Female participation in the Indian innovation ecosystem: Stylized facts from patent data

India received 8,954 patent applications in 1998 and 110,375 in 2024. What nobody could say is who had been doing the inventing. The Indian Patent Office did not record inventor gender until 2023, and that gap kept the world's fastest-growing patent system out of three decades of international research on gender and innovation.

Neha Raghuwanshi and Madan Dhanora, both at Maulana Azad National Institute of Technology in Bhopal, reconstructed the missing variable from names.

Building the dataset

Their source was PatSeer, from which they pulled every patent granted by the Indian Patent Office between 1995 and 2023: 329,034 patents naming 1,006,270 inventors. Application numbers, years, inventor names, IPC classifications and assignee details were cleaned in Python and cross-checked against Indian Patent Office records, with the records that resisted automated standardization corrected by hand.

Gender came from two tools that were chosen because they fail differently. WIPO's World Gender Name Dictionary is deterministic: a name is in it carrying one gender, or it is not there at all. The deconflicted WGND 2.0 file holds 3,491,141 names, having thrown out every name that disagrees with itself across countries, sources or languages, and India is its best-covered country at roughly 2.9 million names. Genderize.io supplied the probabilistic half, returning a classification and a confidence score for names the dictionary had discarded or never held.

Rather than choose between them, the authors ran the analysis twice. Every headline figure has a Genderize.io counterpart in the appendix, and the API classifications were re-cut at confidence thresholds of 0.5, 0.6, 0.7 and 0.9, with names falling below each threshold dropped from that pass. The patterns held at all four.

Diversity was then measured two ways per inventor team. The Blau index runs from 0 for a single-gender team to 0.5 for an even split; the Shannon index runs from 0 to 0.693. Both were reported alongside a simpler figure, the share of a patent's inventors who were women.

An uneven climb

The trend rises, but not smoothly. Diversity sat low through 1995 to 2001, dipped around 2002, and turned upward after 2006, when India came into compliance with the TRIPS agreement. The sustained rise arrives only after 2016, the year of both the National IPR Policy and the Startup India scheme, when the two indices first average above 0.10. Against theoretical maxima of 0.5 and 0.693, that is a long way from balance.

Residency splits the picture. Patents assigned within India climbed from a female inventor share of about 0.10 in the early 2000s to 0.25 by 2023, with the acceleration after 2015. Non-resident patents, filed in India by foreign assignees, stayed consistently below that line for the whole period. The growth in women's participation is domestic.

Where the women are patenting

Female participation concentrates in chemistry and the life sciences: pharmaceuticals, biotechnology, organic fine chemistry, macromolecular chemistry and food chemistry. Mechanical engineering runs the other way, close to uniformly male across handling, machine tools, engines, and transport. This is not a small corner of the data. Chemistry accounts for 109,926 of the granted patents and mechanical engineering for 73,684, so the two blocks the genders sort into are comparable in size.

Institutions sort the same way. Firms file the most patents across electrical, mechanical and chemical fields and show the lowest diversity of any assignee type. Government bodies and research institutes patent far less but come closer to balance, which the authors read against India's women-in-science schemes, among them KIRAN, INSPIRE and the Women Scientist Scheme.

What the authors flag

Raghuwanshi and Dhanora are direct about the limits. Gender here is inferred from names rather than self-identified, and they note that some misclassification survives even a hybrid approach. Patents also measure inventive output rather than participation in research, so contributions that never reach a filing go uncounted. They present the work as a descriptive benchmark, with causal questions left to later research.