Croxton, F. E. & Stryker, R. E. (1929). Bar charts versus circle diagrams. Journal of the American Statistical Association, 22, 473-482.

(Conclusion.) (...) It is not possible to make a general statement to the effect that either the bar or the circle was in all cases more desirable for the presentation of component parts. In conclusion a few important points may be noted:

1. For presenting a 50-50 relationship the circle was undoubtedly superior to the bar.
2. For presenting the percentage 25 and 75 the circle was clearly preferable to the bar. This was especially true when the circle was constructed with the 25 per cent section centered at 1:30 o'clock (...) but also obtained when the 25 per cent section was centered at 6 o'colock (...).
3. In addition to the 50-50 and 25-75 relationships four other pairs of two-art diagrams were included. For these four cases judgment based upon the bar seemed to have been more accurate, since in two instances there was a significant difference in favor of the bar, in one instance a significant difference in favor of the circle, and in one instance a difference that was not significant in favor of the circle.
4. For the three pairs of diagrams of three parts the circle appeared to have been preferable, there being two cases of a significant difference in favor of the circle, and one case of a difference in favor of the circle, but not significant.
5. The four-part diagrams, of which there were three, yielded the same results as the three-part diagrams. There were two instances of a significant difference in favor of the circle, and one in favor of the circle, but not significant.
6. In the case of the single pair of diagrams of five parts no choice could be made, since the difference between the mean errors, while in favor of the bar, was not significant. It was apparently true that the greater the number of sections into which a diagram was divided, the greater the error that a reader was apt to make.
7. Of a total of 21,789 estimates of diagrams, 466, or slightly more than 2 per cent, failed to total 100 per cent. This was in spite of the fact that the observers were carefully instructed that the parts of each diagram totaled 100 per cent. This error was present to a greater extent in the case of the multiple part diagrams, than in the case of the simpler ones.