An Introduction to the Column
By Michael Seadle
What is Information Integrity?
Information integrity is fundamentally about what makes information true or false, both at the scholarly level (research integrity) and for public and policy discourse. There are reports about false information almost daily. A recent example involves the BBC, which has long been a model for the integrity of its reporting. (Sweney, 2018) This column will focus mainly on the scholarly aspects of information integrity, but the effect of integrity problems on policy matters (public health issues, for example) will not be ignored.
The topic includes a broad range of problems, including data falsification, image manipulation, and plagiarism. While plagiarism is perhaps the most prominent issue, it is primarily an ethical and legal issue and generally does not undermine scholarship that builds on it because the results are not necessarily false. This column will discuss all aspects of information integrity, but will focus especially on data problems, since no generalized detection tools exist, though a few disciplines (such as psychology) are working on them.
A core concept in my book on “Quantifying Research Integrity” (Seadle, 2017) is the greyscale approach: integrity issues rarely separate neatly into simple black and white, guilty or innocent, categories. Many scholarly works have imperfections, and problematic works may still contain valid information. From the viewpoint of a university or a publisher, formal decision-making processes involving punishments and retractions may make black-and-white decisions about integrity problems preferable, but such black-and-white decisions can themselves be an integrity issue, since an overly simplistic label is at least partly untrue.
Scholarly literature contains a wealth of examples of integrity problems going well back in historical time. Today there are tools for investigating plagiarism and for examining some kinds of image manipulation. Data falsification presents more of a challenge because of its variety and complexity. Simple cases such as that of Diederik Stapel, who admitted manufacturing his results, are rarer than scholars who make poor choices about data or its interpretation. (Bhattacharjee, 2013) Unintentional error is also an information problem, even if it is not falsification.
Selecting problematic research may have lasting effects on political discourse as well as on scholarship. While the evidence for climate change appears to be overwhelming, studies by a small number of skeptics have given oil and coal lobbies in the US a tool for opposing effective measures to reduce hydrocarbons in the atmosphere. Natural science builds on the ability to reproduce results, and when many scientists produce the same results based on a wide range of measures, the conclusions are normally accepted as valid. Lay persons unfamiliar with the scholarly literature sometimes select flawed studies that confirm their own personal preferences.
Other more historical examples of selection bias can be found in claims about the inferiority of people in the US who were not of northern European descent — not merely those from Africa, but also from Italy, Ireland, and eastern Europe. Such claims were popular among the right wing in many European countries in the Nazi era, and are still popular among some groups today. A basis for them reaches back to Christoph Meiners (Grundriß der Geschichte der Menschheit, 1785) in the 18th century and is as modern as “The Bell Curve” by Richard Herrnstein and Charles Murray (1994). These studies did not fake their data and used scientific methods that seemed appropriate at the time, but they were selective about what evidence they included, and today it is widely accepted that the exclusions skewed results in a particular direction.
Selection bias may have social and cultural origins that can change over time. For those who believe in the inerrancy of Holy Scripture, the data confirming evolution is invalid. A scholar of research integrity needs in some sense to be an historian, in order to understand the research in time and place, and to be an ethnographer, in order to understand integrity violations across cultures and disciplines. No one should imagine that integrity research involves simple labels.
The Research Integrity Literature
This column will focus on discussing papers about research integrity and will look at specific cases, whose complexity gives opportunities to apply a greyscale analysis. There are many good sources of information, not the least of which is Retraction Watch (Oransky, 2018), which provides an excellent news feed and classifies cases of retractions by type and field. Retractions may represent only part of the problem, simply because discovering problems is hard and because false positives may distract from more important issues. The ability to reproduce results is a classic hallmark of good science, but there is good evidence that results in behavioral and social science studies are harder to reproduce than natural-science results for the simple reason that social circumstances change.
The goal of this column is scholarly, not investigative. It does not actively seek out new cases where research integrity may have been violated, but seeks to examine existing cases in order to apply a greyscale understanding of what happened and what the consequences are. As Principal Investigator for the research integrity part of the HEADT Centre, I will be the primary columnist, but others will likely contribute as well, including Dr. Thorsten Beck, who specializes in image manipulation.
Bhattacharjee, Yuduit. 2013. “The Mind of a Con Man.” New York Times, April 26, 2013. Available online.
Seadle, Michael. 2017. Quantifying Research Integrity. Morgan Claypool: Synthesis Lectures on Information Concepts, Retrieval, and Services. Available online.
Sweney, Mark. 2018. “No Title.” New York Times, April 4, 2018. Available online.
Oransky, Ivan, and Adam Marcus. 2018. “Retraction Watch.” 2018. Available online.
Honest Error: a Look at the Literature
By Michael Seadle
Problems with data are arguably the most serious issue for information integrity in the research world, because they undermine the ability of scholars to build on past results. These problems come in many variations, including people who make up fake data, people who manipulate data to get specific results, and people who leave out data or sources. Each of these represent some form of misconduct when done deliberately. Nonetheless not everyone is guilty of malicious intent. Ordinary negligence plays a role too. The results remain unreliable and irreproducible, but the persons involved may be innocent of intentional wrongdoing. This column looks at the scholarly literature on “honest” errors.
Resnik (2012) explains that recognizing honest error is important but hard:
“It is important to distinguish between misconduct and honest error or a difference of scientific opinion to prevent unnecessary and time-consuming misconduct proceedings, protect scientists from harm, and avoid deterring researchers from using novel methods or proposing controversial hypotheses. … the line between misconduct and honest error or a scientific dispute is often unclear,”
Precisely what constitutes honest error may depend on personal judgment. An older study by Nath (2006) in the Medical Journal of Australia looked at ” [a]ll retractions of English language publications indexed in MEDLINE between 1982 and 2002…” and “[t]wo reviewers categorised the reasons for retraction of each article…”. Nath concluded that:
“Of the 395 articles retracted between 1982 and 2002, 107 (27.1%) were retracted because of scientific misconduct, 244 (61.8%) because of unintentional errors, and 44 (11.1%) could not be categorised.”
The percentage of unintentional errors suggests surprisingly high rate of unintentional error. While it is possible that misconduct has increased significantly over time (see below for more recent numbers), the more likely lesson here is that it matters how the classification is made. It is hard to know how accurate the classifications of misconduct are under circumstances where the assumption of innocence is not always strictly observed after an accusation has been made.
Estimates of Size
Later studies do not confirm the Nath estimate about the number of unintentional errors. An article by Arturo Casadevall (2014) argues that
“Analysis of the retraction notices for 423 articles indexed in PubMed revealed that the most common causes of error-related retraction are laboratory errors, analytical errors, and irreproducible results. … The database used for this study includes 2047 English language articles identified as retracted articles in PubMed as of May 3, 2012…“
This suggests that the cause of just under 12% of the PubMed retractions are essentially ordinary human error. A different study by Moylan and Kowalczuk (2016) looks at the BioMed Central journals finds a similar percentage:
“Honest error accounted for 17 retractions (13%) of which 10 articles (7%) were published in error. … A total of 13 articles (10%) of retractions were due to problems with the data. Often these issues occurred through honest error in how the data were handled, for example … although in some cases it is difficult to determine whether honest error or misconduct was the cause. “
Daniele Fanelli (2016) offers a somewhat higher percentage of honest error:
“However, retractions reliably ascribed to honest error account for less than 20% of the total, and are often a source of dispute among authors and a legal headache for journal editors. The recalcitrance of scientists asked to retract work is not surprising. Even when they are honest and proactive, they have much to lose: a paper, their time and perhaps their reputation. Much reluctance to retract errors would be avoided if we could easily distinguish between ‘good’ and ‘bad’ retractions.
In this case good retractions are generally ones where the authors recognize their own mistake and ask for the paper to be withdrawn. Fanelli (2016) makes the further argument that:
“Self-retractions should be considered legitimate publications that scientists would treat as evidence of integrity. Self-retractions from prestigious journals would be valued more highly, because they imply that a higher sacrifice was paid for the common good.“
This could, as he notes, be open to abuse, but some abuse could well be tolerable in the interests of providing an incentive for researchers to withdraw misleading results so that they do not mislead other scholars. Considering present publication pressure and the effect of public opinion, researchers may be unwilling to admit honest errors because they will be thought guilty of misconduct. It may be hard to escape censure regardless of the choice.
One of the measurements that can help define honest error is the degree to which errors confirm the desired conclusions. This is not to say that every error in favor of the authors’ arguments is dishonest, but errors that weaken the conclusion are more likely unintentional. There is of course a human tendency to believe confirming results and to doubt disruptive ones, and a part of research training that may need more emphasis is a healthy skepticism toward desired results. Another form of measurement has to do with the frequency of error. Everyone makes some errors. When authors repeatedly make errors, it may be reasonable to think that the errors follow a standard distribution where some are for and some against the conclusions. A pattern that is consistently in favour of the desired conclusion may imply more bias than honesty.
Those judging integrity should not forget that honest errors exist, and that people under career or social pressure may be more error prone without particular ill intent.
Casadevall, Arturo, R. Grant Steen, and Ferric C. Fang. 2014. “Sources of Error in the Retracted Scientific Literature.” FASEB Journal 28 (9): 3847–55. Available online.
Fanelli, Daniele. 2016. “Set up a ‘self-Retraction’ System for Honest Errors.” Nature. Available online.
Moylan, Elizabeth C., and Maria K. Kowalczuk. 2016. “Why Articles Are Retracted: A Retrospective Cross-Sectional Study of Retraction Notices at BioMed Central.” BMJ Open 6 (11). Available online.
Nath, Sara B., Steven C. Marcus, and Benjamin G. Druss. 2006. “Retractions in the Research Literature: Misconduct or Mistakes?” Medical Journal of Australia. Available online.
Resnik, David B., and C. Neal Stewart. 2012. “Misconduct versus Honest Error and Scientific Disagreement.” Accountability in Research. Available online.