Second-level analyses were conducted based on the statistical programming language R (http://cran.r-project.org/) (i.e., a mixed-model analysis used the R program lme). Specifically, a voxel-wise model was used to examine the fixed effects consisting of emotion type, group, education, and response latency, and the random effects were subjects (i.e., an individual intercept was fitted for each subject). These were performed within the constraints of ROIs to reduce Type 1 errors. Voxel-wise multiple linear regression analyses were conducted with latency to respond to angry, fearful, or happy faces as independent measures, and the percent signal change between faces and the sensorimotor control condition as the dependent measure using the lm program of R. Subsequently, clusters of activation within the anatomically-constrained functional regions of interest were considered significant only if the clusters were larger than the above-mentioned volume thresholds. The average of each cluster was tabulated across the different experimental conditions. All analyses covaried for both CBF (overall blood flow differences in ml/100g/min for placebo vs. alcohol sessions) and for usual drinks per occasion.
