Citation, the act of connecting text statements through reference to the broader literature, is not simply an impartial scholarly method for joining related published knowledge. Citation may be used for self serving purposes311
or as a tool for persuasion312
(see web extra note 14). These aspects of citation might be called social citation. I studied how distortions of the persuasive aspect of social citation may result in broad acceptance of unfounded claims as fact. These distortions can be detected and interpreted through social network theory309
because citation as persuasion is a social behaviour. Network theory applied to citation networks constructed from entire paper bibliographies, such as the science citation network,313
can disclose societal attitudes to journals and specific papers (for example, impact factors), but these networks are not suitable for understanding the foundation for belief in specific claims. When networks are instead confined to citation pertaining to one set of related claims (a claim specific citation network), they become sharply focused tools for understanding social communication pertaining to the claims—what is in effect the published record of a belief system shared by a community. These allow for study of not just what is said about a belief (the traditional scope of review papers), but also who hears it and how it is retold.
The general approach taken here (fig 7) addressed belief in claims; no experiments were done addressing their truth. The computational analysis of the claim specific citation network representing this belief system detected certain distortions in the patterns of citation that would not have been expected had only scholarly citation been used. Primary data that weakened or refuted claims on which the belief was based were ignored (citation bias) and a small number of influential papers and citations exponentially amplified supportive claim over time without presenting new primary data (amplification). Certain related claims were invented as fact. The combined effects of these citation distortions resulted in authority of the belief (acceptance of it) according to social network theory.
Fig 7Overview of approach. After construction of the claim specific citation network, a combined manual and computational endeavour, steps on left (determination of authorities [yellow papers] and identification of amplification) require only (more ...)
There are varied forms and consequences of distorted persuasive citation seen in this study (see box). Citation bias against critical content can be used for the systematic support of claim,314
results in the loss of implications of isolated data (see web extra note 15), and can be used to justify construction of animal models, which can then be circularly used to amplify claim (see web extra note 16). Such animal models have enormous appeal, and some publications describing them achieved authority status in this network (fig 1) despite reporting no data addressing the claim—that is, whether these β amyloid related molecules are present in human inclusion body myositis muscle. Amplification involves repetitive citation of review papers or other papers lacking data, often through self citation, features noted previously in a variation of a claim specific citation network.315
Invention has multiple variations.
Vocabulary of citation distortions
- Both scholarly and social forms: the scholarly form connects statements to the broader medical literature, the social form (social citation) includes self serving and persuasive subtypes
- Self serving citation is always a distortion
- Persuasive citation may be necessary to communicate new, sound claims to the scientific community; it may, however, have distorted uses—citation bias, amplification, and invention
- Systematic ignoring of papers that contain content conflicting with a claim
- Bolster claim; justifying animal models to provide opportunities to amplify claim
- Expansion of a belief system without data
- Citation made to papers that don’t contain primary data, increasing the number of citations supporting the claim without presenting data addressing it
- Citation diversion—citing content but claiming it has a different meaning, thereby diverting its implications
- Citation transmutation—the conversion of hypothesis into fact through the act of citation alone
- Back door invention—repeated misrepresentation of abstracts as peer reviewed papers to fool readers into believing that claims are based on peer reviewed published methods and data
- Dead end citation—support of a claim with citation to papers that do not contain content addressing the claim
- Title invention—reporting of “experimental results” in a paper’s title, even though the paper does not report the performance or results of any such experiments
Three factors may account for how citation distortions created authority in this belief system. Foremost is the power of citation through the choice of which papers to cite and which to ignore (citation bias), by citing but distorting content (citation diversion), and by using citation to invent fact (citation transmutation, dead end citation, and back door invention).
Second is an inherent property of negative results, which failed to spread through the network. These were not repeatedly cited by their authors in subsequent papers (only one instance was present274
) as perhaps there was simply nothing further to say about them. Unlike “positive results” there is nothing exciting to be repeatedly written about how something was not found in an experiment. Thus the progression from data to accepted claim is different within a single paper compared with across many papers in a specialty. Within a single paper readers generally view new claims as false until proved true through convincing methods and results. Across a network of papers, however, the barrier to the propagation of negative results biases claims as being viewed as true until proved false.
Thirdly, this belief system is possibly an information cascade (also called an informational cascade),316 317
an entity resulting when people perceive advantage in accepting the prevailing view over any private information they may have when making choices. Indeed certain mathematical properties of information cascades (preferential attachment) would be expected to produce a network with properties seen here (a biased network with a power law distribution of node degrees; see web extra note 3). Many authors may just not be aware of the critical data, as these data are effectively isolated from the discourse about this claim and not mentioned in any review articles. Although unsound information cascades are in theory fragile and fall apart quickly when exposed,316
this may not occur in biomedical belief systems, where contradicted claims may persist.318
Many published biomedical belief systems may be information cascades because repetition of claims is ubiquitous in the biomedical literature. Many are built on sound data, with authors repeating claims after trusting the published expert opinion of their colleagues. However, there are incentives for generating and joining information cascades regardless of their soundness. Joining an information cascade aids publication as articles have to say something and negative results are biased against.319
Generating and joining an information cascade may improve the likelihood of obtaining research funding because hypothesis driven research is an essential requirement320
at many research funding agencies such as the National Institutes of Health, and successful funding generally requires a “strong hypothesis . . . based on current scientific literature”320
—that is, the published belief system of a claim. Chances for successful funding may therefore be increased through joining the cascade (repeating the claim and proposing experimental plans around it). In the extension of this citation network into text within grant proposals that have been funded by the National Institutes of Health, citation bias, diversion, or invention were often present. Once research funding has been used to join a cascade there are further incentives to interpret results through confirmation bias (“in a way that confirms one’s preconceptions and to avoid information and interpretations which contradict prior beliefs”321
) to demonstrate success of the research for subsequent funding. Although joining an information cascade may be an optimal behaviour for some people, it reduces the likelihood that future investigators can discover whether it is sound.317
Methods for the construction and analysis of comprehensive claim specific citation networks present challenges and limitations. These include interpreting meaning of text, as people may reasonably interpret text differently, and understanding the distinct phenomena observed (see web extra note 17 for a discussion of these issues). In principle many biomedical claims have an associated citation network, the study of which provides a powerful approach to detecting citation bias, amplification, and invention, and understanding the nature of the authority of the claim.
What is already known on this topic
- In addition to its scholarly use, citation has social uses, both self serving and as a tool for persuasion
- One distortion of this persuasive aspect of citation, citation bias, has been recognised in clinical trial reporting where it may lead to false belief about a therapy’s efficacy
What this study adds
- Distortions in the persuasive use of citation—bias, amplification, and invention—can be used to establish unfounded scientific claims as fact
- Categorising these distorted uses of citation and having vocabulary for them aids in their recognition
- How scientific data evolve into entire published biomedical belief systems around specific claims can be studied through a device called a claim specific citation network and the use of social network theory