In this paper it is shown that it matters a lot for empirical research whether policy is taken to be exogenously set or to be endogenous. In the model investment depends on policy which depends on economically important fundamentals and is, thus, endogenous. Conditioning on factor accumulation in growth regressions that also include endogenous policy variables may then be problematic. When policy is endogenous the measured effects of policy on growth will generally be biased. Based on the model and OECD data, the signs of the biases for tax variables related to the tax base and for redistribution are derived. Based on these signed biases the paper discusses some empirical results that seem puzzling from a theoretical viewpoint. The paper argues that regressing growth on policy may still yield important information if policy endogeneity is taken account of.