Both Experiments 1 and 2 were unable to demonstrate advantage of exposure to three expressions of a face relative to one expression in between-expression transfer. In this experiment, we tested the possibility that the potential advantage of three-expression training is more detectable when the result is compared to a baseline using a neutral expression.
Cross-experiment comparison
Were the results in Experiment 3 fundamentally different from the other two experiments? To understand whether the pattern of the d’ results differed among the three experiments, we performed a three-way repeated-measures ANOVA, using experiment as an additional between-participant factor. If the better between-expression transfer were unique to Experiment 3, we would expect a significant three-way interaction among experiment, exposure, and expression change, where only Experiment 3 should display better transfer to a new expression following multiple-expression training.
The ANOVA showed no main effects of experiment, F (2, 238) = 0.01, partial η
2 = 0.00, p = 0.99, or level of exposure, F (1, 238) = 0.16, partial η
2 ~ = 0.00, p = 0.68. Recognition was impaired when learned faces were tested in a different expression relative to the same expression, F (1, 238) = 167.38, partial η
2 = 0.41, p < 0.001. There was no three-way interaction, F (2, 238) = 2.07, partial η
2 = 0.02, p = 0.13, or two-way interaction between exposure and experiment, F (2, 238) = 0.02, partial η
2 ~ = 0.00, p = 0.98. However, the results were qualified by two-way interactions between exposure and expression change, F (1, 238) = 10.43, partial η
2 = 0.04, p < 0.01, and between experiment and expression change, F (1, 238) = 3.09, partial η
2 = 0.03, p < 0.05.
The interaction between exposure and test expression was due to a better transfer of training to a new expression following exposure to three expressions relative to a single expression. When trained faces were tested in the same expression, the results for three-expression (M = 2.78, SD = 1.19) and single-expression (M = 3.01, SD = 1.10) were comparable, t (242) = −1.59, p = 0.11. However, when studied faces were tested in a new expression, the result for three-expression (M = 1.94, SD = 1.41) was significantly better than single-expression (M = 1.61, SD = 1.35), t (242) = 2.20, p < 0.03.
To identify the source of interaction between experiment and expression change, we first conducted simple main effects analyses for same and different expression separately. The results showed that when faces were tested with the same expression as the trained, there was no difference among the three experiments, F (2, 241) = 1.15, partial η
2 ~ = 0.00, p = 0.32. There was also no difference among the experiments when the trained faces were tested in a different expression, F (2, 241) = 0.99, partial η
2 = 0.01, p = 0.37. As our analysis for each individual experiment already indicated, there was a strong effect of expression change in all experiments. However, effects of this varied across the three experiments. We calculated the effect of expression change in each experiment by subtracting the d′ for different expression from the d′ for same expression. The effects for Experiments 1 through 3 were 1.15 (SD = 1.52), 1.37 (SD = 1.17), and 0.84 (SD = 1.44), respectively. An ANOVA showed a significant difference among these, F (2, 241) = 3.01, partial η
2 = 0.01, p = 0.05. A Tukey pairwise comparison of means showed that the change of expression had significantly less impact in Experiment 3 compared to Experiment 2. Other pairwise comparisons did not find significant difference. The interaction between experiment and expression change was, therefore, due to relatively smaller effect of expression change in Experiment 3.
The cross-experiment analysis showed that multiple-expression training in all experiments created similar effects relative to single-expression training. This was evident in the two-way interaction between exposure and expression change. In other words, the analysis showed no evidence that only the baseline condition in Experiment 3 of this study resulted in the interaction. The analysis merely showed that the effect of expression change was smaller in Experiment 3 than the other two experiments. The lack of three-way interaction or two-way interaction between level of exposure and experiments suggests that both multiple-expression training and single-expression training conditions produced comparable recognition performance across the three experiments.