Background To tackle population health challenges, we must address the fundamental determinants of behaviour and health. Systematic reviews frequently conclude that the available evidence about the effects of population health interventions is too diverse, flawed or inconclusive to support a more general conclusion about what should be done. However, merely increasing the supply of intervention studies is not enough. The pivotal link between research and policy or practice should be the cumulation of insight from multiple studies. In spite of all the developments in quantitative methods for evidence synthesis, however, we struggle to derive meaningful generalisable inferences to guide and support public health action.

Methods We review theoretical, methodological and case study material from a variety of disciplines and propose a more eclectic, flexible and reflexive approach to building and interpreting the evidence.

Results If conventional evidence synthesis can be thought of as analogous to building a wall, then we can increase the supply of bricks (the number of studies), their similarity (statistical commensurability) or the strength of the mortar (the statistical methods for holding them together). However, many public health challenges seem akin to herding sheep in mountainous terrain, where ordinary walls are of limited use and a more flexible way of combining dissimilar stones (pieces of evidence) may be required. This would entail shifting towards generalising the functions of interventions, rather than their effects; towards inference to the best explanation, rather than relying on binary hypothesis-testing; and towards embracing divergent findings, to be resolved by testing theories across a cumulated body of work. We present case studies of mixed-method primary research and evidence synthesis to illustrate ways of doing this in practice.

Conclusion We should look beyond simple notions of ‘interventions’, search for patterns and embrace the mess in evidence synthesis in order to better understand what makes for an effective public health strategy. In this way we might channel a spirit of pragmatic pluralism into making sense of complex sets of evidence, robust enough to support more plausible causal inference to guide action, while accepting and adapting to the reality of the public health landscape rather than wishing it were otherwise. The traditional art of dry stone walling can serve as a metaphor for the more ‘holistic sense-making’ we propose.