The linear mixed-effects modeling program 3dLME has been around for many years. However, one bottleneck of 3dLME has become suffocating: the limited adaptivity of random-effects specifications. A new program 3dLMEr has been just created to expand the adaptivity of 3dLME.
3dLMEr is a revised and advanced version of its elder brother 3dLME in the sense that the former is much more flexible in specifying the random-effects components than the latter. In addition, 3dLMEr uses the R package ‘lmerTest’ while 3dLME was written with the R package ‘nlme’, and the statistic values for main effects and interactions are approximated with the Satterthwaite’s approach. The greater flexibility of 3dLMEr lies in its adoption of random-effects notations by the R package ‘lme4’, as nicely summarized in the following table (change the variable ‘group’ in the table to ‘subject’ in the neuroimaging context) :
Similar to 3dLME, all the main effects and interactions are automatically available in the output while simple effects that tease apart those main effects and interactions would have to be requested through the option -gltCode -glfCode. Also, the 3dLMEr interface is largely similar to 3dLME except
the random-effects components are incorporated as part of the model specification, and thus the user is fully responsible in properly formulating the model structure through ‘-model …’ (option -ranEeff in 3dLME is no longer necessary for 3dLMEr);
the specifications for simple and composite effects through -gltCode and -glfCode are slightly simplified (the lable for each effect is part of -gltCode and -glfCode, and no more -gltLabel is needed); and
all the statistic values for simple effects (specified through -gltCode) are stored in the output as Z-statisc while main effects, interactions and the composite effects (specified through -gltCode) are represented in the output as chi-square with 2 degrees of freedom. The fixed number of DFs (i.e., 2) for the chi-square statistic, regardless of the specific situation, is not related to the real number of factor levels and is adopted simply for convenience because of the varying DFs due to the Satterthwaite approximation.