A fresh look at infant-directed speech preference through an updated meta-analysis

• In Event: Poster Session 11
  In Poster Session: PS 11 Topic - Language, Communication

Abstract

Meta-analyses are a robust source of evidence when evaluating the existence and magnitude of an effect. However, meta-analyses also have shortcomings. For example they are essentially outdated at the time of being published and decisions the meta-analysts had to take are not always transparent (e.g. which effect and paper to include, how to include it, which effect size to compute). To address these concerns, transparent, "living" community-augmented meta-analyses have been proposed (Tsuji, Bergmann & Cristia, 2014), which includes an open repository and allows for updates and corrections.

Here, we update an existing meta-analysis (Dunst et al., 2012). This meta-analysis assessed whether infants show a preference for infant- over adult-directed speech, and which factors might moderate this effect and found a comparatively large overall effect (Cohen’s d = 0.67) that was modulated by infant age and the type of stimulus used among other moderators. This original meta-analysis was based on 34 studies testing 840 children.

Since publication of this meta-analysis, relevant studies have been published, including ManyBabies1 (The ManyBabies Consortium, 2020)., a large-scale replication study which included data from 2,329 infants from 67 laboratories across four continents. This replication study found a substantially lower overall effect size, and the opposite direction of an effect for infant age.

Updating the original meta-analysis allows us to re-evaluate the published literature on preferences for infant-directed speech, and also explore whether moderating factors may explain differences in findings between the published meta-analysis and the ManyBabies1 replication. We explore a subset of the moderators discussed in Dunst et al. (2012) and focus here on participant age and the type of speech stimulus used (naturalistic, simulated, synthesized, or filtered IDS).

The process was eye-opening: Recoding the previous meta-analysis led to the modification of a number of records, including changes affecting the effect sizes contributing to the original meta-analytic estimate. The validation process is ongoing, and these modifications were necessary for a variety of reasons, including lack of access to data, re-visiting decisions on effect sizes included or excluded in the meta-analysis, and recalculations of previously reported effect sizes. In augmenting the original meta-analysis, we screened 1481 papers published since 1982 and identified 29 as potential papers to be added, in addition to a first wave of updates in 2015 that comprised adding 7 papers. The coding process is ongoing, and we have added 30 experiments to the meta-analysis at the time of writing.

Preliminary analyses with a random effects meta-analytic intercept model without moderators conducted with the R package metafor reveal an overall effect size of Cohen's d = 0.49, 95% CI [0.36, 0.62] (compared with 0.67 reported by Dunst et al., 2012). We further investigated the effect of age and speech, finding no significant moderating effect for either (as opposed to Dunst et al., 2012; see also Figure 1 and 2, respectively).

This project can serve as a model for the process of updating and sharing meta-analyses, in line with recent proposals for community-augmented meta-analyses and shows that new insights can be gained this way.

Authors

• Lucille M Anderson, Brigham Young University
  Presenting Author

• Heewon Hwang, Brigham Young University
  Non-Presenting Author

• Sarah Kamhout, Brigham Young University
  Non-Presenting Author

• Stephen Gilliat, Brigham Young University
  Non-Presenting Author

• Rebecca A. Lundwall, Brigham Young University
  Non-Presenting Author

• Alexis K Black, University of British Columbia
  Non-Presenting Author

• Natalia Kartushina, University of Oslo, Norway
  Non-Presenting Author

• Jessica Elizabeth Kosie, Princeton University
  Non-Presenting Author

• Sin Mei Angeline Tsui, Stanford University
  Non-Presenting Author

• Martin Zettersten, Princeton University
  Non-Presenting Author

• Christina Bergmann, Max Planck Institute for Psycholinguistics
  Non-Presenting Author