You can explore the map online here, check it out: mpib.berlin/vFVqU

With @bolenz.bsky.social, @kevinetiede.bsky.social, Yujia Yang, @stepalminteri.bsky.social, Ralph Hertwig, and @dirkwulff.bsky.social

8/n Freely accessible, pre-formatted version of the manuscript: mpib.berlin/zWaGS
Thank you to my PhD advisor Christin Schulze & co-author Ben Newell for their support in this project. Happy to see it out now 💐!

7/n We argue to more strongly consider the structure of real-world learning environments (where rewards may not be random but rather clumped or autocorrelated) when examining the development of learning and decision-making strategies.

6/n Our work connects to research on the benefits of cognitive immaturity: For instance, perseveration tendencies in early childhood may hold advantages when repetition fosters learning. Similarly, exploring a lot (maybe too much) helps older kids to adapt to changing environments.

5/n Computational modeling results reinforce behavioral analyses: Children's exploratory behavior and adaptivity increase with age, but adults seem better able to balance exploration and exploitation.

4/n By age 6, children increasingly diversify choices, for instance, through probability matching. They explore widely (sometimes more than adults) and are quickly able to adapt to an ecologically plausible statistical task structure.

3/n 3–4-year-olds showed strong perseveration tendencies: They often stick with one option, even if switching would be beneficial. Our findings integrate into previous research showing increased maximization in young kids; however, we suggest that low implementation effort is key.

2/n Children (3–11 years, N = 362) & adults (N = 121) repeatedly chose between two options with different reward probabilities. We tested three between-subjects conditions: two static (50% vs. 50% and 70 vs. 30%) and one ecologically plausible environment (initially 70% vs. 30%).

In sum, our findings integrate into literature showing that probabilistic inferences in childhood may be highly context-dependent, that previous findings may not easily generalize to a new task, and that the learning format is an important aspect to consider in future work.11/n