2.7. fMRI data analysis

Whole-brain general linear model (GLM) analyses were performed using SPM 12. First-level analyses were performed at an uncorrected threshold of p < .001, contrasting each participant’s experimental task runs against the control task (i.e., [Task > Control]) and against parametric levels (i.e., [Near > Far]) within tasks (e.g., [Arabic_Near > Arabic_Far]). Both erroneous and correct trials were included in the analysis. Voxels surviving first-level analysis were included in a second-level nonparametric two-sample T-test, conducted using the Statistical Nonparametric Mapping (SnPM version 13.1.08; http://nisox.org/Software/SnPM13/) toolbox for SPM 12. Analyses were performed at a height threshold of p < .001 and a familywise error corrected (FWE) cluster-forming threshold of p < .05. Second-level analyses were variance smoothed in accordance with the smoothing kernel (i.e., 4 mm) and subject to 10,000 permutation tests. Cluster extent thresholds for each analysis were calculated as the critical suprathreshold cluster size (STCS), as implemented in SnPM.

The use of nonparametric second-level analyses was motivated by the group size imbalance (30 versus 44 subjects) together with the use of an aggressive denoising pipeline. Both factors yield imbalanced comparative analyses, particularly given that child participants are subject to more movement and thus a loss in temporal degrees of freedom. Moreover, nonparametric statistics require few assumptions and offer strong control over the type I error rate (e.g., Nichols and Holmes, 2001).

Conjunction analyses of overlapping activity in child and adult participants, across all three tasks, were performed by subjecting FWE-corrected T-maps from individual one-sample T-tests (e.g., [Arabic_Child > Control_Child]) to a minimum statistic conjunction null analysis (Nichols et al., 2005) using SPM’s ImCalc (min) function. The largest STCS cluster extent found throughout one-sample T-tests was used as the minimum extent for conjunction analyses.