A couple days ago, Senator Ted Cruz (R-TX) released a spreadsheet of about 3.5k “woke DEI grants” made by NSF. This spreadsheet was the basis for a report released last October by the minority of the Senate Commerce, Science, and Transportation committee (of which Cruz was the minority chair). Congressional Republicans have a long history of complaining that NSF funds “wasteful” or “silly” projects, and this report and spreadsheet continue that tradition, using the anti-DEI framing that happens to be popular today.
As a computational social scientist and philosopher of science, one thing that struck me about the report was a “methods section,” a pair of appendices that narrated the text mining methods used by congressional staff to assemble the list of awards and apply labels such as “social justice.” Staff used “N-gram analysis” to find “the most-often-used terms and phrases” across about 30,000 NSF award abstracts, then “parsed those terms for DEI keywords and phrases,” “using over 800,000 variations and combinations.” The second appendix is a series of term lists — one for each of the five labels — with a total of about 750 words or short phrases. The authors of the report used multiple rounds of filtering, “a categorization ‘qualifier’ tagging formula in Excel,” and three rounds of manual review, to arrive at the final list of about 3.5k awards and the five labels.
At this level of description, the methods used to assemble this report are similar to methods my research group uses in our own text mining projects. For example, in a paper currently under review, using the fulltext of tens of thousands of journal articles we trace the presence of race science across mainstream and fringe scientific journals over the second half of the twentieth century. Such work is often grouped under the interdisciplinary heading of “metascience,” utilizing scientific methods to better understand scientific research processes and communities.
In this sense, Senator Cruz is engaged in weaponized metascience: using metascience in an attempt to undermine the political and cultural standing of scientific research.
Like congressional complaints about NSF, weaponized metascience is not new. During the first Trump administration, the Environmental Protection Agency proposed (and briefly adopted) a rule, originally called “Strengthening Transparency in Regulatory Science,” that appealed to concerns about a widespread replication crisis and the open science movement in an effort to undermine the use of public health research at the agency. Numerous scientists and advocacy organizations criticized the rule as “anti-science,” without noting that it appealed to widespread sentiments within the scientific community itself (D. J. Hicks 2022b, 2023).
It should not be surprising that the quality of weaponized metascience is often low by the standards of academic research. In text mining, one way to avoid “false positives” is to filter out documents that only match a single keyword. This filter is idempotent, meaning that it has no effect after being run once: all of the single-match documents are removed on the first pass. Nonetheless, the “methods section” of the Cruz report says that this filter was applied twice, at two different stages of review. The first pass removed 3,118 awards; and, somehow, the second removed 5,715. At best, this is a serious misreporting error, perhaps if the second pass involved a much larger keyword list; at worst it indicates a severe error in the analysis scripts.
The report also makes claims that these methods, and the resulting database, cannot possibly support. For example, the report claims that “Instead of identifying the best or most talented scientists, NSF funded researchers who prioritized filling out research teams and programs based on ethnicity, cultural background, or political perspectives.”
Most of the researchers funded by NSF grants — students, postdoctoral researchers, professional research staff — are hired by the universities that receive the awards, not NSF. Some personnel are named in the initial grant application, but otherwise universities don’t notify NSF who they hire using NSF-awarded funds, much less report them for release in a public database. The particular data portal used by Cruz’ staff does not even contain the names of the primary investigator for each award (though these names are available through NSF’s award search portal). And, of course, even if Cruz’ staff did somehow know the names of the members of the research teams, they still wouldn’t know their qualifications, ethnicity, cultural backgrounds, or political perspectives, much less what role these factors played in the hiring decisions.
Similarly, a distinguished statistician — a fellow of both the American Statistical Association and the American Association for the Advancement of Science, and a member of EPA’s Science Advisory Board during the first Trump administration — published numerous papers using a fundamentally flawed statistical methodology (D. J. Hicks 2022a). Many of these papers claimed to show that research misconduct was widespread in air pollution epidemiology. At least some were funded by the American Petroleum Institute (D. Hicks 2022).
As a final example, recently the Trump administration attempted to unilaterally impose a cap of 15% on indirect costs for NIH-funded awards. Reporters at Wired (Reynolds n.d.) traced this policy to a report by two affiliates of the Heritage Foundation, which alleged that indirect costs were used to “subsidize the agenda of the political left through funding its research agendas and Diversity, Equity, and Inclusion (DEI) staff on university campuses.” The primary evidence for this claim was a collection of simple univariable regressions of the number of “DEI staff” against indirect costs for 82 universities. The authors of the report made no effort whatsoever to adjust for potential confounders in this regression — despite the lead author being a former Distinguished Professor and education researcher at the University of Arkansas.
But the low quality of weaponized metascience is beside the point. The purpose of weaponized metascience is to appear scientific, to adopt the trappings of science in order to discredit science. Over the past 150 years or so, our society has become deeply “scientistic,” holding up science (or things that look like science) as the best and most reliable source of knowledge. Science and its defenders need to understand that, by and large, the contemporary far right and its allies are not interested in challenging this scientism. Project 2025 doesn’t say anything about replacing evolution with intelligent design in schools, for example. Instead the goal is to turn science into a tool of their political and cultural project, dominated by conservative, nationalistic, straight white men. To Make Science Great Again.