Reanalysis of Individual Differences in Face Identity Processing: A Robustness Reproduction, One-Factor Test, and Exploratory Network Analysis of McCaffery et al. (2018)

Background. Face-identity processing involves several related abilities, including matching unfamiliar faces, learning new faces, and recognising familiar people. McCaffery et al. (2018) examined individual differences in these abilities and their relationships with perceptual, cognitive, and personality measures. We report a robustness reproduction of their Study 2 analyses, using the publicly available participant-level data. Methods. The Study 2 dataset contains 107 participants and 22 measured variables: five face-processing measures and 17 comparison measures. We recalculated the 85 Pearson correlations between face-processing and comparison measures and compared them with the published Table 4. We applied the Benjamini–Hochberg (BH) procedure across the 85 tests and, separately, across the ten face–face correlations. We fitted a one-factor confirmatory factor model with a nonparametric bootstrap of variance explained, followed by exploratory two-factor and sensitivity models. We estimated a regularised partial-correlation network (Spearman correlations, EBICglasso) and assessed edge uncertainty and centrality stability with bootstrap and case-dropping procedures. Results. Of the 85 recalculated correlations, 77 matched the published values to two decimal places and eight did not. Eight correlations met the BH-adjusted threshold of .20, reproducing the eight findings the original article marked as corrected; five remained below .05. The one-factor model did not meet conventional CFI and RMSEA cutoffs, χ²(5) = 9.10, p = .105, CFI = .920, TLI = .841, RMSEA = .087, SRMR = .069. The estimated variance explained was .265, but 19.3% of bootstrap resamples produced boundary solutions. An exploratory two-factor model fitted better but was specified after the one-factor results had been examined. The γ = .10 network retained 17 edges, three with bootstrap intervals excluding zero. Centrality stability was limited (strength CS = .206; closeness and betweenness CS = 0). Conclusions. The reanalysis largely reproduced the published correlation pattern, although eight numerical discrepancies remain unexplained, two of them large. The factor analyses give limited support to a single common face-processing dimension. The network findings are exploratory.