What could possibly go wrong? - A heuristic for predicting population health outcomes of interventions

Austin Bradford Hill offers a general heuristic for causal inference in epidemiology, but no general heuristic for prediction is available. This paper seeks to identify a heuristic for predicting the outcome of interventions on population health, informed by the moral context of such interventions. It is suggested that, where available, robust predictions should be preferred, where a robust prediction is one which, according to the best knowledge we are currently able to obtain, could not easily be wrong. To assess whether a prediction is robust, it is suggested that we ask why the predicted outcome will occur, rather than any other outcome. Firstly, if, according to our current knowledge, we cannot identify the likeliest ways that the other outcomes could occur, then the prediction is not robust. And secondly, if, according to our current knowledge, we can identify the likeliest other outcomes but we are unable to say why our predicted outcome will occur rather than these, then, again, our prediction is not robust. Otherwise, it is robust. The inaccurate but memorable short version of the heuristic is, "What could possibly go wrong?"