23.04.2024 - Erik Ringen and Sabine Stoll [CANCELLED]

Many researchers in language science and related fields aim to estimate the rate of behaviors, such as vocalizations and signals. The precision of these estimates—and consequently our statistical power—is a function of sample size. However, the sampling effort required to achieve a desired precision is heavily influenced by both: (i) the rarity of the behavior and (ii) the degree of individual differences. To illustrate these points, we utilize a longitudinal child speech corpus (Chintang) and model-based simulations to investigate the impact of both the number of participants and the frequency of recordings on speech statistics. While recruiting additional participants is more efficient in general, we highlight the conditions under which repeat sampling can be a more cost-effective design. We suggest a broadly applicable framework that can assist researchers in some pragmatic—but consequential—aspects of study design.