method to avoid over fitting.
Namaste 🙏🏻
As promised last Saturday, here comes the video on AIC & BIC.
I use it in clustering
That's how I was first introduced to it.