Abstract:
Cox et al. (2023) investigated the acoustic features of infant-directed speech. They used Bayesian meta-analyses to investigate five acoustic features with data from 88 studies. In the present robustness reproduction, we first check if the reported results are reproducible based on the data and code provided by the original authors. We then perform robustness analyses by using different measures of model comparison and by investigating the robustness of the Bayesian sampling approach used by the authors. We find that the main results of Cox et al. (2023) can be reproduced. However, some minor coding errors and unclarity in the results hindered us from reproducing all results in the paper, including those regarding model comparison. In our robustness analyses, we find that alternative ways of model comparison also contradict some of the original results. Further, while the original models were complex and required very specific sampling settings to avoid convergence issues, simplifying them did not show a noticeable impact on the results. Taken together, the reproducibility of the core findings of the paper strengthens its main conclusions, although some results regarding the relevance of moderator variables remain unclear.