Princeton Notebook — A Question of Honor

“For just the next question, we ask you to get a coin. Flip it twice. If the coin comes up heads both times, then answer this question truthfully. But if it comes up a head one time and a tail the other time, or if it comes up tails both times, then purposively lie. (That is, give us the opposite of the truth.) Since we do not see you flip the coin, WE HAVE NO WAY OF KNOWING IF YOU ARE TELLING THE TRUTH OR THE OPPOSITE TO THE TRUTH.
“Have you ever cheated on an inclass examination while you’ve been at Princeton?”

— Daily Princetonian Poll

The question used a complicated form of what pollsters call the Randomized Response Technique, a device designed to assure respondents that their answers cannot be traced back to them. If those being surveyed follow the instructions, the coin toss method theoretically should improve a poll’s accuracy by overcoming the respondents’ natural reluctance to incriminate themselves on sensitive issues like cheating. Since statistically two heads should come up one fourth of the time, the actual number of cheaters can be computed from the results by solving two simultaneous mathematical equations. * When the Daily Princetonian processed the responses it got from 519 randomly selected students last month, its calculations indicated that 34 percent had cheated on an examination at least once during their stay at the university.

            Obviously uncomfortable with the size of the figure, the Princetonian warned not only that its poll’s “sampling tolerance” produced a margin of error of plus or minus eight percentage points on the cheating question, but also that its estimate “may be inflated because some students might not have complied with the detailed instructions.” It quoted Michael R. Kagay, a Wilson School lecturer who has been a consultant for the New York Times-CBS Poll, as saying the inflation could be as high as 10 percent. On the other hand, George Gallup was reported to have said, “I see no reason to challenge or doubt the 34-percent figure.”

            Yet most observers on campus saw plenty of reasons to challenge the Prince poll. Robert M. May, a biologist who has done statistical studies of animal populations, pointed out that the same 34-percent figure would result if no students had cheated but 23 percent failed to follow the instructions. “While I do not for the moment suggest no Princeton students have cheated on exams, this numerical example serves to show how minor disobedience to the rules of the poll can produce quite misleading conclusions.”

            In fact, when members of the Honor Committee informally questioned students who had been surveyed by the Prince poll, they found that perhaps as many as 23 percent had failed to follow its instructions. Dean of the College Joan S. Girgus observed, “This polling technique asked students to respond in an unusually difficult way, in that it asked 75 percent of those who had never cheated to lie and say that they had cheated. It seems entirely plausible — indeed likely — that many students faced with such an instruction would simply tell the truth and say they had never cheated. [But] the statistical analysis that was used would treat these truthful statements as lies…thus inflating the estimated amount of cheating quite dramatically.”

            Similarly, Student Government Chairman Eliot L. Spitzer ’81 told reporters, “My feeling is that the figures are inflated and that the technique produced unusually high numbers.” One television newsman who came to cover the story could not find a single undergraduate who would admit to having observed any cheating at Princeton. Kagay noted, however, that such anecdotal impressions could be misleading: “When you consider the number of exams in which a student sits during the course of his career, and think of them as a number of temptations, and if you assume each student who admits to cheating has cheated only once, the Princetonian’s figure could be generated by as little as one student cheating in an exam room of 200 students.”

            The Honor Committee, which consists of the current and former undergraduate class presidents, issued a statement declaring that a poll it took last spring “found the level of cheating on in-class examinations at Princeton to be significantly lower” than the Prince’s estimate. The panel added that in its opinion, “the difference in the two figures is due to peculiarities in the polling techniques used rather than any change in the level of cheating since last spring.” But it declined to give the precise percentage it had come up with, pending the completion of a comprehensive report offering a detailed analysis of how the Honor System is working today and recommendations for improving it.

            The Honor Committee’s survey reportedly used a simpler form of the Randomized Response Technique, involving a single coin-toss and a dummy question, and it was conducted by telephone. This was said to provide more assurance that the respondents followed the poll’s instructions but — in contrast to the Prince’s procedure — it was more likely to err in the direction of underestimating the number of cheaters. One panel member let out that the difference between the results of the two surveys was greater than the combined total of their sampling errors (14 percent), which would put the Honor Committee’s finding under 20 percent, and another member revealed that it was above 15 percent. A non-randomized poll by the Prince two years ago put the figure at 12 percent (PAW, May 2, 1977).

            In any event, Princeton appears to be doing much better than most other universities. During its year-long study, the Honor Committee found that the available national statistics — which include all forms of cheating, not just on in-class exams — average about 80 percent. After comparing the situation at Princeton with that at other colleges, the panel concluded that the Honor Code is a more effective deterrent to academic fraud than any system of proctored exams. While acknowledging that the cheating problem at Princeton is serious, the committee members insist that it does not amount to a crisis. By and large, they feel the Honor System is working, although they plan to propose improvements in its mechanics that “might even help to reduce the level of academic dishonesty which does occur.”

            Meanwhile, the administration is waiting for the Honor Committee to complete its report before beginning an evaluation. “The Honor Code is a fundamental part of undergraduate academic life at Princeton,” Dean Girgus said, “and any suggestion that the Code has been violated — even by a small number of students, quite apart from more substantial numbers — is a matter of the most serious possible concern and deserves prompt, but also very thoughtful, attention.” She added, “Conversations with both faculty and students suggest that, while there undoubtedly are violations of the Honor Code, there is nothing resembling widespread cheating on examinations at Princeton. Nonetheless, the entire Princeton community clearly must do all that it can to prevent cheating in the first place, and to see to it that students who are accused of cheating receive prompt and fair hearings and, if convicted, appropriate penalties.”

* The number of respondents answering “yes” should comprise 25 percent of the cheaters polled plus 75 percent of the non-cheaters. Conversely, the number of students answering “no” should consist of 75 percent of the cheaters surveyed plus 25 percent of the non-cheaters. From these two equations, the number of cheaters can be calculated by multiplying the number of “no” answers by three, subtracting the number of “yes” answers, and dividing the difference by two. The percentage is then obtained by dividing that result by the total number of respondents.

This was originally published in the May 7, 1979 issue of PAW.