AbstractIntroduction: Science is playing a big role in our society.Therefore, scientists should adhere to ethical standards. The number ofmisconduct in scientific work differs from study to study. This paper focusesonly on plagiarism as one type of scientific misconduct. The following questionwill be discussed: how often does plagiarism occur in different scientificdisciplines?Methods:Every study has adifferent approach to assess the prevalence of Plagiarism.
The researchsubjects of the first chosen study (5) are scientists. This meta-analysis (5)summarizes the results of 17 survey studies, published between 1987 and 2010. Theother two studies (3, 6) analyzed retracted scientific articles (3) and papersthat were submitted in the Croatian Medical Journal from 2009 to 2010 (6). Results:According to themeta-analysis (5), 1.7 % of therespondents admitted of having plagiarized at least once. Furthermore, nearly30 % are knowing that one of their colleagues have committed plagiarism atleast once.
The second study (3) found that 9.8% of the 2047 articles theyanalyzed, were retracted because of plagiarism. In the analysis of manuscripts(6) 85 of 754 articles (11,3%) contained plagiarized parts. Discussion: Because of the limitations and the difference in theresults in many studies, it could be assumed that the results are inaccurate. That´swhy it´svery hard to determine the prevalence of plagiarism in scientific work. Though,I believe that we can estimate the occurrence somewhere between the results ofthe 3 studies (1,7-11 %), provided that most of the studies, containingplagiarism, were detected. IntroductionScience, scientists and scientific research areplaying a big role in our society. Although most people are unaware of it,gaining scientific knowledge affects a large part of our lives and the humandevelopment.
That´s why it´s important for scientists to adhere to ethicalstandards and don´t commit fraud. According to the TUM Code of Conduct forSafeguarding Good Academic Practice (1) “academic misconduct has occurred whena scholar, deliberately or through gross negligence, makes false statements,infringes upon intellectual property rights or the research activities ofothers”. This includes e.g. fabrication and falsification of data, falsestatements in letters of application and about the scholarly achievements ofapplicants, plagiarism, intellectual theft especially of unpublished work,claiming academic authorship, sabotaging the research activities of others andmaking false accusations.In Fanelli´s meta-analysis (2), nearly 2% of therespondents admitted that they fabricated or falsified Research at least once,while 14% of the respondents have noticed scientific misconduct of colleagues.
Fangand colleagues (3) noted that we have a huge increase in retracted articlesbecause of fraud since 1975. As we shallsee there is a big discrepancy in the data about scientific misconduct. I decided to focus only on results about plagiarism,which is probably gaining the most attention in the media whenever a new caseoccurs to the public. For example, the plagiarism scandal involving Mr.
Guttenberg in 2011 created sensation and caused a questioning of theself-regulation of science. Plagiarism is defined as the “unauthorized use orclaims of authorship” (1) or, more detailed, “plagiarism takes many forms, frompassing off another’s paper as the author’s own paper, to copying orparaphrasing substantial parts of another’s prepare, without attribution toclaiming results from research conducted by others” (4). These forms could beself-plagiarism, true/direct plagiarism or patchwork/mosaic plagiarism. (4, 6)This paper discusses the following question: how oftendoes plagiarism occur in different scientific disciplines? Therefore, I chose 3Studies, which will be described in detail.
The first one is a meta-analysis ofsurveys, published in 2014 (5). The second one is a review of retracted biomedicaland life-science research articles (3). Lastly, I will assess a review of manuscriptsin the Croatian Medical Journal (6). MethodsIn this part I´m goingto present you the methodological aspects of the studies. Every study differsmore or less in its approach. There are two reviews (3, 6) which we will lookat, that have a quite similar research object.
At first, we´ll have acloser look to the work of Pupovac and Fanelli (5). In 2014, the authorspublished a systematic review and meta-analysis of anonymous surveys, in whichscientists of different academic disciplines were asked if they´ve everplagiarized. “From May to December 2011 they searched 35 bibliographicdatabases, five grey literature databases and hand searched nine journals forpotentially relevant studies. (5)” Studies,that asked scientists whether they have plagiarized or knew of a colleague whocommitted plagiarism, were included. While studies, which were asking foracademic (e.g. students) plagiarism and studies which didn´t measured theactual prevalence of plagiarism, were excluded.
(In the end 18 relevant studieswere found.) The main outcome of effect was the extent of scientists whoadmitted or witnessed plagiarism at least once. For the statistical analysis “proportions and relative standard errorswere logit-transformed …. Analyses used standard inverse-variance weighting,and heterogeneity was measured with Chochran’s Q and I2 statistics.Both quantities are directly proportional to the amount of between-studyvariability.
Cochran’s Q tests heterogeneity for statistical significance,whereas I2 expresses the proportion of total variation in studyestimates that is due to heterogeneity. Irrespective of whether Cochran’s Q wasstatistically significant, they pooled effects using a random effects modelbased on restricted maximum likelihood estimation. (5)” Furthermore, theyused a mixed-effects meta-regression to evaluate the reasons for differentadmission rates and also tested every factor. “In secondary and exploratoryanalyses, a step-forward approach was used, in which significant factorsretained in the model as covariates whilst each remaining factor was testedagain, one at a time. (5)” Characteristics of the respondents (e.g. country ofthe study, participants discipline) and methodological factors (e.g.
samplingmethod, response rate) were assessed. In addition, the authors compared data ofplagiarism to other forms of misconduct. Therefore, a matched-control paradigm andthe Wilcoxon´s signed rank test was used. Lastly, funnel plots were used tomeasure the risk of publication bias and the robustness of results was tested. Fang and colleagues (3)are following a completely different approach to detect fraud.
In 2012, theypublished a detailed review of retracted scientific articles. 2.047 biomedicaland life-science research articles from PubMed, which have been publishedbetween 1973 and 2011 were included in their review. In contrast to theprevious study, the aim was not only to detect plagiarism. The main outcome wasto find out the reasons of the retraction of the selected articles. Secondary,they wanted to discover temporal trends, geographic origin, journal impactfactors and the “Time-to-Retraction” of the articles. For this reason, they hadto categorize and classify the articles.
“Retracted articles were classifiedaccording to whether the cause of retraction was documented fraud (datafalsification or fabrication), suspected fraud, plagiarism, error, unknown, orother reasons (e.g., journal error, authorship dispute). (3)” 158 articles werereclassified (e.g. from error to fraud) after cross-checking them againstreports of the Office of Research integrity. Any other missing informationabout causes of retraction was searched in Google and included RetractionWatch, news media, and other public records. “Impact factors were based on the2011 edition of Journal Citation Reports Science Edition ….
Journals withoutan impact factor were assigned a value of 0.1. Statistical analyses wereperformed using Prism (GraphPad Software). (3)”Inthe next paper (6) the research objects were, similar to the previous study andin contrast to the first one, original research manuscripts, review manuscriptsand case reports submitted for publication in the Croatian Medical Journal. Theobjective was to identify the prevalence of plagiarism of lodged papers from2009 to 2010. The papers came from 64 countries. 754 documents (abstracts andfull texts) were checked by using the following plagiarism detection software:eTBLAST, Cross-Check, WCopyfind. Those where analyzed through Déjà vu databaseand manually verified.
The findings were categorized into true plagiarism,self-plagiarism, patchwork plagiarism and also were divided into countries oforigin. Furthermore, “manuscripts were grouped into three plagiarism categoriesaccording to the extent of plagiarism (text similarity rate): minor (11-24%),moderate (25-49%), and major (50% and more). (6)” The authors set criteria for plagiarism:Criterion A for abstracts were 6 words in a row that are equivalent to words inalready existing abstracts. Criterion B for the full texts was a text similarity rate higher than10% in parts of the manuscript. If this was the case, the text washand-checked, which included reading the whole paper and set it against thesuspected original. The study procedure is summarized in figure 1. In thestatistical analysis categorical variables were constituted with absolute and relative frequencies; “distributions were compared with test of proportions with power estimation. Continuous variables are expressed as median with 95% confidence interval (CI), because data did not follow the normal distribution (Kolmogorov–Smirnov test).
(6)” For assessing Between-group differences Mann–Whitney and Kruskal–Wallis tests were used. Additionally, the Mann-Whitney test adjusted formultiple comparisons was deployed for Post-hoc comparisons. Spearman´scorrelation coefficient exhibited the degree of association between variables. “A P value of less than 0.05 was considered significant. (6)” Data were analyzed with the MedCalc statistical software.
Russel-Lenth´s power and sample sizewebpage calculated Power estimation for test of proportions. ResultsThis part containsa summary of the most important results of the 3 chosen studies. In Fang andcolleagues (3) article there are results about other types of misconduct. Forthe purpose of this paper I will only mention the results on plagiarism. But before that, we´llexamine the findings of Pupovac and Fanelli (5). Their meta-analysis included17 survey studies (Appendix 1), from which, 7 had been self-reports and theothers asked about colleague’s behavior. The studies were published between1987 and 2010, mostly in the US and in peer-reviewed scientific journals.
1.7% of therespondents confessed to have plagiarized at least once. When the word”plagiarism” was explicitly used in a survey, the number amounts to 1.8%. Bothresults were significant. “Between-studiesheterogeneity was not significantly different from that expected by samplingerror (Cochran e Q= 7.01, df=6, P=0.
32; I2 =30%), althoughstatistical power to detect heterogeneity was low. Meta-regression identifiedfour study characteristics that, taken individual, had a statisticallysignificant influence on the outcome. Admission rates were negativelyassociated with response rate, delivery method (mailed surveys yielded loweradmissions), and year of survey (admission rates have declined over time). Inaddition, surveys conducted amongst students had significantly higher admissionrates than the others (5).” A mean pooled estimated 29,6% of respondentreported, that they know at least one colleague, who committed plagiarism. Insurveys, which explicitly used the word “plagiarism”, the estimate was 23,61%,but the difference not statistically significant. The heterogeneity betweenstudies was large and statistically significant. The exploratory analyses foundout that surveys that asked more generic/indirect questions had higheradmission rates and that there is an interaction between the year of survey andstudy size.
“To assess the robustness of the meta-analytical estimates,they recalculated admission rates on self and non-self reports leaving outone study a time. Admission rates thus recalculated were never significantlydifferent from the overall summary estimate. (5)”The second study (3) found that 9.8% of the 2047articles they analyzed, were retracted because of plagiarism. The rest of themwere revoked because of fraud, error, duplicate publication or another reason. Theissue of plagiarism in retracted articles is a recent but increasing event, asyou can see in Figure 2. Besides, plagiarism appears stronger in emerging andpoorer countries then e.g.
in the USA. The journal impact factor does not showa significant correlation with retractions because of plagiarism. In the analysis ofmanuscripts (6) 57 papers, analyzed by eTBLAST, applied to Criterion A and B.Further WCopyfind-analysis found that 33 of these manuscripts had similar textcontent to already existing papers. 228 abstracts met criterion A, while 151 ofall examined met criterion B, when analyzed with CrossCheck. After all furtheranalysis and manual verification 85 (of 754) articles contained plagiarizedparts. “Of them, 63 (8%) contained trueplagiarism and 22 (3%) contained self-plagiarism ….
Of the 63 manuscriptsthat contained true plagiarism, 18 contained patchwork plagiarism from twosources and one manuscript contained plagiarism from three sources. Ninemanuscripts contained major plagiarism, with eight of them containing trueplagiarism …. (6)” Nearly one fifth of the 85 papers were peer reviewed.Most of the plagiarized manuscripts came from China, Turkey and Croatia, whilethe proportion of plagiarized than non-plagiarized articles from China (21% :8%) was higher than the proportion of articles from Croatia (14% : 25%). DiscussionAs I´ve alreadymentioned in the beginning, there is a big discrepancy in the data.Nevertheless, we saw that the 2 reviews (3, 6) with similar techniques todetect plagiarism found out that 9.8% (3) and about 11% (6) of the examinedpapers contained plagiarism.
It can be concluded that different approaches leadto different results. Douglas L Weed (7) described three types of empiricalestimation studies of misconduct and came to the same conclusion. In Pupovacand Fanelli´s work (5) 1,7% of respondents admitted their own misbehavior,while about 30% noticed one colleague who had plagiarized at least once. Astudy conducted in Iran (8) had 4,9 % of the respondents admitting toplagiarism. Mass surveys asking students about plagiarism had prevalence ratesup to 40% (9, 10).
That leads us to the question, why does the results differfrom another. It can be assumed that the results are possibly inaccurate. AsKrumpal (11) pointed out, survey questions asking about taboo topics are oftendistorted by the social desirability bias. Moreover, Pupovac and Fanelli (5) consideredthat their survey data could be influenced by the study design. Hence, theheterogeneity of the studies and the results is another limitation, which couldnot be explained by their statistical methods.
Besides these issues and thepublication bias I would like to point out one last limitation that was notmentioned by the authors: It was not described what it meant to “know that acolleague plagiarized”. We don´t know if this was only a rumor or the scientisthas real evidence for this accusation. Maybe that´s why the prevalence of plagiarismin self-reports is much lower compared to the prevalence in non-self-reports.
Our other two studies(3, 6) are not vulnerable for social desirability bias, because of theirresearch object. But, there are two other issues. The first one is, as alreadynoticed by Fang and colleagues (3) that “not all articles suspected or fraudhad been retracted”.
Second, the results aren´t representative. We cannot drawconclusions from this data to an entirety, because not all articles wereassessed. The same limitation occurs in the other study (6). The lack ofrepresentativeness here is due to the plagiarism detection software, whichcannot search all databases. Because of those limitations and others indifferent studies it´s very hard to determine the prevalence of plagiarism inscientific work.
Though, I believe that we can estimate the occurrencesomewhere between the results of the 3 studies. (1,7-11 %). Future studies shouldtry to cope with the social desirability bias by using recommended methods (12)e.g. use of forced choice items. In general, we need a lot more studies dealingwith this topic. The number of studies is rare. To provide completeness,studies, that are using the same methodology as our second and third (3, 6)should be conducted as systematic review.