Distraction of attention by novel sounds in children declines fast

New task-irrelevant sounds can distract attention. This study specifies the impact of stimulus novelty and of learning on attention control in three groups of children aged 6–7, 8, and 9–10 years and an adult control group. Participants (N = 179) were instructed to ignore a sound sequence including standard sounds and novel or repeated distractor sounds, while performing a visual categorization task. Distractor sounds impaired performance in children more than in adult controls, demonstrating the long-term development of attention control. Children, but not adults, were more distracted by novel than by repeated sounds, indicating increased sensitivity to novel information. Children, in particular younger children, were highly distracted during the first presentations of novel sounds compared to adults, while no age differences were observed for the last presentations. Results highlight the age-related impact of auditory novel information on attention and characterize the rapid development of attention control mechanisms as a function of age and exposure to irrelevant novel sounds.


Part A -Supplemental Tables and Figures
Note. CI = 95% confidence interval. The dependent variable was RT (ms). Age (reference = 6/7), sound (standard = 0, distractor = 1) and condition (0 = novel, 1 = repeated) were dummy-coded predictors. The intercept reflects the estimated RT in the youngest age group for targets after distractors in the novel condition. Note that all results reported in the main article can be computed from this model by simple recoding (i.e., changing the reference categories). Note. CI = 95% confidence interval. The dependent variable was RT (ms). Age (reference = 6/7), sound (standard = 0, distractor = 1), block (reference = first block) and randomization (reference = novel condition first) were dummy-coded predictors. For instance, the intercept reflects the estimated RT in the youngest age group for targets after distractors in the first block and for children who were first presented with the novel condition. Note that all results reported in the main article can be computed from this model by simple recoding (i.e., changing the reference categories). Exemplary comments are added in the right column.  Note. The values are differences in distraction effects (i.e., block 1 minus respective block) in the respective combination of group, block and randomization. Positive values indicate higher distraction effects in the first versus the respective block. For instance, the distraction effect in the first block for 6-7-year-olds who started with the novel condition was 84.22 ms higher than in the second block. 95% confidence intervals are given in square brackets. Bold numbers indicate significantly increased distraction effects in the first versus the respective following block (CIs not-overlapping with zero in the respective fields). Please note that these blocks were also highlighted in bold in table 7 for the reader's convenience.

Part B
During the first recording sessions the repeated deviant sound was accidentally presented as novel in the novel condition. This has no implication when the participant starts with the novel condition as this sound was always new when firstly presented. In the following repeated condition, the distractor was never new as it was always presented several times in the training block. When the participant starts with the repeated condition then in the novel condition 1 out of 24 distractor sounds was not new. We therefore removed this single trial in the respective participants from further analyses. An age-matched comparison between samples with and without correction revealed no differences.

Part C
Sample size considerations in the context of linear mixed models are relatively complex because the sample size has to be considered on both levels of the analysis and many diverging recommendations were proposed [e.g., 1,2,3 ]. Although power computations are possible via simulations for basically any mixed model [e.g., 4,5 ], they require a priori knowledge regarding the many parameters of a mixed model (i.e., random effect (co-)variances and effect sizes of the fixed effects) that was not available for our study. We aimed at testing as many participants as possible but at least 30 participants per group. In the following, we present the results of some simulations that we conducted to investigate the sensitivity of our approach to the effects of interest.
In absence of specific a-priori knowledge, we investigated the sensitivity of our research design toward the hypothesized effects using power curve analyses. That is, we approximated the smallest effect size for which we would have reasonable power (i.e., 90%). In order to run a power simulation in a feasible amount of time using the simr package [ 5 ], we made the following simplifications to the model: 1) The residual correlations between adjacent trials were fixed to zero (as these were very small, this should not bias the results severely).
2) We estimated a homoscedastic model because the package simr does only include lme4 models for which this feature is not yet available.
We identified the following parameters as most central to our substantive hypotheses:  Conclusion. The simulation demonstrates that our research design was especially suited to detect differences between the adults and the groups of children for which the effect sizes were well above the minimum effect size with sufficient power. Comparisons between the children groups, however, should be interpreted bearing in mind that the power was lower, implying that more detailed studies of developmental trajectories across childhood probably require samples even larger than the one tested here.
Regarding the block-wise analyses, the simulation implied that the effect sizes for the most important results (larger distraction effects in children in first block compared to adults) were by far larger than the minimum effect size for sufficient power. Conclusions regarding changes in distraction effects between further blocks (where differences were considerably smaller) and, again, regarding comparisons of children groups should made with caution. That is, the absence of significant differences in distraction effects between children groups should not necessarily be interpreted as a stagnating development.

Part D
An anonymous reviewer encouraged us to test an alternative interpretation for our finding of larger distraction effects in children than in adults. Specifically, it was suggested that the results pattern could have arisen simply from a general development towards faster processing.
According to this reasoning, the reduced distraction effects could be the result of a simpler developmental process: With increasing age, processing becomes faster as indicated by reduced overall RTs. Hence, the absolute difference between distractor and standard sound RTs decreases but relative to the standard RT, the distraction effect remains the same. Formally, such a proportional linear model implies equal distraction effects relative to the intercepts (i.e., the average standard RT) across all groups. Such a model can be construed as a restricted version of our full model with interactions in which an equality constraint is placed on the ratios across all groups.
In a next step, we investigated the pattern of RT ratios further. Descriptively, the RT ratio was similar across children groups (all blocks / 1 st block only; 6-7 years: 0.09 / 0.19, 8 years: 0.08 / 0.26, 9-10 years: 0.08 / 0.192) but different between children and adults (0.04 / 0.09). That is, the size of the distraction effects in children was about 9%/20% (all blocks / 1 st block only) of a standard RT while it was only 4%/9%) in adults. This indicates that the larger distraction effects in children indeed exceeded what would be expected based on a proportional model. Pairwise comparisons confirmed significant ratio differences (i.e., changes exceeding a proportional model) between children and adults (see Table D1). Overproportional differences between children groups could not be established. In light of power considerations (cf. Suppl. part C), this result is not surprising since smaller differences would be expected between children groups than between children and adults. We conclude that a proportional linear model is not sufficient to explain differences between children and adults.