We are concerned about two potential problems of previous studies on the relative income hypothesis. First, previous studies assume that reference groups are geographically defined, because data are readily available for geographical reference groups. We examine non-geographical reference groups for the first time in the literature on the relative income hypothesis by utilizing subjective responses to the questions asking individuals to compare their own living standard with the living standard of others in multiple potential reference groups. The second motivation of our study is that previous estimates of the effect of relative income on health could be biased due to unobserved community characteristics that may be correlated with both health and the typical measures of relative income such as mean income and income inequality within communities. We address this potential bias by making use of township fixed effects. We show by comparing the results with and without township fixed effects that our subjective measures of relative income are unlikely to be correlated with unobserved community characteristics affecting health. Further, we address a potential weakness of our subjective measures of relative income by estimating individual dispositions (pessimism/optimism) and controlling for them in our empirical exercises. Our findings support the relative income hypothesis and reveal that relatives, villagers, and others in the same county would be important reference groups when people compare their living standard with someone else's.