If you break down your system into larger sub-pieces (with bigger boundaries), you coarse-grain the scale of analysis. Larger boundaries generally mean longer timescales, at the scale of which shorter timescale changes mostly act as noise. I find this paper helpful:
https://royalsocietypublishing.org/doi/10.1098/rsif.2017.0792
You might enjoy this preprint: https://psyarxiv.com/6afs3