In one of the lab courses I took as an undergraduate, I remember that the professor noted while discussing some statistical test (almost certainly chi-squared) that one could use it to show that a lot of early scientists (prior to the development of rigorous statistical analysis) had data that were too good to be realistic, and thus (with very high probability) were either deliberately falsified or the result of poor experimental design. He in particular cited Mendel as an example of a scientist whose data were far too good to be believable.

Wikipedia has a section related to Mendel's case specifically, and some discussion as to the possibilities. I'm more interested in the general case.

Is it true that a large fraction (relative to today) of prominent scientists before the 20th century presented data to support their conclusions which were statistically too good to be true? And if so, how did they avoid being wrong about a lot of their results?

  • 3
    $\begingroup$ Are you interested in data which is statistically implausible (as your question at the end suggests), or in cases where the data was really faked? I think that most prominent scientists didn't resort to outright fakery, by which I mean that the data was freely invented. $\endgroup$
    – Felix
    Oct 30, 2014 at 12:57
  • 1
    $\begingroup$ "common" and "large fraction" are subjective. I recommend changing the language to "more than today" or similar wording. $\endgroup$
    – BMS
    Oct 30, 2014 at 15:56
  • 3
    $\begingroup$ Apparently, this goes far back. Some of Ptolemy's astronomical "observations" were fabricated by extrapolating Hipparchus's data, Richard Newton even called him "the most successful fraud in the history of science". adsabs.harvard.edu/full/1980QJRAS..21..253G $\endgroup$
    – Conifold
    Oct 30, 2014 at 19:26
  • 1
    $\begingroup$ @Felix I don't think statistics alone can hope to tell us whether the data was statistically implausible because the scientists simply made it up or because of poor experimental design. Hence, I'm not specifying explicitly in the question, but I'm really interested in both cases. $\endgroup$
    – Logan M
    Oct 30, 2014 at 23:32
  • 2
    $\begingroup$ the result for Buffon's needle experiment by Mario Lazzarini (en.wikipedia.org/wiki/Buffon%27s_needle) was probably cooked up. $\endgroup$
    – mau
    Nov 6, 2014 at 17:15

2 Answers 2


The notion of "rigged data" evolved with time. Some ancient scientists are accused (by modern scientists) in rigging of the data.

One notable example is Ptolemy. I do not want to discuss here the accusation of Ptolemy by Robert Newton, but here is another well-known example.

In his Optics, Ptolemy gives a table of refraction. It looks like he measured refraction of the light ray passing from the air to other media (water, glass). He gives a little table, in such a way that a modern reader may conclude that these are experimental data. Some even consider this the oldest surviving example of an experiment in physics. However they are most likely made up. The data are inconsistent with the Snell law. If you plot the numbers they better fit a parabola rather than a sinusoid (as it should be). It is clear from the text that Ptolemy had some theory about refraction (incorrect theory), and his table was most likely computed according to this theory.

On the other hand, he never states unambiguously that he actually performed this experiment, and does not give any detail.

The most serious accusation against Ptolemy is that he forged his star catalog. That is he took the 200 years old catalog of Hipparchus, and corrected the data for precession. And used the wrong constant of precession which was not known exactly at that time. There is a large controversy about this among historians of astronomy, and as I understand, there is no consensus.

EDIT. Let me add that I am not aware of any famous stories of forgery and rigged data in physics and mathematics. None prior to 20 century and VERY few in 20 century. Unlike in humanities like archeology or "finding some ancient manuscripts" which was quite common in 18 and 19 centuries.

  • 1
    $\begingroup$ Not that few in 20th century, unfortunately. One famous piece of fake data, published in a very respected peer-reviewed medical journal, led to anti-vaccination movement that in turn caused preventable outbreaks and deaths. Wakefield, MMR vaccination and autism, The Lancet $\endgroup$
    – Michael
    Jun 13, 2019 at 20:47

Charles Babbage (the grandfather of the computer) in the Reflections on the Decline of Science in England, and on Some of Its Causes (1830) writes about misconduct in science and criticizes British science for lack of disciplined data recollection. He provides an insight of how common it was.

He divided fraud in four practices:

  • Hoaxing: falsifying the object of study, creating creatures or phenomena that do not exist. Not so commonly practiced.
  • Forging: fabricating data. Not so widespread, but much more damaging because it can be difficult to notice.
  • Trimming: taking bits of data out that did not fit. More extensively used.
  • Cooking: cherry-picking observations, conclusions that do not follow from the facts, removing unfavorable results. Widespread among reputable scientists, who make a name out of their results.

He also list various examples of fraud throughout the book. He emphatically calls out problems with Edward Sabine's extraordinary accuracy with pendulum calculations for which Sabine had received some grant. However, Babbage's accusations did not affect Sabine's career as he became later the president of the Royal Society.

Babbage also cites an example of data forgery from the 18th century

The observations of the second comet of 1784, which was only seen by the Chevalier D'Angos, were long suspected to be a forgery, and were at length proved to be so by the calculations and reasonings of Encke. The pretended observations did not accord amongst each other in giving any possible orbit. But M. Encke detected an orbit, belonging to some of the observations, from which he found that all the rest might be almost precisely deduced, provided a mistake of a unity in the index of the logarithm of the radius vector were supposed to have been made in all the rest of the calculations. ZACH. CORR. ASTRON. Tom. IV. p. 456.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.