ABSTRACT We have applied stellar population synthesis to 500-pc-sized regions in a sample of 102 galaxy discs observed with the MUSE spectrograph. We derived the star formation history and analyse specifically the ‘recent’ ($20\,\rm {Myr}$) and ‘past’ ($570\,\rm {Myr}$) age bins. Using a star formation self-regulator model, we can derive local mass-loading factors, η for specific regions, and find that this factor depends on the local stellar mass surface density, Σ*, in agreement with the predictions form hydrodynamical simulations including supernova feedback. We integrate the local η–Σ* relation using the stellar mass surface density profiles from the Spitzer Survey of Stellar Structure in Galaxies (S4G) to derive global mass-loading factors, ηG, as a function of stellar mass, M*. The ηG–M* relation found is in very good agreement with hydrodynamical cosmological zoom-in galaxy simulations. The method developed here offers a powerful way of testing different implementations of stellar feedback, to check on how realistic are their predictions.