Random effects structure for confirmatory hypothesis testing: Keep it maximal
Citations
1,679 citations
1,210 citations
Cites background from "Random effects structure for confir..."
...Burnham, Anderson & Huyvaert (2011) demonstrate how AIC approximates Kullback–Leibler information and provide some excellent guides for the best practice of applying ITmethods to biological datasets. Vaida & Blanchard (2005) provide details on how AIC should be implemented for the analysis of clustered data....
[...]
...Schielzeth & Forstmeier (2009); Barr et al. (2013) and Aarts et al. (2015) show that constraining groups to share a common slope can inflate Type I and Type II errors....
[...]
...Therefore, the approach of fitting the ‘maximal’ complexity of random effects structure (Barr et al., 2013) is perhaps better phrased as fitting the most complex mixed effects structure allowed by your data (Bates et al., 2015a), which may mean either (i) fitting random slopes but removing the…...
[...]
...Barr et al. (2013) suggest that researchers should fit the maximal random effects structure possible for the data....
[...]
...Therefore, the approach of fitting the ‘maximal’ complexity of random effects structure (Barr et al., 2013) is perhaps better phrased as fitting the most complex mixed effects structure allowed by your data (Bates et al....
[...]
1,045 citations
Cites methods from "Random effects structure for confir..."
...All simulations were run using the SIMGEN package (Barr et al., 2013) in R to fit models to simulated data, varying number of subjects and items systematically....
[...]
...To test this, corrected power (power’) was computed, as described by Barr et al. (2013); separate simulations were conducted, identi-...
[...]
...To test this, corrected power (power’) was computed, as described by Barr et al. (2013); separate simulations were conducted, identical to those described earlier in the paragraph, but with the null hypothesis set to true....
[...]
...Barr et al. (2013) use this test repeatedly in their simulations, and suggest that for the numbers of subjects and items typical of cognitive research these likelihood ratio tests are not particularly anti-conservative....
[...]
928 citations
Cites background or methods or result from "Random effects structure for confir..."
...One simple option, when numerically possible, is to t the full variance-covariance structure of random e ects (the maximal model; Barr et al., 2013), presumably to keep Type I error down to the nominal α in the presence of random e ects....
[...]
...…is supported by the data, usually by fixing some of the small variance components or correlation parameters to zero (for example, see discussion in Barr et al., 2013, p. 276).1 Fortunately, with enough data for every subject and every item, the programs may converge and may provide what looks…...
[...]
...Finally, we want to emphasize that we are not proposing a new dogma that is an alternative to the “keep it maximal” proposal of Barr et al. (2013)....
[...]
...Barr et al. (2013) refer to the first two cases as confirmatory and the last case as exploratory hypothesis testing....
[...]
...Fourth, the distinction between design-driven and data-driven, as introduced by Barr et al. (2013), misses an important confirmatory aspect in multivariate statistics: Any hypothesis about the support of variance components by the data requires a model comparison....
[...]
889 citations
Cites background or methods from "Random effects structure for confir..."
...The online supplement to Barr et al. (2013) fits such a model to data from an experiment described in Kronmüller and Barr (2007)....
[...]
...A full factorial model in the fixed-effects can be described by the formula , 1 + S + P + C + SP + SC + PC + SPC. Barr et al. (2013) analyzed Kronmüller and Barr (2007, Exp. 2) with the maximal model for this design comprising 16 variance components (eight each for the random factors SubjID and ItemID, respectively)....
[...]
...Barr et al. (2013) argue that failure to specify a maximal random effects structure amounts to violating the compound symmetry assumption in classical analysis of variance (anova)....
[...]
...Barr et al. (2013) base their recommendations on simulation studies comparing different procedures for model selection....
[...]
...Inclusion of these random slopes was motivated in part by the wish to provide stringent tests for the significance of main effects (cf. Barr et al., 2013), and in part by interest in individual differences....
[...]
References
272,030 citations
25,607 citations
24,665 citations
23,126 citations
19,282 citations
"Random effects structure for confir..." refers background in this paper
...For further information about random effect variance–covariance structures, see Baayen (2004, 2008), Gelman and Hill (2007), Goldstein (1995), Raudenbush and Bryk (2002), and Snijders and Bosker (1999a)....
[...]