Children’s Weighing of Local and Global Effectiveness in Information Search

• In Event: The Dynamics of Children’s Information Search

Abstract

Over the past decade, behavioral science has been challenged by the realization that generalizing insights about a phenomenon across diverse populations or contexts requires diverse data. Once aware that the results obtained from restricted, homogeneous or convenient (e.g., WEIRD) samples in specific situations do not necessarily generalize to other subpopulations or contexts, researchers began calling for more diverse epistemic strategies. In this project we investigate how the underlying intuition about effectiveness and robustness in testing strategies emerges and develops across childhood.
Four- to 8-year-old children are presented with a large global population of blue and yellow items, and tasked with identifying which ones among them are desirable “blickets”. To this end, they are shown a monolingual and a bilingual “blicket detector” that vary in their color-specific sensitivity. Children learn about the detectors on a small subset of items (local population) sampled from the global population: The monolingual detector reveals whether items of the minority color (e.g., yellow) are blickets, but cannot see whether items of the majority color (e.g., blue) are blickets or not, whereas the multilingual detector is informative about both colors, but reveals only some of the items. After familiarization with the detectors on the local population, children are asked to select one to use both locally and later on the global population. Crucially, in the representative condition, color ratios in the local sample match the global population, whereas in the unrepresentative condition, the proportion of yellow and blue items in the local sample is reversed. Therefore, in the non-representative condition, if children take the global population into account, they should select the multilingual detector, the one that responds also to the global majority color and is therefore more useful to detect blickets overall (advantage 2.4:1), even if it performs poorly when initially applied to the local population. However, in the representative condition, children should be at chance, as both detectors reveal similar numbers of items in the global population (1.3:1 advantage for the monolingual detector), even if the monolingual detector is still performing better in the local sample. Children’s detector choice therefore resembles a choice between two research instruments that strike different trade-offs between effectiveness over an immediately available but restricted population on the one hand, and generalizability on the other.
Data collection is ongoing, but our first results (N=42, Mage=78 months, SD=15 months) are in line with our hypotheses, suggesting a strong preference (17 out of 21) for the multilingual detector in the non-representative condition, whereas in the representative condition choices are at chance level (11 out of 21). This suggests that children value generalizability in their information search: they prefer a versatile strategy when it is superior, and this preference is not entirely cancelled even when the restricted sample does represent the global population.
In the planned follow-up, we will test an identical paradigm, but increase the number of colors in the global population, further emphasizing the need for generalizability and controlling for the prevalence of a single color in the global population.

Authors

• Andreas Domberg, Max-Planck-Institut für Evolutionäre Anthropologie
  Presenting Author

• Caren M. Walker, University of California - San Diego
  Non-Presenting Author

• Azzurra Ruggeri, Max-Planck Institute for Human Development
  Non-Presenting Author