On s-convexity and risk aversion