The uncertainty in the shot noise rescaling parameter a versus the number of mocks used to calibrate it. Our covariance matrix has 350 bins, and so we indicate with a vertical dashed line N_mocks = 350. As the |$\mathcal {L}_1$| likelihood (see equation 39) does not require that we invert the mock covariance matrix, it yields a precise estimate of a even with |$\mathcal {O}(100)$| mocks. The |$\mathcal {L}_2$| likelihood (see equation 43) does require that we invert the mock covariance matrix, and so only yields a precise estimate of a for N_mocks well above 350. Typical values for a in our model, calibrated against the QPM mocks, were 1.14, so 1 per cent precision on a can be achieved with a small number of mocks.