Sequential Bayes Factor designs in developmental research: studies on early word learning

Developmental research, like many fields, is plagued by low sample sizes and inconclusive findings. The problem is amplified by the difficulties associated with recruiting infant participants for research as well as the increased variability in infant responses. With sequential testing designs providing a viable alternative to paradigms facing such issues, the current study implemented a Sequential Bayes Factor design on three findings in the developmental literature. In particular, using the framework described by Schönbrödt and colleagues (2017), we examined infants’ sensitivity to mispronunciations of familiar words, their learning of novel word-object associations from crosssituational learning paradigms, and their assumption of mutual exclusivity in assigning novel labels to novel objects. We tested an initial sample of 20 participants in  each study, incrementally increasing sample size by one and computing a Bayes Factor with each additional participant. In one study, we were able to obtain moderate evidence for the alternate hypotheses despite testing less than half the number of participants as in the original study. We did not replicate the findings of the cross-situational learning study. Indeed, the data were five times more likely under the null hypothesis, allowing us to conclude that infants did not recognize the trained word-object associations presented in the task. We discuss these findings in light of the advantages and disadvantages of using a Sequential Bayes Factor design in developmental research while also providing researchers with an account of how we implemented this design across multiple studies.

Data, code and manuscript available at: https://osf.io/kpsy3/