Evidence from cognitive science suggests that people compress multi-dimensional information into hierarchies, yet we lack causal evidence on how these representations shape learning, beliefs, and choice. I study this question experimentally. Participants learn success probabilities of firm–project pairs over multiple rounds and repeatedly choose the pair most likely to succeed. I exogenously vary whether outcomes are grouped by firms or by projects, inducing the grouped dimension as the top level of participants' hierarchy while holding information constant. The representation governs how participants engage with the data and where they direct their attention: they process information conditional on the grouped dimension, forming summaries of the data – conditional averages – that they were never asked to form. I elicit behavior at both ends of the resulting hierarchy: beliefs about the conditional averages at the top and incentivized choices over pairs at the bottom. Participants hold accurate beliefs about the averages along their induced dimension, while compressing information along the other. Optimizing locally within these top-level groups, they thereby translate belief distortions into systematic distortions of pair-level choices. Procedural data confirm that attention tracks the representation. Endogenous experiments demonstrate that individuals are not passive recipients of imposed structure: they adapt their representations to both cognitive costs and environmental diagnosticity.