It's a mistake to assume that anything that can be analyzed in terms of smaller pieces is less real than the pieces. Neither is real, or both are.
I think your dichotomy is false. Human simplifications of the world are "real" in so far as they correspond with what they model (or else we'll get into a debate about what "real" means). The higher up in the abstraction stack you go, the bigger things (generally) you can model, but the details vary more. This may be more "real", or less "real", depending on what you're trying to model and how well the conceptual framework fits; e.g. you won't get far with modelling fluid as Newtonian marbles. The point being that it's not a binary proposition, the models are all approximations, and any "cause" you find is only with respect to a story people have made up about the way the world works.