Quasi-likelihood estimation in the presence of noise. Synthetic CFSE datasets were generated with branching process Model 1, in which probabilities of division and death change after the first division; parameter values were γ0 = 0.1, γ1+ = 0.6, α0 = 0.05, α1+ = 0.2, starting with 106 cells. QL estimation was used to identify all four best-fit parameter values from a single timepoint – the counts in generations 0–6 observed after 6 timesteps – as increasing levels of Gaussian noise were added either to the counts in each generation (open circles) or the total cell counts, keeping the proportions of cells in each generation constant (filled circles). Noise level σ indicates that the cell counts (or total numbers) were multiplied by a factor (1 + ε) where ε is a random number drawn from N (0, σ2)). We show the mean and standard deviation of 100 simulations. Dotted horizontal lines indicated the true values of the parameters.