Try to replicate in a more rigorous statistical setting. Do nothing until the result is replicated. I’d be willing to bet a significant amount of money that this result is due to a missed latent variable (or a few) that explains everything without making it seem like gender is causal to the outcome.