A colleague asked me for some details on bottom up attribution. I then asked for the context in which he was speaking, since knowing the context is important.
If we're talking about a "bottom up" manager, who doesn't take allocation into consideration, one could argue that contribution is sufficient. However, there will still be allocation, though perhaps unintended, unless the manager is successful at being market neutral throughout (i.e., regular rebalancing to match the benchmark). Otherwise, "allocation risk" could occur, which could detract from the portfolio's performance. The "Brinson type" models will still be effective to evaluate a bottom up manager, by showing the unintended allocation effects, and should arguably be considered.
If we're speaking in the terms of "nested attribution," for example where we evaluate the portfolio at the asset class (equities, fixed income, cash) and styles (large cap growth, large cap value, etc.), then we're speaking of adding the lower level effects to arrive at the higher level ones. There ARE attribution systems that behave this way, but they're arguably flawed, as I've shown in the past: the results are sometimes simply nonsensical.
Steve Campisi wrote an article for The Journal of Performance Measurement ("Balanced Portfolio Attribution," Winter 2008/2009) where he provides an interesting way to handle nested attribution.
A colleague and I will discuss this topic at a webinar in early 2013; details to follow. With the absence of agreed upon "best practices," we are bound to have various approaches, some of which don't quite work as well as they should. Steve's does, and it will be presented during this webinar.
As an aside, I think there's a fundamental belief that some hold, that summing up from the bottom, be it for returns or attribution effects, yield a more precise number; I generally disagree. I'll try to take this up in greater depth at another time.