This paper examines issues related to the estimation of the government spending multiplier (GSM) in a DSGE context. We stress a source of bias in the GSM arising from the combination of endogenous government expenditures and Edgeworth complementarity between private consumption and government expenditures. Due to cross-equation restrictions, omitting the endogenous component of government policy at the estimation stage would lead an econometrician to underestimate the degree of Edgeworth complementarity and, consequently, the long-run GSM. An estimated version of our model with US postwar data shows that this bias matters quantitatively. The results are robust to a number of perturbations.