This paper compares retrospective and prospective analyses of the effect of flip charts on test scores in rural Kenyan schools. Retrospective estimates suggest that flip charts raise test scores by up to 20% of a standard deviation. Yet prospective estimators based on a randomized trial provide no evidence that flip charts increase test scores. One interpretation is that the retrospective results suffered from omitted variable bias. If the direction of this bias were similar in other retrospective analyses of educational inputs in developing countries, the effects of inputs may be more modest than retrospective studies suggest. A difference-in-differences retrospective estimator seems to reduce bias, but it requires additional assumptions and is feasible for only some educational inputs.