Bradford Hill Criteria for Causal Inference

We think we’re good at determining causality, but we suck at it One of the great challenges in evaluation is determining whether the results we’re seeing are because of the program we’re evaluating, some other influences out there in the big world, or random chance. At one level, this is an everyday, common sense task. As a species we’ve been making judgments about causation for a million years or so. Unfortunately, though, the way we are wired does not predispose us to logical thinking. We are inclined to be led astray by all sorts of biases and heuristics. Over time, the rocket science for dealing with causation has become more sophisticated – a key example being the experimental study design or randomised controlled trial (RCT). And our evidence base about what works has been enriched as a result. But deep down we’re still biased, heuristical beings and not very good at thinking things through. We’ve become so enthusiastic about experimental designs we’re a little inclined to think they are the only way to determine whether A causes B. Such a rigid view is no much use in the real world, where there are all sorts of ethical, conceptual, practical and economic barriers that mean we can’t always conduct RCTs. Even where technically possible, they are not necessarily the best tool for every job.