Multiple Linear Regression with Microsoft Excel

[@evathiel1455]

The music is the most annoying thing in this video, however the video itself is helpful! But damn this noise!!!!

[@ic9135]

Me: trying to focus on the video for my assignment Music: DUM DUMMMMM DUMDUMMM

[@sndpks156]

Multi linear regression is explained very well thank you for this video. Suggestion : Not to have background music while explaining, because some of your words are overlapping with music which is reducing your words clarity.

[@mikeba8879]

what is that music???? really? for what reason?

[@adamdagosto570]

Well I thought the music was fine, but with the song kept going!! More importantly, this was a really, really good explanation. Well done!!

[@sivachaitanya6089]

i can't believe i just watched the last video of this man. 😭 such a helpful video

[@morningjavajam2986]

oooof the sound. i want to watch but it's too distracting

[@karimmahdi6896]

keep on going man, great and simple explanation.

[@sammyespana7768]

How about the P-value of the Width IV, should we exclude it first before obtaining the Mileage formula?

[@MatthiasKullowatz]

Yes, we should probably drop width from the model. A few things to consider: 1. What is the p-value? This gives us some idea of whether the effect is statistically significant. 2. How correlated is this variable to others in the model? High correlation (absolute correlation coefficient > 0.8) is a good reason by itself to remove some variables from the model (keep one of the correlated variables). 3. How big is the effect? In many cases with small sample sizes, the coefficient may be relatively large, but the p-value is > 5%. This leaves us in a bit of a pickle. This could be an important variable with a great effect on the response, but we just don't have the sample size to say for sure. The compromise is to adjust the coefficient to some value between what it is now and zero. For width, maybe something like -0.05. One solution for these pickles is more advanced modeling. It's beyond the scope of this video/course, but penalized regression and ensembles of decision trees (random forest, GBM) can help find that compromise in a mathematically robust way (so you don't have to guess).

[@sammyespana7768]

Thank you for your response. I actually re-run the regression excluding the insignificant IVs before the calculations using the reduced regression model.

[@sammyespana7768]

You were such a big help. I'm an accountancy student.

[@ganesharya3512]

in the case of one independent variable equation can be guessed(by a plot) but how can I guess the equation in the case of more than one independent variable. suppose two independent variables (volume fraction and temperature ) then how can I guess the equation for fitting on excel (only by excel)

[@ericapowell9736]

is there a way to turn off the background music in this video...? i saw someone else comment the music was distracting. i agree...

[@my_nicholas]

That was a great explanation! Thank you!

[@dianacruz9352]

Where we can fin the database

[@narendrapulikonda5837]

which version of excel is this? my excel is not showing analysis tab

[@MatthiasKullowatz]

This is an old version of excel, but I think they all have the option. I think this link answers your question: support.office.com/en-us/article/load-the-analysis-toolpak-in-excel-6a63e598-cd6d-42e3-9317-6b40ba1a66b4

[@ghulamhabib948]

what should be the best goodness fit criteria between reduced chi square or adjusted R square

[@MatthiasKullowatz]

The adjusted R^2 is easier to understand in concept, I think. I have not personally used the reduced Chi Square metric directly, as described here: en.wikipedia.org/wiki/Reduced_chi-squared_statistic. In my opinion, there are more intuitive ways to understand if a model is "overfit" or "underfit". In practice, when I work with clients to build predictive models to help them make decision, we use a data science best practice known as "partitioning". This is where you split your data into multiple pieces, and build your model on a subset of the full data. The simplest approach might be to build a model on like 70% of your data, randomly sampled. This is often referred to as the "training" data. Then once you're happy with your model, you use it to make predictions on the other 30%--known often as the "holdout" dataset. There many ways to use these holdout predictions (and the corresponding actual Y responses) to determine "goodness of fit". Here's on nice visual way. With all the rows in the holdout dataset--now with predictions from the model--order the predictions from least to greatest and make 10 (or 20, or whatever) equally sized buckets. So if you had 100 rows in your holdout dataset, put the lowest 10 predictions into the first bucket. Then the next-lowest 10 predictions into the second bucket, etc. Each of those predictions corresponds to an entire rows of data, remember, and one piece of information in that row is the response, or Y, variable. Now, finally, within each bucket calculate the average predicted value and the average response Y value. This gives you two values for each bucket. Plot each set of values (predictions and responses) as two line graphs on the same plot. You're looking to see how close the average predictions are to the average responses.

[@teejmd91]

hey great vid! background music is a bit loud/distracting though

[@MatthiasKullowatz]

I know. I hate that music. I forgot to turn it off, recorded the whole thing, then never took the time to re-record. I'm a poop.

[@marnomiddelbos1209]

Still a great video, thanks

[@tlofrank5689]

@@MatthiasKullowatz music is annoying but the information is still great. thank you!

[@ganesharya3512]

how can i do a multiple nonlinear regression in excel

[@MatthiasKullowatz]

I suppose that by nonlinear regression you mean detecting curved relationships between x and y. The most straightforward way to do this is to create new columns in Excel that are functions of existing columns. The most straight-forward approach might be to just square a few columns. In the example, where weight is in column C, you could move mileage over to E and create a new column in D (=C2^2 and copy down). Then build the regression with all 4 X columns, including weight^2. There is some nuance here, but that should get you started. There is another whole thing called nonlinear regression, but it is typically overkill. See here: en.wikipedia.org/wiki/Nonlinear_regression

[@audreypalosse2356]

This is a great video ! Really well explained and easy to follow. However when I run my multiple regression all my p values show as #NUM!, but when I do all variables one by one the p values are normal - does anyone know how to fix this? Thanks!

[@MatthiasKullowatz]

I don't know that error off the top of my head. I would check that if you selected the column header, that you also checked the box that you included column headers.

[@yamax1]

why do you proceed if all the p-values are >0,05 means insignificant ?

[@MatthiasKullowatz]

Are you referring to the p-values for the individual slope coefficients in cells J18 - J20? Can you reference a time stamp in the video?

[@entrepreneursunited3097]

Loved this. My question is a little more complicated. Wish I could get some help from you on this! lol.

[@ryccoh]

I liked the music

[@MatthiasKullowatz]

Yes! That's what I'm talking about. Love the positivity! Let's do some stats!

[@meiligal]

can any body please explain What is "df", "SS, MS, F" in this video? im confuse About the formula to get the result : example : the gas mileage? i just learn this today, through this video, out of curiousity. thankyou.

[@MatthiasKullowatz]

Probably the most important of those metrics is the mean squared error (MSE), where the MS column meets the Residual row. That value represents a typical, or standard, error of prediction. Anytime you make a prediction, you have to be aware of how wrong it could be. This number helps define that error. The most intuitive way to derive the MSE value is by comparing all the Y values to all the predicted Y values. For each observation, take the difference between Y and predicted Y, then square it. Add all those values together, then divide by the Residual df (which, you'll notice, is quite close to the overall sample size). By dividing the sum of squared differences by a number close to the sample size, it's like find the average squared error. That is, the average squared difference between what actually happened and what the regression predicted would happen. The last step is to take the square root, since we squared all those differences earlier, so that the units are no longer squared. It would require a lot more explanation that can fit on this reply to go over each of those things you asked about, but I hope I've summarized the most important one adequately.

[@fourscoreand9884]

Very interesting, indeed. But, yes, the music is quite distracting. I understand it is fun to work on audio production, but, rather than adding distraction, why not try to improve the voice clarity by adding EQ, compression, etc?

[@jakob2746]

annoying music or sound or whatever shit that is

[@MatthiasKullowatz]

Agreed.

[@Janelif]

it's just so so so so so annyoying..........................

[@kanglarette]

Thank you so much

[@black-catmusic3619]

Is the data dependent? or indep

[@MatthiasKullowatz]

The observations, cars in this case, are independent of one another. The features of the cars, the columns, are dependent on one another. Mileage, for example, depends on the length, width, and weight of a car.

[@johnphillips4887]

Do we run into issues that the weight is dependent on the length & width?

[@katherinecheng3801]

omg thank you so much

[@connorrhodes6936]

I don't know that I hate the music. Kind of a nice sound.

[@anishjha7059]

Perfect!!!

[@nathanventurini2829]

Great video! Please get rid of the music.

[@rifatabdullahpranto9141]

Nice and easy explanation :) Loved it Keep up the good work. The music u used is shitty though :P

[@pavanbheesetty3207]

that music in the middle really sucks everytime

[@MatthiasKullowatz]

But what about at the ends?

[@BodmonBullet]

please no music man, heading to diff video

[@MatthiasKullowatz]

I'm crushed!

[@MatthiasKullowatz]

@@BodmonBullet I wrote that song!

[@BodmonBullet]

@@MatthiasKullowatz Its a nice song, but for the wrong video loool

[@harrytorres9101]

Great information...but P L E A S E....REMOVE!!!! the sound, music... or whatever it is....

[@Kimpressions333]

The music is annoying and not necessary.

[@jeromeavornu5211]

Which sought of barbarian video is this? The video isn't clear . Besides we use money to buy bundle in search of such videos. If you can't make a quality and stop fooling yourself

[@MatthiasKullowatz]

I was working on this, but fell a little short. www.imdb.com/title/tt0082198/

[@hubwells-testingwonders9274]

please eliminate this annoying sound

[@FirstnameLastname-et3fy]

This is the worst choice for background music ive ever seen, possibly.