A ratio-based logistic model developed to assess elephant harvest rates, based on a study at Nagarhole Tiger Reserve in India, was recommended as a management tool to control human–elephant conflict through culling. Considering this reserve among others violates an assumption of the logistic model: isolation. Nevertheless, assuming this violation was irrelevant, we re-evaluated the model, with minor modifications, for the neighbouring Mudumalai Tiger Reserve, where we used data from 13 elephant Elephas maximus population surveys to derive bootstrapped sets of population ratios, and mortality records. We generated arrays of harvest regimes and examined which ratio outputs were closest to the bootstrapped ratios. Our results indicated that (1) model outputs corresponded best with the Mudumalai population structure when harvest regimes were extreme and unlikely, (2) there were significant differences in population structure and harvest regimes between Nagarhole and Mudumalai, and (3) only 49% of adult male deaths predicted by model outputs were recorded in official governmental records. The model provides significantly different results among reserves, which invalidates it as a tool to predict change across the entire elephant population. Variability in survey data and inaccuracies in transition probabilities are sufficiently large to warrant caution when using them as a basis for deterministic modelling. Official mortality databases provide a weak means of validation because poaching incidents are poorly recorded. We conclude that the model should be based on validated transition probabilities and encompass the entire regional population.