Radio interferometers, such as the Low-Frequency Array (LOFAR) and the future Square Kilometre Array (SKA), are attempting to measure the spherically averaged 21-cm power spectrum from the Epoch of Reionization (EoR). Understanding of the dominant physical processes which influence the power spectrum at each length-scale is therefore crucial for interpreting any future detection. We study a decomposition of the 21-cm power spectrum and quantify the evolution of its constituent terms for a set of numerical and semi-numerical simulations of a volume of (714 Mpc) 3 , focusing on large scales with 𝑘 0.3 Mpc −1 After ∼ 10 per cent of the Universe has been ionized, the 21-cm power spectrum follows the power spectrum of neutral hydrogen fluctuations, which itself beyond a certain scale follows the matter power spectrum. Hence the signal has a two-regime form where the large-scale signal is a biased version of the cosmological density field, and the small-scale power spectrum is determined by the astrophysics of reionization. We construct a bias parameter to investigate which 𝑘-scales of the large-scale 21-cm signal can be utilised as a probe of cosmology. We find that the transition scale between the scale-independent and scale-dependent bias regimes is directly related to the value of the mean free path of ionizing photons (𝜆 MFP ), and is characterised by the empirical formula 𝑘 trans ≈ 2/𝜆 MFP . Furthermore, we show that the numerical implementation of the mean free path effect has a significant impact on the shape of this transition. Most notably, the transition is more gradual if the mean free path effect is implemented as an absorption process rather than as a barrier.