Youmans, Robert J. (2011) Does the adoption of plagiarism-detection software in higher education reduce plagiarism? In: Studies in Higher Education. Vol. 36, Issue 7, pp. 749-761. DOI: 10.1080/03075079.2010.523457.[It just became available online at our university library, as this journal participates in the 18-month embargo trying to force researchers to pay more for timely access to journals, or make a physical trip to the library to read the article.]
Youmans reports on a fascinating study that divided up students into a 2x2 group: being required to cite 3 peer-reviewed publications in a 10-page paper or only receiving the suggestion of using peer-reviewed publications, and then half of each group randomly selected and publicly warned that plagiarism-detection software would be used. All of the students were actively educated about plagiarism and signed an honor code prior to writing the papers.
The surprising results were that there was no difference in plagiarism incidence between the groups. They used Turnitin, and modified the papers before submitting to remove some of the extraneous text parallels that this system tends to flag on items such as bibliographic information. They also had all of the papers graded by an instructor who did not know the Turnitin result for each paper. Still, there was intentional plagiarism, even in the group of students who were warned that their papers would be checked!
A second study looked into whether the students' knowledge of how the system works would affect the plagiarism amount found. This, too, did not predict whether they would plagiarize or not.
Youmans has a few theories about why this is so, but the study itself only addresses the question of whether just publicly using plagiarism-detection software decreases the amount of text parallels found. The result is: no.
This has quite important consequences for the decision of a university to adopt the use of such systems. Youmans notes that the reports of the system have to be analyzed by a teacher who understands what the results are reporting, as there is a good bit of non-plagiarism reported as text overlap, as he calls it, but that the systems are a useful tool for finding candidate sources.
Let me repeat myself: Software cannot solve social problems. There is no magic software that will detect plagiarism. This means that we have to work hard at educating students on how to write properly, and as Youmans suggests, it needs to be regularly repeated.