These were some long days – particularly for someone who had to fight through traffic to get there. I’m going to try to wrap up Day three in this post so I can then summarize for some work people tomorrow.
Dani Bassett – Science as a Branched Flow
Profile page. Flow through irregular media – wave dynamics scattering… branches and makes a tree-like pattern. Non-uniform and non-obvious paths
Citation trails like academic bricks (Ahmed, 2017 – likely this) to build careers, to build fields.
Papers with white authors in first authorship overperform by 8% articles by AoC undercited by 17.2%… This trend is getting stronger. (can’t read my hastily jotted notes. They cited a number of different studies showing women were cited less than men.
We can be part of the change we want to see by checking our reference lists – are we citing the same old guys from our grad school days? Are we citing only men?
(see their slide: https://twitter.com/ICSSI2022/status/1534888861477781504)
Rebecca Eisenberg Science of Patents
(I felt like there was a lot of Patent 101 here which I certainly don’t need. I didn’t take many notes. Disclaimer: the one question I tried to ask was sort of scoffed at? So that colors my memory of the session)
Main points:
- Patents unlike publications because inventors get to decide what to patent, if it meets basic requirements (novel, non-obvious, described … )(and yet, not really? don’t we all work in organizations that own our IP? patenting takes money? patenting takes know-how?)
- Difference in who counts as an inventor vs who counts as an author
- Studies in SoS misuse patent information – mistaking terms applied by inventor for those applied by examiner (see this study: https://www.nber.org/papers/w25742) to quote from that:
We argue that there is a fundamental problem in interpreting front page citations as knowledge flows. These citations are not simply a list of earlier patents, academic articles, and other documents which were relevant to the invention. Rather, they derive from a legal “duty of disclosure” requiring inventors to list documents material to the patentability of their claims. Material to patentability means there is no need to cite tools that help build the invention, information used to avoid unpromising paths, basic facts that focus effort, or research suggesting a technological hole or a market need
Joseph Staudt – Specialized Human Capital and STEM Career Outcomes
His page. Best of markets for some fields but worst for others. Is the US producing too many or too few STEM PhDs?
He looked at “quality” of job placements as faculty, postdocs, industry using UMETRICS, ProQuest Dissertations and Theses, IRS W-2 (!), LEHD (about).
Faculty placement – assume “good” position. In some fields like Chem, Geo, lowest fraction of faculty placements.
Post Docs – may indicate weak demand or demand but position requires additional skills after PhD. So look at if there is higher wage growth after postdoc. He found fields with higher % in post docs had lower earnings (d’oh!)
Industry – depends on quality of placement. (his way of measuring this is fancy and huh). Pay + dissertation abstract vs firm portfolio patent classification. He used the Census Business Register to see if the person’s dissertation was in the same general field as the company patented. (so yeah, i mean like if you have a Boeing or something or gosh, my place of work. I have a good placement but ?)
1 s.d. higher relevance 5% higher earnings.
In the end, it depends on field.
Julia Barnett Intersectional Inequalities in the Impact of Online Visibility on Citations
Her page. Looked at Altmetric % researchers listed by field
Ratio of women on the team (pub 2013-2019). Ethnicity by Blau’s index.
Women are less visible online and work visible online is more cited. Intersectional Penalty
Mengyi Sun Effects of Peer Review Format on Bias
Profile. (preprint but looks like it also made it to JASIST) Does blinding reviews reduce prestige bias, increase accuracy and reliability? A natural experiment of a large computer science conference that changed from single-blinded review to double-blinded in 2018. Prestigious scholars got lower scores but did not generally change acceptance status (threshold was lower). Papers rejected under single-blind (and then appearing elsewhere I guess) were more highly cited than those rejected under double-blind so better acceptance decisions.
Sarah Rajtmajer Synthetic Market for Estimating Confidence in Published Work
Part of larger DARPA SCORE project described by Nosek earlier. Paper > features > trader bots . Trained with output from major replication projects. Some papers are too dissimilar from training so no bots participated (they know what they don’t know). Next, multiple claims and hybrid markets.
Dakota Murray – Understanding Disagreement in Science
publications. Looking to quantify disagreement in science. Retrieved citation context for articles. Looked for cues of disagreement. Found the most in social sciences and humanities, and health policy/health economics, and Paleontology/geology.
Sadamori Kojaku Who Wrote This Paper
page Disambiguating authors at scale. Usually use author traits, but when you try to do that with standard information retrieval methods – like string similarity or edit distances – you get horrible accuracy (most affiliations include University, for example). He used DisAmBert (built on BERT) which outperformed. Most important feature turned out to be article title, which is typically omitted from string matching methods.
Binglu Wang Dynamics of Innovation Abandonment
Using MAG, USPTO, commercial products. Abandonment probability changes over time. Are their structural/social effects? Preferential abandonment. Collapse of network with removed nodes – as happens with sudden disappearance of things like MySpace.
Panel – Innovation and Inequality
Donna Ginther (Kansas) – reviewing her important work 11 years ago in disparate funding at NIH (and NSF) for race/ethnicity, and gender. In the US, the gender gap has closed. Race/ethnicity persists;however, there is a paucity of data on race/ethnicity in the US and none in Europe. Women’s application rates are up and their success rate is higher than mens (31%-28%). Women are also more likely to apply for NSF and be successful. Gap is larger at NSF than NIH. Cites Nakamura et al – attributed to “halo effect” for white investigators instead of bias against black/Hispanic investigators.
Rem Koning (HBS) – who do we invent for? Labor market has bias and gaps, but how about the product market? Patents by women 1976-2010 – USPTO NBER. Their data set is available on Harvard dataverse. Fed text in, spit out MESH terms. Female management of teams more likely to patent in areas of interest to females (like say ovarian cancer).
Innovation platforms like Kickstarter – response on there is taken to VC companies for funding. But majority of users there are men (75-90%). 30% of female products have a female inventor. These products grow 45% less than neutral or male tech
***Marcella Alsan – Excellent talk. Clinical trials… Life expectancy gap. Black less likely to participate in clinical trials and less likely to be prescribed new medicine.
Regulators and interest groups pressuring to diversify trials but market penalized Moderna when slowed recruitment to get a more diverse base for COVID-19 vaccine.
Does lack of racial diversity in the innovative process affect how Black patients benefit from new technologies? Efficacy and diversity both matter in doctor’s prescribing intentions. Diversity matters more for doctors who treat Black patients. If the trial is diverse, prescribing gap depends on efficacy. Same efficacy, diverse trial no different for white but 0.78 s.d. for black patients.
In responding to audience questions, Dr Alsan spoke with the importance of working together with doctors of the community – how that impacted recruitment, study design, and understanding the results.