Kernel density estimation of the metallicity gradient distribution for the LG sample (with and without merger candidates as solid and dashed blue lines, respectively), compared with the simulated sets of Revaz & Jablonka (2018, red line) and the DCJL plus MARVEL ones (orange line) analysed in Sects. 5.2 and 5.3, respectively. We assumed in all cases a Gaussian kernel with a bandwidth of 0.04 dex , which is comparable to the average error of the metallicity gradients.