A blog on statistics, methods, philosophy of science, and open science. Understanding 20% of statistics will improve 80% of your inferences.

Saturday, September 19, 2020

P-hacking and optional stopping have been judged violations of scientific integrity

On July 28, 2020, the first Dutch academic has been judged to have violated the code of conduct for research integrity for p-hacking and optional stopping with the aim of improving the chances of obtaining a statistically significant result. I think this is a noteworthy event that marks a turning point in the way the scientific research community interprets research practices that up to a decade ago were widely practiced. The researcher in question violated scientific integrity in several other important ways, including withdrawing blood without ethical consent, and writing grant proposals in which studies and data were presented that had not been performed and collected. But here, I want to focus on the judgment about p-hacking and optional stopping.

When I studied at Leiden University from 1998 to 2002 and commuted by train from my hometown of Rotterdam I would regularly smoke a cigar in the smoking compartment of the train during my commute. If I would enter a train today and light a cigar, the responses I would get from my fellow commuters would be markedly different than 20 years ago. They would probably display moral indignation or call the train conductor who would give me a fine. Times change.

When the report on the fraud case of Diederik Stapel came out, the three committees were surprised by a research culture that accepted “sloppy science”. But it did not directly refer to these practices as violations of the code of conduct for research integrity. For example, on page 57 they wrote:

 “In the recommendations, the Committees not only wish to focus on preventing or reducing fraud, but also on improving the research culture. The European Code refers to ‘minor misdemeanours’: some data massage, the omission of some unwelcome observations, ‘favourable’ rounding off or summarizing, etc. This kind of misconduct is not categorized under the ‘big three’ (fabrication, falsification, plagiarism) but it is, in principle, equally unacceptable and may, if not identified or corrected, easily lead to more serious breaches of standards of integrity.”

Compare this to the report by LOWI, the Dutch National Body for Scientific Integrity, for a researcher at Leiden University who was judged to violate the code of conduct for research integrity for p-hacking and optional stopping (note this is my translation from Dutch of the advice on page 17 point IV, and point V on page 4):

“The Board has rightly ruled that Petitioner has violated standards of academic integrity with regard to points 2 to 5 of the complaint.”

With this, LOWI has judged that the Scientific Integrity Committee of Leiden University (abbreviated as CWI in Dutch) ruled correctly with respect to the following:

“According to the CWI, the applicant also acted in violation of scientific integrity by incorrectly using statistical methods (p-hacking) by continuously conducting statistical tests during the course of an experiment and by supplementing the research population with the aim of improving the chances of obtaining a statistically significant result.”

As norms change, what we deemed a misdemeanor before, is now simply classified as a violation of academic integrity. I am sure this is very upsetting for this researcher. We’ve seen similar responses in the past years, where single individuals suffered more than average researcher for behaviors that many others performed as well. They might feel unfairly singled out. The only difference between this researcher at Leiden University, and several others who performed identical behaviors, was that someone in their environment took the 2018 Netherlands Code of Conduct for Research Integrity seriously when they read section 3.7, point 56:

Call attention to other researchers’ non-compliance with the standards as well as inadequate institutional responses to non-compliance, if there is sufficient reason for doing so.

When it comes to smoking, rules in The Netherlands are regulated through laws. You’d think this would mitigate confusion, insecurity, and negative emotions during a transition – but that would be wishful thinking. In The Netherlands the whole transition has been ongoing for close to two decades, from an initial law allowing a smoke-free working environment in 2004, to a completely smoke-free university campus in August 2020.

The code of conduct for research integrity is not governed by laws, and enforcement of the code of conduct for research integrity is typically not anyone’s full time job. We can therefore expect the change to be even more slow than the changes in what we feel is acceptable behavior when it comes to smoking. But there is notable change over time.

We see a shift from the “big three” types of misconduct (fabrication, falsification, plagiarism), and somewhat vague language of misdemeanors, that is “in principle” equally unacceptable, and might lead to more serious breaches of integrity, to a clear classification of p-hacking and optional stopping as violations of scientific integrity. Indeed, if you ask me, the ‘bigness’ of plagiarism pales compared to how publication bias and selective reporting distort scientific evidence.

Compare this to smoking laws in The Netherlands, where early on it was still allowed to create separate smoking rooms in buildings, while from August 2020 onwards all school and university terrain (i.e., the entire campus, inside and outside of the buildings) needs to be a smoke-free environment. Slowly but sure, what is seen as acceptable changes.

I do not consider myself to be an exceptionally big idiot – I would say I am pretty average on that dimension – but it did not occur to me how stupid it was to enter a smoke-filled train compartment and light up a cigar during my 30 minute commute around the year 2000. At home, I regularly smoked a pipe (a gift from my father). I still have it. Just looking at the tar stains now makes me doubt my own sanity.

 


This is despite that fact that the relation between smoking and cancer was pretty well established since the 1960’s. Similarly, when I did my PhD between 2005 and 2009 I was pretty oblivious to the error rate inflation due to optional stopping, despite that fact that one of the more important papers on this topic was published by Armitage, McPherson, and Rowe in 1969. I did realize that flexibility in analyzing data could not be good for the reliability of the findings we reported, but just like when I lit a cigar in the smoking compartment in the train, I failed to adequately understand how bad it was.

When smoking laws became stricter, there was a lot of discussion in society. One might even say there was a strong polarization, where on the one hand newspaper articles appeared that claimed how outlawing smoking in the train was ‘totalitarian’, while we also had family members who would no longer allow people to smoke inside their house, which led my parents (both smokers) to stop visiting these family members. Changing norms leads to conflict. People feel personally attacked, they become uncertain, and in the discussions that follow we will see all opinions ranging from how people should be free to do what they want, to how people who smoke should pay more for healthcare.

We’ve seen the same in scientific reform, although the discussion is more often along the lines of how smoking can’t be that bad if my 95 year old grandmother has been smoking a packet a day for 70 years and feels perfectly fine, to how alcohol use or lack of exercise are much bigger problems and why isn’t anyone talking about those.

But throughout all this discussion, norms just change. Even my parents stopped smoking inside their own home around a decade ago. The Dutch National Body for Scientific Integrity has classified p-hacking and optional stopping as violations of research integrity. Science is continuously improving, but change is slow. Someone once explained to me that correcting the course of science is like steering an oil tanker - any change in direction takes a while to be noticed. But when change happens, it’s worth standing still to reflect on it, and look at how far we’ve come.

No comments:

Post a Comment