Subjective well-being and the gender composition of the reference group: evidence from a survey experiment