Governed by a methodical process of repeated observation and experimentation and inspired by the lofty philosophical principles of rationalism and objectivity, it’s easy to think of science as an infallible and incorruptible tool with which to explore pure truths about the universe. But in fact, scientific research is and (until the robots take over) always will be a human endeavor. Long story short, it’s important to consider science results carefully because separating conclusions from the lingering effects of our biases and egos is difficult, and humans have a tendency to simply up and cheat when things don’t go as expected.
A new study in the Annals of Internal Medicine revealed that more than half of the biostatisticians surveyed had been asked by researchers to severely manipulate or falsify data in the past five years. These requests for “inappropriate analysis and reporting” covered a broad range of unethical number-fudging tactics, including removing or altering records to better support the research hypothesis, shifting secondary outcomes to sound like the primary outcome, and modifying measurement scales for some desired results.
Biostatisticians are specialized scientists who translate experimental datasets into comprehensible figures and calculate the degree to which the findings support the research team’s hypothesis. As such, many less-than-scrupulous researchers will try to enhance the apparent outcome of their studies by changing or cherry-picking data during the analysis process.
The authors – whose data analysis is hopefully trustworthy – were inspired to conduct their study after finding that only one other survey quantifying shady requests made to biostatisticians had been published, and that was back in 1998. For their own survey, authors Min Qi Wang, Alice Yan, and Ralph Katz sent online questionnaires to 800 randomly selected members of the American Statistical Association. Information from the 390 biostatisticians who completed the questionnaire was used in the analysis.
According to the responses, 79 percent of biostatisticians aged 23 to 39 had been asked, at least once, to commit one of the eight worst statistics integrity no-nos (a ranking determined in another part of the questionnaire) in the past five years. Seventy-eight percent of those 40-59 years old and 54 percent of those 60-88 years old had received such requests.
And while, as the authors note, “our survey provides no data on whether the request reflected researchers' maleficence or inadequate knowledge about statistical and research methods”, several of the manipulations are impossible to dress up as anything other than blatant fraud. For example, 3 percent of the statisticians had been asked at least once to falsify the statistical significance in order to support the desired result, and a whopping 24 percent had been asked to remove or change information to better support the research hypothesis.
Wang, Yan, and Katz conclude that the best way to reduce the number of these requests in the future is to change the toxic scientific publishing culture. “[T]hese findings should be used to encourage research universities and companies to develop or improve institutional efforts to reduce workplace- and publication-related stress to alleviate the pressure that may be contributing to these inappropriate requests.”