R - Hierarchical Multiple Regression

[@Artyom109Zinchenko]

Dear Dr. Buchanan, thank you very much for the videos and for sharing everything at OSF! I really like your style of teaching.

[@StatisticsofDOOM]

Thanks! Appreciate the kind words.

[@jasoncain1575]

You started this off just like I start off my R video walkthroughs for students. I laughed pretty hard.

[@StatisticsofDOOM]

Many many spelling errors will be found across these videos and life just in general. Much easier to notice when recording ;).

[@VictorianoOchoa]

Very very helpful. Thank you so much!

[@StatisticsofDOOM]

Thank you!

[@eamoncolvin1748]

Has anyone cited this video in a manuscript? It's the clearest procedural description of hierarchical regression I've seen and I'd like to credit it for an upcoming paper!

[@StatisticsofDOOM]

Thank you! APA has a new version for youtube: www.scribbr.com/apa-examples/youtube/ (I've done this for tutorials when I've used them and they aren't papers officially).

[@kristinabeau]

Thank you so much for sharing your knowledge with us ! I use your videos so often in order to enhance my statistical skills. I was wondering, what do we do if a variable of the first step model isn’t significant ? So I run a new model without it and then go to the next step ? Or do I need to keep it, as it was part of my hypothesis ?

[@StatisticsofDOOM]

I would say you’d keep running your planned steps, just noting step 1 was not significant.

[@mr235]

Hey, Dr. Buchanan, thanks for your videos, very helpful stuff! I have a question: If I run a Hierarchical Multiple Regression where I add my control variables in the first model and then add 4 IVs in the second model, what is my number of IVs for the leverage? Do I put "k=4" or "k = number of control variables" or "k=4 IVs + Amount of control variables?" Thanks in Advance for the help!

[@StatisticsofDOOM]

When you are doing the screening you would run the last model (i.e. with all the variables). Then k would equal how many variables you have total, including any control variables, dummy coded columns, etc. If you want to use control variables, it's normal to put them in the first step yes.

[@mr235]

@@StatisticsofDOOM Thank you for your answer!

[@eamoncolvin1748]

Any tips on how to make a plot that showcases the *differences* between the models? The plot you suggest includes all 3 IVs (Age + Gender + Extraversion). However, if the research question of interest is how much value Extraversion adds to predicting Car Care (over and above Age and Gender) this summary figure isn't the most informative. Any thoughts?

[@StatisticsofDOOM]

True ... hmmm, that's usually an R2 question sooooo maybe, you could add a separate regression line to the plot which has the predicted values for the model without extraversion for comparison? One of them should be "closer" to the participant points. I"m thinking of those textbook type pictures which have like a flat line then have a tilted line to show you how much better it matches the dots.

[@eamoncolvin1748]

@@StatisticsofDOOM Thanks for the idea! I'll give it a shot. Very grateful for all of your videos and insights :)

[@daphnenakhid2881]

Thank you for the video! When calculating power, why is the partial r-squared 0.09 for a medium effect size? Shouldn't f squared be 0.15? Thank you!

[@StatisticsofDOOM]

You can go either way! I am using r = .3 as a medium correlation, which squared is .09. You should always use values that make sense your own research field.

[@hollysullivan-toole4872]

Thank you so much for all of your videos--you are a great teacher!!! About the partial correlations, when I used your code for computing pr on my own dataset (with more control variables than your example dataset), I got different values than the ones SPSS gave me. I used this code instead and got the same values as my SPSS output: partial_X = pcor.test((Data[ , c(X)]), (Data[ , c(Y)]), (Data[ , c(Z1,Z2,Z3)])) where X = the column with X variable, Y = column with Y variable (criterion), and the Z values are columns for all variables I wanted to control for. Maybe I did something wrong when I tried your code on my data, but I wanted to add this other method here in case anyone else had trouble computing partial correlations.

[@StatisticsofDOOM]

I’d have to look at that function - don’t forget to square them for effect size.

[@ibrahimbashimam9149]

Thanks a lot !!!!

[@stevehawley6737]

Fantastic tutorial. Thanks!