Figure 2. A: Mean percentage of correct responses as a function of object configuration (grouped, partially grouped, and ungrouped) for the color and orientation changes (solid and dashed lines, respectively). B: Mean percentage of correct responses as a function of change type (color and orientation) in the partially grouped triangle condition. Accuracies in B are plotted separately for trials on which the probe was one of the three pacmen that gave rise to the illusory triangle (inside), or, respectively, on which the probe was one of the three non-grouped pacmen (outside). Error bars denote the 95% (within-subject) confidence interval.
A subsequent analysis examined whether change detection performance was influenced by the probe location in partially grouped displays (with triangle groupings). Figure 2B presents the percentage of correct responses for color and orientation changes, separately for trials on which the probe was presented at one of the three pacman locations that formed the illusory triangle (inside) and, respectively, trials on which the probe appeared at one of the three other, “non-grouped” pacmen (outside). A corresponding two-way repeated-measures ANOVA of the accuracies, with the factors Change Type (color, orientation) and Probe Location (inside, outside), revealed both main effects to be significant: Change Type, F (1, 23) = 15.96,p < .001, ηp2= .41; and Probe Location, F (1, 23) = 10.09, p = .004,ηp2 = .31. Accuracies were higher for color changes (68%) than for orientation changes (64%), mirroring the analysis described above. In addition, the accuracies were increased when the probe was presented inside the partially grouped triangle (68%) as compared to an outside location (64%). The Change-Type × Probe-Location interaction was not significant, F(1, 23) = 0.55, p = .47,ηp2 = .02,BF10 = 0.34. Thus, the behavioral results directly replicate our previous findings (Chen et al., 2021a) and show an object-benefit for both grouping-relevant and -irrelevant features.
Moreover, a final analysis was performed which computed an overall estimate of VWM capacity K (Cowan, 2001) in order to determine how the change in grouping strength across our stimulus configurations affected the capacity estimate. Each individual’s memory capacity was computed using Cowan’s formula: K = (H – FA) × N, where K is the memory capacity, H is the observed hit rate, FA the false alarm rate and N the number of (pacman) items presented. The resulting capacities for orientation and color change trials were then combined to yield an ”overall” capacity estimate for a given configuration. Next, a one-way repeated-measures ANOVA was performed on the mean K estimates, which (again) revealed a reliable effect of Object Configuration, F (2, 46) = 70.97, p < .001,ηp2 = .76. The K estimates were largest for the grouped configuration (5.5), intermediate for partially grouped configuration (3.8), and smallest for the ungrouped configuration (3.0; all p’ s < .001,dz s > 0.65, for the pairwise comparisons between configurations). This shows that grouping can lead to a substantial enhancement of the overall VWM capacity beyond the usual capacity estimates of around 3-4 items (Luck & Vogel, 1997).
ERP data. The corresponding ERP waves at parieto-occipital electrodes (averaged across electrodes PO3, PO4, PO7, PO8, O1, and O2) for the different object configurations are plotted inFigure 3A . Visual inspection of the ERP waves suggests that major differences between the different object configurations occurred in the PPC, N1pc, N2pc, and CDA components. For analysis, we examined these amplitude variations across conditions separately for each component in a series of one-way repeated-measures ANOVAs with the within-subject factor Object Configuration (ungrouped, partially grouped, and grouped; see also Figure 3B ).
The ANOVA of the mean PPC amplitudes revealed the Object-Configuration effect to be significant, F (1.43, 32.78) = 9.56, p = .002, ηp2 = .29: there was a graded difference across object configurations, with the positive deflection being largest for the ungrouped (0.89 µV), intermediate for partially grouped (0.72 µV), and smallest for the grouped (0.50 µV) configurations (all p’ s < .008,dz s > 0.53, for the pairwise comparisons between configurations).
The analysis of the N1pc also yielded a significant Configuration effect, F (1.47, 33.83) = 5.08, p = .019,ηp2 = .18, with a larger negativity for the grouped (-0.29 µV) as compared to the ungrouped (0.03 µV, p = .006, dz = 0.55) and partially grouped (-0.11 µV, p = .004, dz = 0.58) configurations, but no reliable difference between ungrouped and partially grouped configurations (p = .12,dz = 0.25, BF10 = 0.73).
For the N2pc, the Configuration effect was again significant, F(2, 46) = 10.07, p < .001,ηp2 = .31, due to more negative-going amplitudes for the grouped (-0.95 µV) as compared to the ungrouped (-0.56 µV, p < .001,dz = 0.74) and partially grouped (-0.61 µV,p = .001, dz = 0.69) configurations, but no significant difference between ungrouped and partially grouped configurations (p = .26, dz = 0.13,BF10 = 0.38).
Finally, the analysis of the CDA amplitudes also yielded an effect of Object Configuration, F (2, 46) = 3.57, p = .036,ηp2 = .13. As depicted inFigure 3B , the mean CDA amplitude was more negative for the grouped (-1.26 µV) as compared to the ungrouped (-1.08 µV, p = .01, dz = .51) and partially grouped (-1.15 µV,p = .046, dz = .36) configuration. There was again no reliable difference between ungrouped and partially grouped configurations (p = .16, dz = .21,BF10 = 0.56).
The result patterns of the PPC, N1pc, N2pc, and CDA thus mirror (at least to a large extent) the pattern of behavioral performance, evidencing an effect of Object Configuration, which was driven particularly by the fully grouped star object. Of note, a graded improvement in VWM performance with an increase in grouping strength (across all three configurations) was already evident at early stages of perceptual processing, namely, in the PPC component.
Moreover, the CDA results essentially mirrored the estimated VWM capacity scores (see above), thus supporting the view that the CDA corresponds to the number of effectively remembered items. In addition, the findings are also compatible with the view that the generation of a global shape (in Kanizsa figures) requires additional mnemonic resources, and this increase in the mnemonic activity may likewise be reflected in the increased negativity of the CDA.
Finally, additional correlational analyses between the individual behavioral performance and the corresponding ERP amplitudes revealed significant negative relationships for the PPC components in the grouped and partially grouped configurations for orientation changes (grouped:r = -0.47, p = .01; partially grouped: r = -0.36,p = .04; see Figure 3C ), that is, the PPC amplitude scaled with behavioral performance for the grouped (and partially grouped) memory configurations. The correlations thus show that larger performance benefits for the (partially) grouped memory configurations were associated with less positive PPC amplitude deflections. No other significant correlations between behavioral performance and ERP components were revealed. Statistical significance of the correlation coefficient was determined by comparing the observed correlations with results derived from 20000 permutations of the two variables, thus excluding the influence from any outliers in the data.