There is currently a debate between Daron Acemoglu and Simon Johnson and David Bloom, David Canning and Güther Fink on whether health improvements lead to higher growth. Acemoglu and Johnson claim it does not, and overturn previous results on the basis that they address critical endogeneity issues. The instrument they propose is exogenous improvements in health technology and how they impact national health.
Bloom, Canning and Fink acknowledge this but think that timing is crucial. Health improvement take time to translate into higher incomes. For example, health in utero in only reflected in physical and cognitive development many years later. Concretely, this means that countries that already benefitted from good health subsequently did not enjoy improvements in health conditions, yet could show growth from previous improvements. Cross-country regressions with appropriate lags show this.
This debate makes me think again that development is in dire needs of proper modelling. All its debates are about instrumental variable choices or specifications of linear reduced form "models" with frighteningly small samples One can debate forever on such reduced forms if one does not have a strutural model to work from. Lay down a theory, derive from it testable implications, which may not be linear by the way, and then argue. This is the scientific way.