Read our COVID-19 research and news.

Bill Dickinson/Flickr (CC BY-NC-ND 2.0)

Data check: NSF sends Congress a garbled message on misconduct numbers

When a senior National Science Foundation (NSF) official told the House of Representatives science committee this month about a “significant rise in the number of substantive allegations” of research misconduct, her testimony set off alarm bells.

Legislators from both parties were clearly disturbed by this trend, which had led to three dozen findings of misconduct a year, and asked what the Arlington, Virginia–based NSF was doing about it. Committee Republicans unhappy with NSF’s current system of awarding grants saw her words as further proof that Congress needs to keep a closer eye on the $7.5 billion agency.

Well, it turns out there is no such trend, and the overall size of the problem had been greatly exaggerated. Within days of her appearance at a 9 March hearing to discuss NSF’s business practices, NSF Inspector General Allison Lerner admitted as much in a two-page memo to committee Democrats. But her flawed testimony could rekindle a long-simmering debate over the government’s approach to research misconduct.

Lerner’s memo contained 12 years’ worth of data from her office on allegations, investigations, and findings of research misconduct. It showed that the number of investigations her office launched last year—24—was  just one-third of the 2005 figure, and a quarter of the 2008 peak of 99 cases. Moreover, the caseload for the last 3 years was less than half the number of investigations NSF conducted in 2008–10.

The memo also noted that the annual number of findings of misconduct has remained in the teens for the past decade. (Consistent with that reanalysis, Lerner had informed the committee immediately after the hearing that she had meant to say 75 rather than 175 when citing the number of findings over the past 4 years.)

Democrats were furious, and moved quickly to correct the record. “It would be very hard to discern any clear trend over the last decade, let alone a significant increase,” fumed Representative Daniel Lipinski (D–IL), the top Democrat on the research subcommittee, at the start of a 21 March follow-up hearing on NSF’s research priorities.

Eager to defend NSF’s handling of the problem, Lipinski then used the memo to craft a new message more favorable to the agency. “Looking just at fabrication and falsification … the average [number of findings] is 2.6 per year over 12 years and 3.2 per year over the last 5 years,” he told his colleagues, noting that he was omitting plagiarism, the third category in the federal definition. By Lipinski’s count, the latter number meant that a minuscule 0.0064% of all proposals the agency received were tainted. “Research misconduct is a very serious issue,” he acknowledged. “But I think it is important to keep these numbers in mind.”

Lerner told ScienceInsider after the first hearing that her testimony about a rising number of cases referred to investigations into alleged fabrication and falsification, which she suggested are more serious offenses than plagiarism. However, the data suggest otherwise. The number of fabrication and falsification investigations has held fairly steady over the past 5 years (12, 18, 12, 20, and 15). And the number of findings of misconduct in those two categories for each year of that period sits in the low single digits (0, 3, 7, 2, and 4).

Aside from the numbers themselves, Lerner may have opened up a can of worms by focusing on fabrication and falsification. The Office of Research Integrity, an entity within the Department of Health and Human Services that investigates allegations of research misconduct in biomedical science, spends the vast majority of its time on cases involving those two categories, with scant attention to plagiarism. But the opposite is true for NSF.

Of the agency’s 169 findings of misconduct since 2005, 82% involve plagiarism. That is consistent with the attention NSF devotes to investigating plagiarism versus fabrication and falsification. From 2005 through 2010, for example, its plagiarism caseload never dipped below 90% of the total. In fact, 2016 was the first time that the other two categories represented more than one-quarter of NSF’s portfolio.

Experts in the field have long questioned why there is such a stark difference in the portfolio of the two investigative agencies, and nobody has made a convincing case for why biomedical research should be more prone to misconduct than the broad array of science funded by NSF.

The shifting NSF numbers could also revive the hot-bottom issue of how much misconduct is acceptable. (No points for answering none.) The lawmakers’ reaction to Lerner’s testimony suggests they think NSF isn’t doing enough. “What are you doing to prevent these incidents of misconduct?” the chairman of the committee, Representative Lamar Smith (R–TX), demanded of NSF Director France Córdova, who also testified at the 9 March hearing. “And what are the sanctions?”

Córdova answered by explaining the procedures that NSF follows in investigating every allegation of misconduct and the range of sanctions, from excluding a guilty scientist from serving as a reviewer to a ban of up to 5 years on obtaining any federal grant.

Smith has said the two hearings will lay the groundwork for new guidance from Congress on how NSF should operate. Lerner’s garbled message about the extent of misconduct among those seeking NSF funding could come back to haunt NSF officials if Republicans use it as a reason to tighten oversight of the agency.