Using non-linear methods, this paper finds that existing estimates of government spending multipliers in expansion and recession may yield biased results by ignoring whether government spending is increasing or decreasing. For industrial countries, the problem originates in the fact that, contrary to one's priors, it is not always the case that government spending is going up in recessions (i.e., acting countercyclically). In almost as many cases, government spending is actually going down (i.e., acting procyclically). Since the economy does not respond symmetrically to government spending increases or decreases, the "true" long-run multiplier for bad times (and government spending going up) turns out to be 2.3 compared to 1.3 if we just distinguish between recession and expansion. In the case of developing countries, the bias results from the fact that the multiplier for recessions and government spending going down (the "when-it-rains-it-pours" phenomenon) is larger than when government spending is going up.