What is your threshold in detecting plagiarism when using CrossCheck/ithenticate/turnitin or similar software? How do you choose texts suspicious of plagiarism?
I've recently come over an Australian OA journal stating that they'll check for plagiarism and do not accept submissions with a similarity rate higher than 18%. It would be good to know, how appropriate this measure actually is. Especially when confronted with the findings of recent studies showing that plagiarism SW is not working properly.
We (Croatian Medical Journal) have a completely different algorithm. We manually check all manuscripts that are similar with a published source more than 10% in one source or more than 15% in first three sources. This threshold is artificial, and there were cases when manuscripts were not suspicious according to the CrossCheck but peer-reviewers have indicated that similar articles were already published. Crosscheck is not perfect, but still it is very helpful and it would be very time consuming to scan manuscripts without a threshold. 18% (if this journal uses CrossCheck) is a very low threshold and I think there will be many false positive results.
In my career, I reviewed a few scientific articles and significantly more works of students. I always look what done the author. Plagiarism is very bad, but worse accuse someone of plagiarism (without evidence). Sometimes, we must describe a theory, which was described thousends times (then the quote is very important). It can be reason of false positive plagiarism if we get only a threshold of detecting plagiarism, and we don't check what is actually new in this work.
I believe it shouldnt be less than %20 because it is very important for the researchers to review the literature and it is often to correspond others ideas. We should think how many types of writing we can achieve alternatively to express same thing as a human being in a language. As long as the source is cited appropriately, it should be fine.