When we think about bad science, we often imagine big scandals. However, not all problems in science look like clear fraud. Many of them happen in a more uncomfortable space: the area between grey science and scientific misconduct. This is where the concept Questionable Research Practices (QRPs), become prominent.
What are Questionable Research Practices?
The term Questionable Research Practices gained visibility after the publication of the article Measuring the Prevalence of Questionable Research Practices With Incentives for Truth Telling in 2012 (Figure 1).
Figure 1: “Number of research documents that mention the term questionable research practices over time. Data from the database Scopus up until 2021” by Peter M. Dahlgren from https://howscientistslie.com/qrps-history.html. Licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Consultation date: 08/07/2026.
In that article, the authors distinguish between research misconduct and Questionable Research Practices. Research misconduct refers to the clearest and most serious cases, such as fabrication, falsification or plagiarism. QRPs, instead, are defined as practices located in a grey area: they may not always look like direct fraud, but they can still damage the reliability and transparency of science. For example, changing a hypothesis after seeing the results, trying several statistical analyses and reporting only the significant one, or describing methods in an unclear way may seem less serious than inventing data.
However, QRPs still distort the scientific record. For this reason, this post approaches QRPs from a different perspective. In Defining the Spectrum of Questionable Research Practices (QRPs), Kolstoe suggests that it may be more useful to understand QRPs as a spectrum of behaviours, with research misconduct located at the right end of the spectrum, rather than as clearly separate category (Figure 2). For us, this perspective is especially valuable because it avoids minimizing QRPs: the issue is not only whether a practice is officially classified as “misconduct” or not.
Figure 2: “The spectrum of questionable research practices” by Simon Kolstoe from https://doi.org/10.37672/UKRIO.2023.02.QRPs. Licensed under Creative Commons Attribution-Non Commercial-No Derivatives 4.0 International License. Consultation date: 08/07/2026.
Some Common Examples of QRPs
HARKing
One common example of QRPs is HARKing, which means Hypothesizing After the Results are Known. This happens when researchers create or modify a hypothesis after seeing the results, but present it as if it had been planned from the beginning.
It is important to be careful here. Accusing a specific researcher of HARKing is difficult, because we usually do not know what they really planned before analysing the data. A paper can look suspicious, but proving that a hypothesis was created after seeing the results is not easy. However, some studies have tried to detect patterns that may suggest HARKing. For example, the article Evidence of HARKing in mouse behavioural tests of anxiety analysed 206 studies on the effects of diazepam in two commonly used behavioural tests: the open-field test and the hole-board test. The authors found that the reason given for using each test, and the effect they said they expected, strongly depended on the results that were finally reported. In other words, the interpretation of the study seemed to change depending on the outcome. The authors argue that the most likely explanation for this pattern is HARKing.
P-Hacking
Another example is p-hacking. This happens when researchers try different analyses until they find a statistically significant result. For example, they may test several models and only report the one that “works”, stop collecting data when the result becomes significant, or ignore variables that do not produce convenient results.
Watch the following video to better understand how p-hacking works!
Honorary Authorship
Another important (and really common) case is honorary authorship. This happens individuals are listed as authors on scholarly publications despite lacking a significant intellectual contribution. Authorship should be a form of responsibility, not a reward, a favour or a currency inside academic careers.
Data Fabrication and Manipulation
A more serious example, located on the right-hand side of the spectrum, is data fabrication. This happens when a researcher invents data and reports them as if they were real observations or experiments.
One example is the following retracted article, in which the hospital report described a broad investigation into the author’s research. The investigation found that, in most of the studies, the original patient documentation could not be found. And, where records were available, the report found evidence of scientific fraud, which can be seen as strong evidence of data fabrication or manipulation.
It is worth highlighting that, sometimes, the first idea is not about fabricating the data, but, if the study simply didn’t go as expected, some choose to manipulate it…
Plagiarism and Self-plagiarism
Another really right-hand side example is plagiarism. Basically, plagiarism refers to the use of another person’s work without acknowledgement, presenting it as one’s own. This does not only happen with text, or by directly plagiarising a whole article. It can also happen in more subtle ways, for example by plagiarising images. Look at Figure 3. The figure on the left was literally stolen and presented as original, when it was actually just a zoomed-in version of the original image shown on the right.
Figure 3: Comparison between a plagiarised image and the original image. Left panel from a retracted article: https://doi.org/10.1021/acsomega.3c00979. While right panel from the original article Effect of Core–Shell Rubber Nanoparticles on the Mechanical Properties of Epoxy and Epoxy-Based CFRP by Tatjana Glaskova-Kuzmina et al. Consultation date: 08/07/2026.
A related but slightly different practice is self-plagiarism. This happens when researchers reuse parts of their own previous publications, such as text, data or results, without making this reuse clear. Self-plagiarism may seem less serious than copying someone else’s work, because the material comes from the same author. However, it is still problematic. It can make a new paper look more original than it really is, inflate scientific production, and create unnecessary repetition in the literature.
Nonsense AI-generated Content
A final example is the use of nonsense AI-generated content, which is becoming more widespread than one might expect. If artificial intelligence is doing something in science, it is making visible many problems that were already there, but were easier to ignore before.
Look at Figure 4, which was published in a Frontiers journal. In this case, the problem is not only that the article included strange and scientifically meaningless AI-generated images. The most surprising part is that these images passed peer review.
Figure 4. Literally, a rat with anatomically absurd, disproportionately enlarged male genitalia. From the retracted article: https://doi.org/10.3389/fcell.2023.1339390
And this was not exactly a subtle problem. We are not talking about a small methodological decision hidden in the paper, or a statistical choice that is difficult to detect. The images included obvious anatomical mistakes, strange labels and visual elements that made little scientific sense.
This case connects Questionable Research Practices with Questionable Publisher Practices. On the one hand, including nonsense AI-generated content in a scientific article can be seen as a questionable research practice. On the other hand, the fact that this content passed peer review also points to a questionable publisher practice, in this case by Frontiers. If something so visible was not detected before publication… then the problem is not only in the article itself.
How can Open Science help?
Open Science cannot solve everything by itself… but it can reduce the space where QRPs happen. The solution is not only to punish misconduct. We also need to change the incentives that make questionable practices attractive. Science should not reward only speed, volume and positive results.
Preregistration can help distinguish between hypotheses that were planned before the study and analyses that were explored after seeing the results. Registered reports can reduce the obsession with positive results, because the study is evaluated before the results are known. And sharing materials such as data or code, when possible, can make research more transparent and easier to check. So… what if we also made these practices an incentive?