[Uncertainty of mortality estimators and hypothesis testing: the case of Latin America and the Caribbean, 1850-2010]

We propose a simple procedure to address the uncertainty that arises when multiple estimators of adult mortality indicators are available in statistical analyses. We consider situations in which there are alternative estimators for the same population parameter, each one depending on a set of potentially overlapping assumptions, and some or all potentially characterizing the target parameter erroneously. Uncertainty arises because of the varying sensitivity of estimators to assumption violations or lack of information about how estimators have been calculated. The proposed procedure allows researchers to use all of the (plausible) estimators, instead of having to choose only one that, ex ante, is considered "the best or right one". This is achieved by assigning a precision score to each estimator depending on: (i) known errors attributable to violation of the assumptions on which the estimator is based, and (ii) (estimated) probability that the assumptions are violated in one particular case. The ensuing inferences on mortality determinants or trends can now be based on all estimators, leading to more robust and conservative hypotheses tests. Notwithstanding its use for mortality in this article, the methodology can be applied to any type of demographic parameter.