Do you really need a multilevel model? A preview of powerlmm 0.4.0

In this post I will show some of the new simulation features that will be available in powerlmm 0.4.0. You can already install the dev version from GitHub.

The revamped simulation functions offer 3 major new features:

  • Compare multiple model formulas, including OLS models (no random effects).
  • Evaluate a “forward” or “backward” model selection strategy using LRT.
  • Apply a data transformation during the simulation, this makes it possible to compare e.g. an ANCOVA versus a 2-level LMM, or write your own custom MNAR or MAR missing data function.

I will cover these new function in two examples.

Example 1

Do I really need a LMM? 2-lvl LMM versus ANCOVA

This example will show both the new data_transform argument and the new support for multiple formulas, and formulas without random effects. Each formula is fit to the same data set (or a transformed version) during the simulation. Let’s assume we’ll do a trial where we measure patients for 11 weeks, from baseline to week 10. We can analyze such data in many ways, as an example we will compare 3 popular models:

  • t-test: group differences at posttest
  • ANCOVA: group differences at posttest adjusting for pretest values
  • LMM: group differences in change over time using a 2-level linear-mixed effects model with a random intercept and slope

To make the LMM more similar to the ANCOVA we also fit a constrained model where we assume there is no group differences at baseline (which there isn’t). Let’s setup the models and run the simulation, we will try different amounts of random slope variance compared to the within-subjects variance (var_ratio), and with or without dropout (30 % at posttest).

Let’s summarize the results for the treatment effect.

Since we have 16 × 4 model results, it is probably better to plot the results.

unnamed chunk 6 1

We can see that for complete data ANCOVA is generally more powerful than the standard LMM as heterogeneity increases (“variance ratio”), whereas the constrained LMM is even more powerful. The difference between ANCOVA and t-test (“posttest”) also decrease with increasing heterogeneity in change over time—since this leads to a weaker correlation between pretest and posttest. For the scenarios with missing data, LMM is more powerful, as would be expected—the cross-sectional methods have 30 % of the observations missing, whereas the LMMs can include the available responses up until dropout occurs.

It is worth noting that the 2-lvl LMM with 11 repeated measures is not always more powerful than a “cross-sectional” t-test at posttest. Obviously, this is a limited example, but it demonstrates that it is a mistake to base sample size calculations on a t-test, when a LMM is planned, with the motivation that “a LMM will have more power” (I see such motivations quite often).

Example 2

Do I really need a multilevel model? Using LRT to perform model selection

A common strategy when analyzing (longitudinal) data is to build the model in a data driven fashion—by starting with a random intercept model, then add a random slope and perform a likelihood ratio test (LRT) and keep the random slope if it is significant, and so on. We can investigate how well such a strategy works using sim_formula_compare. We’ll define our model formulas, starting with a 2-level random intercept model and working up to a 3-level random intercept and slope model. The true model is a 3-level model with only a random slope. Let’s first setup the models

Then we run the simulation, the four model formulas in f will be fit the each data set.

During each simulation the REML log-likelihood is saved for each model, we can therefore perform the model selection in the summary method, as a post-processing step. Since REML is used it is assumed the fixed effects are the same, and that we compare a “full model” to a “reduced model”. Let’s try a forward selection strategy, where start with the first model and compare it to the next. If the comparison is significant we continue testing models, else we stop. The summary function performs this model comparison for each of the nsim simulations and returns the results from the “winning” model in each simulation replication.

The point of the model selection algorithm is to mimic a type of data driven model selection that is quite common. We see that this strategy do not lead to nominal Type I errors in this scenario. The cluster-level is left out of the model too often, leading to Type I errors around 12 %. However, it is fairly common to increase the LRT’s alpha level to try to improve this strategy. Let’s try a range of alpha level to see the impact.

unnamed chunk 11 1

The figure shows that the LRT alpha level need to be very liberal to keep Type I errors, for the treatment effect, close to the 5 % level.

We can also see the results from each of the four models. Here we will just look at the time:treatment effect.

We see that the 2-lvl random intercept model lead to substantially inflated Type I errors = 0.495. The 2-level model that also adds a random slope is somewhat better but still not good, Type I errors = 0.215. The correct 3-level model that account for the third level using a random slope have close to nominal Type I errors = 0.054. The full 3-level that adds an unnecessary random intercept is somewhat conservative, Type I errors = 0.043.

When choosing a strategy Type I errors is not only factor we want to minimize, power is also important. So let’s see how power is affected.

We can note that power for the treatment effect based on LRT model selection is only slightly higher than for the correct 3-level model. If we balance this slight increase in power compared to the noticeable increase in Type I errors, it might be reasonable to conclude that for these data we should always fit the 3-level model. Lastly, the overspecified 3-level model that include an unnecessary random intercept loses some power.

Summary

The improved simulation method in powerlmm makes it really convenient to plan and evaluate the analysis of longitudinal treatment trials with a possible third level of clustering (therapists, schools, groups, etc). The support for data transforms and single level (lm()) models also enables a lot of custom models to be fit.

Feedback, bugs, etc

I appreciate all types of feedback, e.g. typos, bugs, inconsistencies, feature requests, etc. Open an issue on github.com/rpsychologist/powerlmm/issues, common on this post, or contact me here rpsychologist.com/about.


Written by Kristoffer Magnusson, a researcher in clinical psychology. You should follow him on Twitter and come hang out on the open science discord Git Gud Science.


Share:

Published May 04, 2018 (View on GitHub)

Buy Me A Coffee

A huge thanks to the 79 supporters who've bought me a 179 coffees!

Jason Rinaldo bought ☕☕☕☕☕☕☕☕☕☕ (10) coffees

I've been looking for applets that show this for YEARS, for demonstrations for classes. Thank you so much! Students do not need to tolarate my whiteboard scrawl now. I'm sure they'd appreciate you, too.l

Someone bought ☕☕☕☕☕ (5) coffees

What a great site

@Daniel_Brad4d bought ☕☕☕☕☕ (5) coffees

Wonderful work!

David Loschelder bought ☕☕☕☕☕ (5) coffees

Terrific work. So very helpful. Thank you very much.

@neilmeigh bought ☕☕☕☕☕ (5) coffees

I am so grateful for your page and can't thank you enough!  

@giladfeldman bought ☕☕☕☕☕ (5) coffees

Wonderful work, I use it every semester and it really helps the students (and me) understand things better. Keep going strong.

Dean Norris bought ☕☕☕☕☕ (5) coffees

Sal bought ☕☕☕☕☕ (5) coffees

Really super useful, especially for teaching. Thanks for this!

dde@paxis.org bought ☕☕☕☕☕ (5) coffees

Very helpful to helping teach teachers about the effects of the Good Behavior Game

@akreutzer82 bought ☕☕☕☕☕ (5) coffees

Amazing visualizations! Thank you!

@rdh_CLE bought ☕☕☕☕☕ (5) coffees

So good!

Someone bought ☕☕☕ (3) coffees

Ruslan Klymentiev bought ☕☕☕ (3) coffees

@lkizbok bought ☕☕☕ (3) coffees

Keep up the nice work, thank you!

@TELLlab bought ☕☕☕ (3) coffees

Thanks - this will help me to teach tomorrow!

SCCT/Psychology bought ☕☕☕ (3) coffees

Keep the visualizations coming!

@elena_bolt bought ☕☕☕ (3) coffees

Thank you so much for your work, Kristoffer. I use your visualizations to explain concepts to my tutoring students and they are a huge help.

A random user bought ☕☕☕ (3) coffees

Thank you for making such useful and pretty tools. It not only helped me understand more about power, effect size, etc, but also made my quanti-method class more engaging and interesting. Thank you and wish you a great 2021!

@hertzpodcast bought ☕☕☕ (3) coffees

We've mentioned your work a few times on our podcast and we recently sent a poster to a listener as prize so we wanted to buy you a few coffees. Thanks for the great work that you do!Dan Quintana and James Heathers - Co-hosts of Everything Hertz 

Cameron Proctor bought ☕☕☕ (3) coffees

Used your vizualization in class today. Thanks!

eshulman@brocku.ca bought ☕☕☕ (3) coffees

My students love these visualizations and so do I! Thanks for helping me make stats more intuitive.

Someone bought ☕☕☕ (3) coffees

Adrian Helgå Vestøl bought ☕☕☕ (3) coffees

@misteryosupjoo bought ☕☕☕ (3) coffees

For a high school teacher of psychology, I would be lost without your visualizations. The ability to interact and manipulate allows students to get it in a very sticky manner. Thank you!!!

Chi bought ☕☕☕ (3) coffees

You Cohen's d post really helped me explaining the interpretation to people who don't know stats! Thank you!

Someone bought ☕☕☕ (3) coffees

You doing useful work !! thanks !!

@ArtisanalANN bought ☕☕☕ (3) coffees

Enjoy.

@jsholtes bought ☕☕☕ (3) coffees

Teaching stats to civil engineer undergrads (first time teaching for me, first time for most of them too) and grasping for some good explanations of hypothesis testing, power, and CI's. Love these interactive graphics!

@notawful bought ☕☕☕ (3) coffees

Thank you for using your stats and programming gifts in such a useful, generous manner. -Jess

Mateu Servera bought ☕☕☕ (3) coffees

A job that must have cost far more coffees than we can afford you ;-). Thank you.

@cdrawn bought ☕☕☕ (3) coffees

Thank you! Such a great resource for teaching these concepts, especially CI, Power, correlation.

Julia bought ☕☕☕ (3) coffees

Fantastic work with the visualizations!

@felixthoemmes bought ☕☕☕ (3) coffees

@dalejbarr bought ☕☕☕ (3) coffees

Your work is amazing! I use your visualizations often in my teaching. Thank you. 

@PsychoMouse bought ☕☕☕ (3) coffees

Excellent!  Well done!  SOOOO Useful!😊 🐭 

@whlevine bought ☕☕ (2) coffees

Thank you so much for these amazing visualizations. They're a great teaching tool and the allow me to show students things that it would take me weeks or months to program myself.

Someone bought ☕☕ (2) coffees

@notawful bought ☕☕ (2) coffees

Thank you for sharing your visualization skills with the rest of us! I use them frequently when teaching intro stats. 

@inthelabagain bought ☕ (1) coffee

Really wonderful visuals, and such a fantastic and effective teaching tool. So many thanks!

Someone bought ☕ (1) coffee

Someone bought ☕ (1) coffee

Yashashree Panda bought ☕ (1) coffee

I really like your work.

Ben bought ☕ (1) coffee

You're awesome. I have students in my intro stats class say, "I get it now," after using your tool. Thanks for making my job easier.

Gabriel Recchia bought ☕ (1) coffee

Incredibly useful tool!

Shiseida Sade Kelly Aponte bought ☕ (1) coffee

Thanks for the assistance for RSCH 8210.

@Benedikt_Hell bought ☕ (1) coffee

Great tools! Thank you very much!

Amalia Alvarez bought ☕ (1) coffee

@noelnguyen16 bought ☕ (1) coffee

Hi Kristoffer, many thanks for making all this great stuff available to the community!

Eran Barzilai bought ☕ (1) coffee

These visualizations are awesome! thank you for creating it

Someone bought ☕ (1) coffee

Chris SG bought ☕ (1) coffee

Very nice.

Gray Church bought ☕ (1) coffee

Thank you for the visualizations. They are fun and informative.

Qamar bought ☕ (1) coffee

Tanya McGhee bought ☕ (1) coffee

@schultemi bought ☕ (1) coffee

Neilo bought ☕ (1) coffee

Really helpful visualisations, thanks!

Someone bought ☕ (1) coffee

This is amazing stuff. Very slick. 

Someone bought ☕ (1) coffee

Sarko bought ☕ (1) coffee

Thanks so much for creating this! Really helpful for being able to explain effect size to a clinician I'm doing an analysis for. 

@DominikaSlus bought ☕ (1) coffee

Thank you! This page is super useful. I'll spread the word. 

Someone bought ☕ (1) coffee

Melinda Rice bought ☕ (1) coffee

Thank you so much for creating these tools! As we face the challenge of teaching statistical concepts online, this is an invaluable resource.

@tmoldwin bought ☕ (1) coffee

Fantastic resource. I think you would be well served to have one page indexing all your visualizations, that would make it more accessible for sharing as a common resource.

Someone bought ☕ (1) coffee

Fantastic Visualizations! Amazing way to to demonstrate how n/power/beta/alpha/effect size are all interrelated - especially for visual learners! Thank you for creating this?

@jackferd bought ☕ (1) coffee

Incredible visualizations and the best power analysis software on R.

Cameron Proctor bought ☕ (1) coffee

Great website!

Someone bought ☕ (1) coffee

Hanah Chapman bought ☕ (1) coffee

Thank you for this work!!

Someone bought ☕ (1) coffee

Jayme bought ☕ (1) coffee

Nice explanation and visual guide of Cohen's d

Bart Comly Boyce bought ☕ (1) coffee

thank you

Dr. Mitchell Earleywine bought ☕ (1) coffee

This site is superb!

Florent bought ☕ (1) coffee

Zampeta bought ☕ (1) coffee

thank you for sharing your work. 

Mila bought ☕ (1) coffee

Thank you for the website, made me smile AND smarter :O enjoy your coffee! :)

Deb bought ☕ (1) coffee

Struggling with statistics and your interactive diagram made me smile to see that someone cares enough about us strugglers to make a visual to help us out!😍 

Someone bought ☕ (1) coffee

@exerpsysing bought ☕ (1) coffee

Much thanks! Visualizations are key to my learning style! 

Someone bought ☕ (1) coffee

Sponsors

You can sponsor my open source work using GitHub Sponsors and have your name shown here.

Backers ✨❤️

Questions & Comments

Please use GitHub Discussions for any questions related to this post, or open an issue on GitHub if you've found a bug or wan't to make a feature request.

Webmentions

There are no webmentions for this page

(Webmentions sent before 2021 will unfortunately not show up here.)

Archived Comments (1)

B
blazej 2018-08-24

Hi, could you please help me out and explain if there is a possibility to simulate data and power for studies where with cont. predictors at different levels of analysis (e.g.level1 and level2) - specifically without any groups? To be more specific I'm trying to simulate `y` (as in `y ~ 1 + (1|id)`) for a given between / within variance ratio, and then add to it continuous `x1` and `x2` covariates and analyse `y ~ x1 + x2 + (x1 | id)`. I'd really use some advice and help. Been stuck with this problem for a week now. Thank you