Lant Pritchett (Harvard Kennedy School and the Center for Global Development) looks at how the bulk of the new empirical research on estimating the learning impact of various programmes and policies is inadequately theorised to be of immediate use in accelerating learning.