A summary of uncanniness ratings across conditions is depicted in
Figure 5. Orientation, familiarity, and distortion were used as fixed effects to predict uncanniness, and face identity and participants as random effects. As the assumption of homoscedasticity was not met, a robust estimation of the linear mixed model was calculated. Distortion significantly predicted uncanniness (
t(6308) = 32.483,
p < 0.001), but neither familiarity (
t(6317) = 0.257,
p = 0.798) nor orientation (
t(19) = −1.073,
p = 0.297). Interaction effects between distortion and familiarity (
t(6308) = −6.204,
p < 0.001), distortion and orientation (
t(6308) = −11.573,
p < 0.001), and familiarity and orientation (
t(6321) = 2.644,
p = 0.008) were found, as well as an interaction with all factors combined (
t(6308) = 2.588,
p = 0.010). The model's regression coefficient was
R²corr = 0.458. The test statistics for all terms are summarized in
Table A1. Data are summarized in
Figure 5.
Post hoc Tukey tests with Bonferroni corrections were performed to test differences between face condition groups (familiar upright versus novel upright, familiar upright versus familiar inverted, novel upright versus novel inverted, and familiar inverted versus novel inverted) for each distortion level. At distortion level 0, novel upright faces were more uncanny than familiar upright faces (t(65) = 4.657, padj < 0.001), familiar inverted faces were more uncanny than familiar upright faces (t(65) = 6.324, padj < 0.001), and novel inverted faces more uncanny than familiar inverted faces (t(65) = 2.748, padj = 0.031) and novel upright faces (t(65) = 5.103, padj = < 0.001). Thus, both novelty and inverted orientation increased uncanniness of base faces. At distortion level 1, no differences between condition groups were significant. At distortion level 2, all differences were nonsignificant except for familiar inverted faces, which were less uncanny than familiar upright faces (t(65) = −4.482), padj = < 0.001). Thus, at distortion level 2, upright orientation increased uncanniness ratings for familiar faces. Familiar inverted faces remain less uncanny than familiar upright faces at distortion level 3 (t(65) = −8.47, padj < 0.001), and novel inverted faces become less uncanny than familiar inverted faces (t(65) = −4.331, padj < 0.001). Thus, at this stage, inversion generally reduces the uncanniness of distorted faces. Finally, at distortion level 4, familiar inverted faces again remain less uncanny than familiar upright (t(65) = −8.072, padj < 0.001), and novel inverted faces less uncanny than familiar inverted faces (t(65) = −4.727, padj < 0.001). In addition, novel upright faces are less uncanny than normal familiar faces (t(65) = −2.963, padj = 0.023), suggesting that both upright orientation and familiarity increase the uncanniness of distorted faces. These results show that uncanniness increases the strongest across distortion levels when faces are upright (versus inverted) and familiar (versus novel). Thus, hypothesis 1 was supported.