Prof. Edgeworth's Problems in Probabilities. Ill 



ment that, whereas the (uncorrected) Mean Deviation for the 

 whole set of 400 pairs is 67, the corresponding determinations 

 for the first, second, and third batches of 133 papers were 

 respectively 63, 67, 72. I have similarly verified other 

 results obtained by the same or a similar method. Thus the 

 Mean Deviation for marks given by two examiners to fifteen 

 papers in Greek Prose is 25 per cent.; for marks given by 

 the same pair of examiners to thirty pieces of Latin Prose, 26 

 per cent. For sixty pieces of Greek and Latin Prose and 

 Verse composition (including some of the Latin Prose but 

 not any of the Greek Verse before mentioned) I obtained, — 

 by a more summary, but in the particular instance at least 

 sufficiently safe, method* — again 25 per cent. 



When the statistics are in the form of marks given by 

 several examiners to the same piece of work, I have extracted 

 the Mean Error according to the usual rules ; and then mul- 

 tiplied the coefficient by V2 in order to obtain the Mean 

 Deviation as above defined. In the only case in which I 

 have been able to compare the two methods of determination 

 the results yielded are fairly consilient. I refer to Latin 

 Prose Composition, for which, according to the first method, 

 I extracted from thirty pairs of marks given to as many 

 pieces of prose the Mean Deviation 26 per cent. By the 

 second method I obtained from twenty-eight marks given to 

 the same piece of prose by as many highly competent ex- 

 aminers the Mean Deviation 20 per cent. — of the average 

 of the twenty-eight marks ; which, being four fifths of the 

 maximum, is not exactly comparable with the general average 

 referred to in the first method. 



I annex a summary statement of the results obtained by 

 one or other of those methods f : — 



* Using the formula : Mean Deviation (in the sense above defined) 

 = V ~ Average Deviation ; which relation had held good for a great 



number of marks given by the same examiners in a variety of Classical 

 subjects including Composition. 



t These computations derive some confirmation from an experiment 

 which Mrs. Bryant, D.Sc, of the North London Collegiate School, has 

 communicated to me. Having examined forty Geometry papers, she re- 

 examined them after some weeks. The discrepancy between the two 

 sets of marks (corrected for a certain difference of scale) proves to be 

 only 12 per cent. For further remarks on the Table see the companion 

 paper. 



Phil May. S. 5. Vol. 30. No. 183. August 181)0. N 



