Why this should depend on convexity at all? The study’s premise is marginal benefit, i.e., estimated risk difference is either minimal or perhaps too poorly known to call it out as marginal. If it was known to be positive or negative, action should be obvious. I am not defending this study because authors seem to rely heavily on survey where participants satisfaction measured only after 6 months. Typically, people tend to comfort themselves in short term for making big decisions such as divorce or job change even though in longer run they may higher accumulated regret. Economists need to develop good models instead of just keep doing surveys.