Scholarly collaboration now drives the frontier of knowledge. Given increasing professional advantages of collaboration such as those related to innovation, it is important to understand gender gaps in women's participation in collaborative teams.
Using a rich database of publications in JSTOR corpus, we explore gender-based homophily -- the tendency for researchers to co-author papers with individuals of the same gender -- in scholarly collaborations. We first use citations to obtain a hierarchical partition of the JSTOR scientific landscape. Given the hierarchical partition, we propose a framework for studying homophily in collaborations across the scholarly landscape and at all levels of science, from terminal fields that represent smallest intellectual communities to major fields. We consider a homophily statistic that is a generalization of the Wright's coefficient of inbreeding, appropriate in the two-author case, to the multi-author case. We develop a non-parametric approach for testing whether the observed homophily value is significantly different from the value that is expected under gender-blind choice of co-authors in the given scientific field, taking into account the field's proportions of male and female authorships, the distribution of the number of authors across multi-author papers, connectivity to other fields and the existing subfield structure. Our conservative approach to carrying out this significance testing finds a substantial deficit of female-male collaborative papers in most major scientific fields and across many terminal fields.