We challenge a commonly used assumption in the literature on social preferences and show that this assumption leads to significantly biased estimates of the social preference parameter. Using Monte Carlo simulations, we demonstrate that the literature's common restrictions on the curvature of the decision-makers utility function can dramatically bias the altruism parameter. We show that this is particularly problematic when comparing altruism between groups with well-documented differences in risk aversion or diminishing marginal utility, i.e., men versus women, giving motivated by pure versus warm glow motives, and wealthy versus poor.