especially in large datasets.

The most important principle is that text columns consume a LOT more resources than numeric columns. Putting them into dimension tables allows the database to fucus on performing your numeric aggregations. From there, it's short work for it to join in the text columns that make the information meaningful to humans.

In most cases it's the most performant data design. It might not be the simplest approach, but other methods usually prove more problematic because they aren't sufficiently flexible or lack the ability to drill down to the atomic level.