Always be suspicious of statistics without references, especially ones on such a difficult subject as self-reported behavior. I'm a medical researcher, and we spend a lot of time ensuring that samples are selected properly, non-response bias is either reported or accounted for, and questions for patients are phrased clearly. It is very, very easy to screw these things up.

It is entirely possible that this sheet honestly reports on the results from one study in the 1940s and another performed recently, but we should still be very careful how we interpret the numbers.

I can't find any data from the 1940s, but data from the 1960s indicates that between 21% and 78% of the students in college admitted to cheating, depending on how 'cheating' was defined (see table 4.13 in Bowers, 1964). This illustrates how much of an effect the type of question can have on the response.

• Questions with relatively high percentages of "yes" responses (>30% for each) included: "copying a few sentences of material without footnoting", "getting questions or answers from somebody who has already taken the same exam",* or "copying answers from a text instead of doing the work independently."
• Questions with very low percentages of "yes" responses (<5%) included: "Taking an exam for another student", or "Arranging with other students to give or receive answers by use of signals".

Unless the people who put this together were careful to ensure that the response levels were reasonably high in both eras and that questions asked in the 1940s were very close to the questions asked in more recent studies, any comparison between the times is very likely to be useless.

*As a side note, a few decades ago when I went to school, the act of obtaining a copy of an old exam from the library was not considered cheating in some of my classes. Even if a question turned out to be exactly the same, the answer might be very, very different if the field had advanced and new statistical tools had been developed.