FAQ/mixedR - CBU statistics Wiki

Revision 26 as of 2012-09-26 09:03:42

Clear message
location: FAQ / mixedR

Where can I find out about using random effects models in R including obtaining proportion of variance attributable to a variable?

Random effects models are becoming increasingly used for both continuous and binary outcomes. Jaeger (2008) recommends using binary logistic mixed models for the latter cases after noting shortcomings using the traditional arcsine transformed responses in ANOVAs to approximating proportions near to zero or one.

Paul Bleise has written an [attachment:nlmeR.pdf introductory guide ] to fitting random effects models in R using the nlme software with some case studies and general tips on using R objects.

The guide mentions, in particular, the function VarCorr() which computes residual variances of models specified to VarCorr. Comparisons of VarCorrs() related to nested models enables the computation of percentage of variance explained by variable sets. For example if VarCorr(y~a+b) gives residual variance A and VarCorr(y~a) gives residual variance B (<=A) the proportion of variance explained by variable B = 1 - B/A. Further details of this R-squared for random effects may be found in Snijders and Bosker (1999).

You can also fit random effects models in R using the lme4 program which computes percentiles of Monte-Carlo Markov Chain (mcmc) for regression estimates and variance components derived from simulations using the random effects model. The usual REML estimates are also produced. The median (50 percentile) of the mcmc estimates should approximately equal the analogous REML ones.

There is also a R guide on multilevel modelling by Baayen, R.H. (2008) in pdf format [attachment:Baayenlmer.pdf here.] SPSS have a 28 page [attachment:spssmixed.pdf document] giving case studies using GLM and MIXED procedures. Data input, syntax and interpretation of output are all considered. This document is also available on-line from [http://www.spss.ch/upload/1126184451_Linear%20Mixed%20Effects%20Modeling%20in%20SPSS.pdf here].

There are some excellent comprehensive descriptions in these selected slides (in pdf format) for fitting Mixed models in R using the lme4 package (Bates and Rahway, 2010) using lmer and glmer procedures:

[attachment:3simple.pdf Part 3: Linear mixed models with simple, scalar random effects]

[attachment:4longitudinal.pdf Part 4: Longitudinal data, modeling interactions]

[attachment:7GLMM.pdf Part 7: Generalized linear mixed models]

References

Baayen, R.H. (2008) Analyzing linguistic data:a practical introduction to statistics using R. Cambridge University Press.

Baguley, T. (2012) Serious Stats. A guide to advanced statistics for the behavioral scoences. Palgrave MacMillan:New York. Comprehensive coverage with R code in Chapter 18.

[attachment:arcsine.pdf Jaeger, T.F. (2008) Categorical data analysis:away from ANOVAs. ] J Mem Lang 59(4) 434-446.

Snijders, T. and Bosker, R. (1999) Multilevel analysis: an introduction to basic and advanced multilevel modelling. Sage:London.

Wright, D.B. & London, K. (2009) Multilevel modelling: Beyond the basic applications. British Journal of Mathematical and Statistical Psychology 62, 439-456. This is a teaching primer article including worked examples of data management prior to analysis. A PDF copy of this paper is available [attachment:randeff.pdf here.]