I have chosen ten well-educated, and in most cases scientifically trained individuals, and determined for each the time necessary for the writing of the Roman and the Arabic numerals from 1 to 100 and the number of errors made, also the time necessary for the reading of the Roman and the Arabic numerals from 1 to 100 when they were irregularly arranged so that the reader did not know what order to expect. In all cases the number of errors made unconsciously was recorded. These measurements furnish the following startling averages: It takes three and one third times as long to write the Roman numerals from 1 to 100 as the Arabic, and the chance of error is twenty-one times as great; it takes three times as long to read the Roman numerals from 1 to 11O as the Arabic, and the chance of error is eight times as great.
In case of a quick and accurate mathematician, whose familiarity with the Roman system
surpassed that of most of the individuals tested, the results were: time for writing
Arabics, 107, errors, 0; time for writing Romans, 357, errors, 5; time for reading Arabics, 62, errors, 2; time for reading Romans, 131, errors, 5. For one well-trained scientist, who has cause to use the Roman system almost every day, the number of errors in the rapid reading of the Romans was 151 These figures certainly indicate the desirability of using the Arabic system wherever there is no urgent need for the simultaneous use of two or more systems of numerals. Even if there were no saving of time and strain by the avoidance of the cumbersome Roman symbols, the far greater accuracy gained by the use of the Arabic system should at once settle the matter for all scientists.
Interesting results. I would like to note two things, however. The Roman system is slightly more aesthetically pleasing and even more appropriate in certain cases. In books, it allows to have separate pagination for the introductory part etc. Also, from the point of view of the information theory, the system that intrinsically involves using more symbols should be more redundant and less prone to transmission errors. I think that, above all, this experiment proves the need to learn Roman numerals and to develop better control over them.