Meta-analytic Procedure

For the current study, we conducted ten meta-analytic models. The first two included data on N170 latency or amplitude to faces from each study regardless of the hemisphere. To examine the influence of N170 lateralization, four additional models were conducted for N170 latency and amplitude data reported exclusively from the left or right hemisphere. Four additional meta-analyses were conducted examining P1 latency and amplitude to faces and N170 latency and amplitude to non-faces.

The ES, using Hedges’ g for small sample correction, was calculated using Comprehensive Meta-Analysis, Version 3 (CMA-3) (54). A random-effects model was used unless otherwise specified (55). Positive g indicated longer latency and less negative (i.e., smaller) amplitude of N170 response in ASD relative to TD (i.e., relatively less efficient or less robust N170 response in ASD). ESs of 0.2, 0.5, and 0.8 were considered to be small, medium, and large, respectively (56). A jackknife sensitivity analysis was conducted by repeating analyses excluding one study at a time to examine impact on the ES estimate (57).

Publication bias was assessed using funnel plots (58) and the tandem method suggested by Ferguson and Brannick (59) (see Supplement). Publication bias was deemed “probable” if indicated by all three criteria, “possible” if one or two criteria were met, and “unlikely” if no evidence of bias was found.

A significant Q statistic (60) and/or nontrivial heterogeneity according to the I² statistic (≥25%) (61) was used to warrant moderator analysis on each effect. Mixed-effects models were used (62). As information was reported separately for adults and children, these data were analyzed separately as an additional moderator. Moderator analyses were conducted using meta-regression for continuous variables (age, sex ratio, cognitive ability, ASD symptom severity), and using analogue-to-ANOVA for the categorical moderators (adult vs. child, use of gold-standard ASD diagnostic measure).