To register for this workshop (open to the academic community having some experience with R): please go to http://www.crblm.ca/events/linear_mixed_models_workshop
Linear Mixed Models (LMMs) are increasingly replacing traditional F1/F2-mixed-model/repeated-measure ANOVAs (rmAOVs; i.e., designs with both between- and within-subject/item factors, e.g., Kliegl, Risse & Laubrock, 2007) or repeated-measures multiple regressions (rmMRs; e.g., Kliegl, 2007) for statistical inference about experimental effects on fixation durations or response latencies.
This workshop focuses on analyses of these two types of dependent variable. The goal of this workshop is to teach how LMMs go beyond the usual rmAOV and rmMR analyses, using the lmer()-function of the lme4 package (Version 1.1-4; Bates, Maechler, Bolker, & Walker, 2014) in the R environment (R Core Team, 2013). The workshop is structured into four sections:
- Transformation of dependent variable (Kliegl et al., 2010)
- Contrast specification of fixed effects (Kliegl & Vasishth, 2013)
- LMMs (Kliegl et al., 2011; Hohenstein & Kliegl, 2014; Masson & Kliegl, 2013)
- Specification of fixed effects, variance components, correlation parameters
- Model selection through model comparison (LRT, AIC, BIC)
- Confidence intervals for model parameters
- Computation and visualization of LMM partial effects with the remef() function (Hohenstein &
- Visualization of individual differences and item differences in (quasi-) experimental effects (Kliegl et al., 2011)
This workshop does not offer an introduction to R. The content of the workshop is aimed at scientists who already have basic knowledge of R and have been carrying out traditional statistical analyses in this environment. To facilitate preparation of the workshop for organizers, participants should choose one of the following four options during registration.
With registration, submit a paper package (zipped file) consisting of (1) PDF of a publication or a PDF of a preprint of an in-press or submitted paper, (2) the data for one experiment in this study, (3) an R Script with the code (a) for reading the data, (b) for computing the summary statistics (Ms and SDs for the design cells), (c) for the rmAOV (or rmMR), and optionally (d) also for an LMM. As to optional (d), the LMM code could represent your best effort; it may not be correct or there may be better or alternative ways to specify the model. Obviously, the goal of the workshop is to learn defensible model specifications and corollary analyses.
Complexity of experiment submitted with registration
If your experiment is typically analyzed with some form of analysis of variance, the design must include at least 3 measures per subject (i.e., one within-subject factor with 3 levels). Preferably, the design should comprise at least a 2 x 2 within-subject factor design (i.e., a minimum of 4 measures per subject). The minimum number of subjects should be 30; preferably around 50. If the experiment contains a second random factor (e.g., items), there should be at least 20 instances (levels). If your data has fewer participants or items than these recommendations, you can still use these data for the workshop, but should be aware that you may lack statistical power for a serious test of your effects.
If the experiment includes continuous covariates, the design should include both a within-subject factor (e.g., experimental condition) and a continuous within-subject/item (repeated-measures) covariate (e.g., log word frequency for subjects; language skill of subjects for items). The minimum number of subjects should be 30; preferably for such a design the number of subjects should be around 50. If the experiment contains a second random factor (e.g., Items), there should be at least 20 instances (levels).
If your experiment is much more complex than specified above, you may want to select a subset of the design for the purpose of the workshop and deal with its full complexity afterwards.
The demonstrations in the workshop will be based on the analyses reported in the cited paper packages (i.e., PDF of paper or preprint plus data and R scripts). They and many other paper packages of such analyses, including also other tutorial material, are available at the Potsdam Mind Research Repository (PMR2; http://read.psych.uni-potsdam.de/pmr2/) or at the Mind Research Repository (MRR; http://openscience.uni-leipzig.de/). Probably there will also be a special LMM-workshop website or moodle account providing access to these paper packages, background reading, links to related websites, and workshop slides.
Other participants / other related topics
There is probably some benefit of the workshop for participants with good knowledge of statistics, but without enough specific knowledge of statistical analyses with R to submit a paper package. Typically, such participants legitimately simply “want to know what this is all about”. As the time for the program as sketched above is tight, there may not be enough time to discuss issues that go much beyond the immediate practical needs of the intended audience, especially with respect to time spent to analyze one’s data during the workshop. So in this workshop the emphasis will be on “how to analyze data with LMMs in R”, not on “what are LMMs”.
Some participants may be interested in other topics. This LMM workshop will not cover Generalized Linear Mixed Models (i.e., analyses of binary dependent variables such as 0/1 accuracy or 0/1 skipping) or other related mixed model analyses such as Nonlinear Mixed Models or Generalized Additive Mixed Models.