How much can children see and report about their experience of a brief glance at a natural scene?

Abstract

Recent studies on brief scene perception have revealed that adults discriminate between what they see and do not see in a photograph with varying degrees of confidence. In this study, we attempt to extend previous studies by asking if these perceptual/cognitive abilities are already established in preschool and school-aged children. In Experiment 1 (n = 122) and 2 (n = 205, registered report), using an online experiment, we briefly presented a natural scene (267 ms in Experiment 1 and 133 ms in Experiment 2) to participants and, subsequently, asked them if a small patch was included in the original scene. Experiment 2 was a registered report. We tested various patch locations to probe "how much" the participants can see and report about it with graded levels of confidence. In Experiment 1, discriminative performance was nearly saturated (the area under the receiver operating characteristic curve (AUC)) = 0.9 across age groups) with no effects of ages, but metacognition slightly improved across ages (AUC = 0.74 in 5-6-year-olds to 0.79 in adults). In a critical registered report (Experiment 2), with reduced stimulus duration, we found a developmental effect (AUC = 0.73 in 5-6-year-olds to 0.91 in adults), and, again, metacognitive accuracy was constant across development (AUC = 0.73 in 5-6-year-olds to 0.75 in adults). Additionally, our analysis of semantic congruence between objects and scenes revealed age-related differences in performance. Contrary to our expectation, the size of the image modification strongly affected task performance, uniformly across ages. Overall, we conclude that 5-6-year-olds' perceptual and metacognitive abilities are much better than we expected when they were tested with briefly presented natural scenes, although their performances were generally lower than adults.

Keywords: children; cognitive development; massive report paradigm; metacognition; natural scene perception; visual experience.

Figure 1

Design of the stimuli and procedure of the task. (a) An example set of “original” and “present” patches. The largest component of the critical object is a disc. Therefore, the red patch (the bottom center) is the “original” patch and the remaining eight blue patches are the “present” patch. (b) An example set of “absent” patches, which are generated from a set of natural images (not used the initial images for this experiment). (c) An example where the initial image is a congruent version of itself: the “original” and the “modified” patches are a disc and a pizza, respectively. (d) An example in which the initial image is an incongruent version of itself: the “original” and the “modified” patches are a pizza and a disc, respectively. (e) The procedure of the task: each trial starts with a 500 ms fixation period, immediately followed by the initial target image probe for 267 (Exp. 1) or 133 ms (Exp. 2). The target is masked by five random textures (60 ms per mask). This target-mask is followed by six probe patches that ask whether the patch was included in the target image. Each probe is presented for 267 (Exp. 1) or 133 ms (Exp. 2) (followed by five-mask patterns, each for 60 ms). Thereafter, we present a response screen asking participants to indicate the presence or absence of the patch with two levels of confidence. (f) The response screen: the upper and lower right panels mean “absolutely present” and “maybe present,” respectively (written in Japanese in a way that children can understand). The upper and lower left panels mean “absolutely absent” and “maybe absent,” respectively. As a reminder for participants, we add the words “confident” above and “not confident” below the response options, respectively, and likewise, we add the words “present” and “absent” to the right and left of the response options, respectively. After responding to this screen, a break screen appears, and when the participants click on it, the next probe patch appears. One trial ends when six probes are tested.

Figure 2

Congruence effects of the image pairs and age groups used for Experiment 2. We defined ∆D × C for a given image pair (identified with a number for each dot in this figure) as (the mean D×C for the original probe patch) − (the mean D×C for the modified probe patch), separately for the initial congruent (x-axis) or the initial incongruent (y-axis) condition (a). The mean is taken across all available participants for one image pair. Each panel is for each age group. Exemplar image pairs in Supplementary Fig. S3 are highlighted by red numbers (59 and 70). ΔD×C for the initial congruent and incongruent image is in x-axis and y-axis, respectively, in Fig. 4. ΔΔD×C quantifies the effects of the congruency of the initial image as: ΔΔD×C = (ΔD×C for an initial congruent image) − (ΔD×C for an initial incongruent image). When the ΔΔD×C is 0, the image pair is located on the diagonal dotted lines. When the ΔΔD×Cs are positive or negative, they are located on the lower right or top left of the diagonal dotted lines. We show the ΔΔD×C (x-axis) with age groups (y-axis) (b). Error bars represent the standard error of the mean across images within each group.

Figure 3

Decision × confidence value: D×C. Mean proportion of responses (y-axis) as a function of decision (“present” = 1, “absent” = −1) × confidence (1 or 3) (x-axis), across participants for each age group in separate panels. Values denoted by each color line sum up to 1 within each panel. Error bars represent the standard error of the mean across participants within each group. The stimulus was presented for 267 ms (Exp. 1, a) and 133 ms (Exp. 2, b). The present patches include the original patches.

Figure 4

Type 1 and Type 2 AUC between the present/original patches and absent patches. (a), (b), (c), and (d) the objective Type 1 AUC (a) and (c) and the metacognitive Type 2 AUC (b) and (d) between the original/present versus absent patches in a raincloud plot with each dot representing a participant. The stimulus was presented for 267 ms (Exp. 1, a and b) and 133 ms (Exp. 2, c and d). Densities along the dots are fitted together with a box plot (median, 25th percentile, 75th percentile).

Figure 5

Relationship between Type 1 AUC and Type 2 AUC. We plotted the Type 1 and Type 2 AUC with each age group for Experiment 1 and Experiment 2. The x-axis represents Type 1 AUC, and the y-axis represents Type 2 AUC. The blue hollow shapes indicate the mean values for each age group in Experiment 1, and the red hollow shapes indicate the mean values for each age group in Experiment 2.

Figure 6

Examples of image pairs with different response trends in Experiment 2. To compute the effects of object-gist congruency, we define ΔD×C as (D×C for the original probe patch) − (D×C for the modified probe patch) for a given image pair, separately for the initial image’s congruency. Image pairs (a) (Image ID = 59) and (b) (Image ID = 70) are images whose ΔD×C for the congruent and the incongruent conditions differed between adults and 5–6-year-olds in the pilot experiment (see Supplementary Fig. S3). In the registered report experiment 2, however, the results were not replicated, implying that the age effect on the congruency in the pilot arose by chance. See the main texts and Supplementary Fig. S7 for further details.

Figure 7

Relationship between the subjective feeling of knowledge and ΔD×C in 5–6-year-olds. Scatterplot of knowledge about the critical object in the images [(1) I do not know it much, (2) I do not know it well, (3) I know it a bit, and (4) I know it very well] and ΔD×C (original − modified) in 5–6-year-olds.

Figure 8

The size of the critical object in the image and ΔD×C. Size of the modification matters regardless of age groups. Scatterplot of the size (log-scale) of the critical object in the image in x-axis and ΔD×C (original patch − modified patch) in the congruent (red) and incongruent (blue) conditions in y-axis. The regression lines, correlation coefficients (r) and P-values (P) are depicted in red (the lines and values above) for congruent and in blue (the lines and values below) for the incongruent initial image conditions. We used the data on the size of the critical object from Qianchen et al. 2022) and the 117 images were analyzed.

Figure 9

Consistency with Watanabe and Moriguchi (2023). We plotted the Type 1 AUC (dots in the figures) of 5–6-year-old children and adults from this study (Experiment 1: 267 ms; Experiment 2: 133 ms) onto 5–6-year-old children’s and adults’ performances of the discrimination task for each SOA from Watanabe and Moriguchi (2023). The upper figure is a modified version of Fig. 3a and b from Watanabe and Moriguchi (2023), and the lower figure is a modified version of Fig. 6a and b. The figures suggest the similar trend in both studies.