(Simplified) Linear Mixed Model in R with lme()

Statistical modeling helps to compress the raw data we have into a simple mathematical formula that we can use for understanding the relationship between two or more variables, or in some situation, use to predict data from new input.

Simple linear model could easily help to model the relationship between two directly correlated variables, but in most cases, the world is too complicated to simple linear model. In this case, linear mixed model comes into play. Incorporating both fixed effects and random effects, this modeling technique attempt to prevent a false negative correlation between the variables, or mis interpretation of the trends.

This video is attempting to summarize the concept of modeling and how you can run LMM in R.

Scripts

Slides

Original tutorial

[salmaasghar]

you saved me dear. if i did not find this video i would quit my Ph.D. i am working in R and my experiments are also based on frequencies and their effect on the tone of the word. this solved my problem. i have no words to say thanks to you.

[charliebielby]

Most clear and concise lmem summary I've seen, showing the syntax makes it so clear to do it on r, no other videos I've seen do this

[andreanarayan]

I've never learned stats so easily, thank you so much for making this video!

[patrycjakasperska]

So great content! Thank you a lot. No one had ever explained LM to me in such a simple way before.

[MrJegerjeg]

(1 | random1) means that it is going to calculate a separate intercept for each one of the categories in "random1"

[christine_ko]

The video is of great use! Can't wait to watch more videos concerning R studio🥰

[AnkitCogn]

Well...never thought a fellow 9gager would one day teach me LMM...thanks op

[tinAbraham_Indy]

You saved me and my time. This video is beneficial. Thanks.

[jacobkenning]

insanely helpful, very clear explanation and workshop!

[veronicawardhani]

Thank you so much for your video! It really helped me during my master thesis. All the best for you :)

[DANICOG]

Hi, very useful and clear, thanks! When you have several fixed factors (e.g. 4 fixed factors), how do you set the null model to test the anova? Should you remove predictors one by one? Run anovas one by one with each null model with -1 predictor?

[genelad]

You made my life easier with this explanation! Thank you so much.

[pastramiking]

A quadratic model is a linear model, (look up polynomial regression). Linear doesn't mean line, it means the model can be expressed as a linear combination of weights and variables. Transforming the predictor variables does not make it non-linear because the model is Y given X so the predictors are always assumed to be fixed. If the coefficient has a square though then that is non-linear.

[kennedygolfhead]

Thank you for creating this video!! So easy to understand the LMM model. Just one question, how did you set your screen so that it shows the script and output all in one screen? That looks so intuitive and easy to look at!!! Thank you for your help!!!

[DM-qbjm]

Really great video! I think regularized regression techniques are outside of GLMs. Generalized linear models have a somewhat rigid statistical definition that is based on the form of the exponential family. Binomial regression, poisson regression, beta regression, gamma regression, etc. are all exponential family, which allows us to develop a GLM regression from those distributions. Regularization involves adding a penalty term that penalizes large coefficients, which does not match the exponential family distributional form. Regularization is a machine learning technique that starts to move away of traditional statistical techniques. At least, this is my understanding of how those ideas fit together!

[hope-g]

Thank you so much Sir for helpful tutorial. Could you show us how can we visualize mixed effect model with more than 1 predictors in R Please ?

[fazlfazl]

Great lecture. 
At 8.01, why will it be a problem to define race as 0, 1, 2, 3? Please explain.

[laurenfindsaway]

Thank you for this video! Your explanation of running the models in R is very intuitive. When writing up your model results, do you ever report statistical significance of your fixed effects? I tried publishing by comparing model fit alone and got strong push-back from reviewers. Thanks!

Thank you for the video. It helped me a lot.

[Tascioni]

Super useful, super clear. Thank you so much 😁