Using cognitive models to interpret value trade-offs in LLMs