Prior research from the U.S. and abroad reveals a sizable lesbian earnings advantage over otherwise-similar heterosexual women. Using data from the 2000 U.S. Census and 2005-2010 American Community Surveys, we estimate traditional earnings equations and find robust evidence of a lesbian premium, corroborating the findings of previous studies. Using within-sample maternity incidence as an estimate of employers' forward-looking expectations, we then examine whether differences in the perceived likelihood of an employee requiring maternity leave, here-labelled 'maternity risk', contribute to the lesbian pay gap. Results from a direct assessment suggest that maternity risk adversely affects income, and that accounting for near-term differences in maternity risk reduces the lesbian premium by approximately ten to fifteen percent. Further analyses, using proxy variables for differential maternity risk, yield similar results. As such, the persistent finding of a lesbian earnings advantage in previous studies can be attributed, at least in part, to employers' aversion to maternity risk and its associated costs.

These findings are also of critical importance to the general labour-market discrimination literature. Given the adverse earnings effect of maternity risk, our analysis suggests that estimates of the well-established gender earnings disparity are likely to be considerably smaller when incorporating maternity risk into the analysis. Absent the ability to adequately control for maternity risk, strict attention should be paid to potential upward bias in estimated earnings differentials. Moreover, policymakers should consider the broader implications of maternity-leave policy on the labour-market outcomes of females. In this respect, maternity-leave policy may influence the hiring and promotion decisions of employers, thereby indirectly affecting sexual-orientation and gender equality in the labour market. However, further research in this area is still required, given the limitations inherent in the direct and indirect analyses.