boundary=statistical is for formally designated statistical boundaries. We use them for things like census-designated places and census county divisions – placeholders to fill gaps in the administrative area hierarchy. On the other hand, an urban area is an algorithmic measurement (plus a modicum of editorial judgment).
In principle, we could import these boundaries, update the geometries annually, and reimport them every decade when the Census Bureau discards the old ones and recalculates new ones from scratch.[1] This would be very convenient for analytical purposes, helping people avoid the pitfalls of aggregating demographic data by administrative area or postal code. However, the formally described edges of an urban area are extremely complex, since they correspond to census blocks and the algorithm doesn’t compact the boundary very aggressively. It’s practically raster data.
I only proposed using the urban areas or something similar to determine more realistic population figures, from which we can derive less arbitrary place=* classifications. Even though we wouldn’t import the geometries, they would allow us to objectively identify places that are suburbs[2] within an urban agglomeration. Some cartographers have opted for a more holistic approach, but although the results are more reliable and generalizable globally, I find the process to be too opaque and elaborate to explain to a mapper, let alone a map user.