Face Recognition and BFRT-c RT Accuracy by Age

RT AND FACE RECOGNITION ABILITY Supplementary Materials Results Associations between Face Recognition and BFRT-c RT/Accuracy when Controlling for Age When including age as a predictor of face recognition scores in a multiple regression in addition to accuracy and RT of BFRT-c, only accuracy was a significant predictor of CFMT (βacc = .43, t = 3.66, p < .001; βRT = -.11, t = -.94, p = .35; βage = -.21, t = -1.75, p = .086; F (3, 57) = 8.04; R2 = .26) and FFMT (βacc = .33, t = 2.82, p = .0065; βRT = .034, t = .30, p = .77; βage = -.31, t = -2.56, p = .013; F (3, 58) = 7.03; R2 = .23), while accuracy and age were significant predictors of the composite score (βacc = .45, t = 4.165, p < .001; βRT = -.047, t = -.45, p = .66; βage = -.31, t = -2.75, p = .0079; F (3, 58) = 11.92; R2 = .35). The binary logistic regression models predicting DP vs. control group membership with age included as a predictor showed that BFRT-c accuracy, but not RT or age, significantly predicted DP diagnosis (accuracy: p < .001, RT: p = .032, age: p = .066; pR2 = .37, AIC = 88.92). Overall, controlling for age did not affect the finding that BFRT-c accuracy, but not RT predicted face recognition ability and DP diagnosis. Examining BFRT-c Raw RT and RT with Outlier Removed Since previous face recognition studies have only used the raw RT of the computerized Benton test, we performed the same set of analysis with the untransformed raw RT data. In the main body of the paper, we took the natural log of Benton RT to better approximate a normal distribution. Results in the control group showed that while BFRT-c accuracy significantly predicted each of the dependent measure (all ps < .001 as reported in the main body of the results section), raw Benton RT was not significantly associated with RT AND FACE RECOGNITION ABILITY 36 CFMT (r = -.03, p = .825), FFMT (r = .05, p = .655), or the composite CFMT-FFMT scores (r = .02, p = .863). Neither did the raw RT explain additive variance beyond accuracy for CFMT (p = .580, ΔR2 < .001), FFMT (p = .907, ΔR2 = .004), or the composite score (p = .761, ΔR2 = .001). To examine whether raw RT can differentiate DPs from the controls, we examined whether raw BFRT-c RT predicted DP/control group membership and examined whether it explained additive variance beyond accuracy in logistic regressions. Results showed that while accuracy was a significant predictor (p < .001 as reported in the main body of the paper) for DP diagnosis, raw BFRT response time was not a significant predictor (p = .505, pR2 = .003). Neither did RT explain unique variance beyond BFRT accuracy (p = .194, ΔpR2 = .011). In the current Benton RT data, one data point from the control group exceeded 3 standard deviations above the mean of the entire sample (i.e., combined DPs and the controls, n=98). We here provide the same set of analysis reported in the main body of the manuscript with the outlier removed RT data (natural log applied to meet normality). In the main body of the paper, we did not exclude any data based on the RT, because 1) RT was what we were aiming to validate, so exclusion criteria should not be based on this measure (exclusion is best based on the robustly validated measures); 2) excluding the outlier from only one group can artificially push the result towards one direction, and we did not want to bias our results; and 3) we examined the trials within that participant and verified that the subject was fully focused on the test, so this should be a valid data point. When examining BFRT-c RT with the outlier excluded, we found that while accuracy was significantly associated with each dependent measure (all ps < .001 as reported in the main body of the results section), Benton RT was only significantly associated with CFMT (r = -.26, p = .049), but not FFMT (r = -.05, p = .729), or the composite score (r = -.16, p = .215). Neither did the RT explain additive variance beyond accuracy for CFMT (p = .066, ΔR2 < .045), FFMT (p = .742, ΔR2 = .002), RT AND FACE RECOGNITION ABILITY 37 or the composite score (p = .166, ΔR2 = .024). We next examined whether BFRT-c RT with the outlier excluded can differentiate DPs from the controls. Results showed that while accuracy was a significant predictor (p < .001 as reported in the main body of the paper) of DP vs. control group membership, BFRT-c RT was not a significant predictor (p = .077, pR2 = .032). Neither did RT explain unique variance beyond BFRT accuracy (p = .088, ΔpR2 = .024). Examining Combinations of BFRT-c Accuracy and RT: Inverse Efficiency and Balanced Integration Scores When calculating the inverse efficiency score (IES, log transformed RT/Accuracy), we found that IES did significantly predict DP vs. control group membership (p < .001, pR2 = .302, AIC = 121.75, see Supplementary Figure S1), but adding accuracy scores in the model resulted in accuracy being the only significant predictor (accuracy: p = .0310, IES: p = .0984; pR2 = .343, AIC = 90.623; ΔpR2 = .02). Further, as can be seen in Supplementary Figure S2, IES was significantly correlated with FFMT (r = -.33, p = .009), CFMT (r = -.47, p < .001), and the composite of these measures (r = -.47, p < .001). However, when including both BFRT-c IES and accuracy as predictors of face recognition scores in a multiple regression, only accuracy was a significant predictor of CFMT (βacc = .32, t = 1.895, p = .063; βIES = -.22, t = -1.30, p = .198; F (2, 59) = 10.39, R2acc+IES = .26, R2acc+IES, adjusted = .24; ΔR2 = .02), FFMT (βacc = .41, t = 2.34, p = .0225; βIES = -.03, t = -.18, p = .861; F (2, 59) = 6.67, R2acc+IES = .18, R2acc+IES, adjusted = .16; ΔR2 = .00), and the composite score (βacc = .43, t = 2.66, p = .010; βIES = -.16, t = -.971, p = .335; F (2, 59) = 12.85, R2acc+IES = .30, R2acc+IES, adjusted = .28; ΔR2 = .01). As shown by the change in R2, compared to the model with only BFRT-c accuracy as a predictor, including BFRT-c RT in the model did not improve the model by a noticeable amount, if at all. RT AND FACE RECOGNITION ABILITY 38 Supplementary Figure S1. Z-scores (computed from the mean and standard deviation of our control group) for developmental prosopagnosic and control groups on the BFRT-c accuracy, BFRT-c RT, BFRT-c IES, and BFRT-c BIS. The error bars represent the standard error of the mean. *** indicates that the difference is significant at p<.001 RT AND FACE RECOGNITION ABILITY 39 Supplementary Figure S2. Associations between face recognition measures and BFRT-c accuracy, IES, and BIS within the control group. A) Scatterplots for BFRT-c score/accuracy against FFMT/CFMT/Composite; B) Scatterplots for BFRT IES against FFMT/CFMT/Composite; C) Scatterplots for BFRT BIS against FFMT/CFMT/Composite. We next examined the balanced integration score (BIS), calculating RT and accuracy z scores for all subjects using the Mean and SD of the control group. As shown in Supplementary Figure S1 and S2, the BIS (p < .001; pR2 = .276, AIC = 97.254) on its own did predict DP vs. control group membership in logistic regressions and was significantly correlated with FFMT (r = .30, p = .018), CFMT (r = .44, p < .001), and the composite of these measures (r = .44, p < .001). But when accuracy was modeled along with BIS, only RT AND FACE RECOGNITION ABILITY 40 accuracy was significant in predicting DP group membership (accuracy: p = .00737, BIS: p = .0987; pR2 = .343, AIC = 90.689; ΔpR2 = .02). Similarly, only accuracy was a significant predictor of CFMT (βacc = .36, t = 2.274, p = .0267, βBIS = .19, t = 1.167, p = .2481; F (2, 59) = 10.17, R2acc+BIS = .26, R2acc+BIS, adjusted = .23; ΔR2 = .02), FFMT (βacc = .42, t = 2.601, p = .0117; βBIS = .015, t = .090, p = .928; F (2, 59) = 6.651, R2acc+BIS = .18, R2acc+BIS, adjusted = .16; ΔR2 = .00), and the composite score (βacc = .46, t = 3.056, p = .00336; βBIS = .12, t = .834, p = .408; F (2, 59) = 12.67, R2acc+BIS = .30, R2acc+BIS, adjusted = .28; ΔR2 = .01). Together, these results suggest that, compared to accuracy alone, integrating RT and accuracy using either IES or BIS does not explain additional variance in DP vs. control diagnosis, nor does it explain additional variance in predicting face recognition scores.