In this paper we show that findings of an apparently instable popularity function of U.S. presidents, as reported in the previous literature, are likely the consequence of the common use of linear estimation techniques. Employing Penalized Spline Smoothing in the context of Additive Mixed Models we allow for a-priori unspecified non-linear effects of possible economic determinants of presidential popularity. We find strong evidence for non-linear and negative effects of unemployment, inflation and government consumption on presidential approval and present empirical evidence in favor of the hypothesis of the existence of interaction effects between the economic variables. Additionally we give supporting evidence for the existence of honeymoon and nostalgia effects as well as general decline of support over time.