78 Messrs. Paterson and Dudding on the 



Table V. — Giving the Probable Errors of the Various 

 Steps of the Cascade and Direct Comparisons. 



Comparison. 



Probable errors. 



Total 

 probable 

 error in 

 mean of 



set. 



Bench 

 error. 



Day-to-day 

 error. 



Error clue to 



differences 



between 



observers. 



Pentane lamp against Set 1. 

 Set 2 against Set 1 



P. cent. 

 001 



001 



o-oi 



001 



o-oi 



001 

 02 

 002 



P. cent. 

 005 



003. 



o-o3 o : 



004 

 0-03 

 002 



0-07 



0-06 



P. cent. 

 000 



015 



0-04 5 



012 



0-11 



004 



036 



0-22 5 



P. cent. 

 005 



0-15 5 

 0-06 

 013 

 011. 

 0-04 3 



0-37 



023 



„ 3 „ „ 2 ... 



„ 4 „ 3 



,, 5 „ 4 



„ 6 ,,5 



Total probable error by 

 cascade method Set 6 

 against Set 1 



Total probable error by 

 direct method, Set 6 

 against Set 1 





3. Intercomparison with other National Laboratories of the 

 Unit of Candle-power in White Light. 



Having fixed the value of the unit in white light by the 

 cascade method, it became of interest to know what order of 

 agreement existed between it and similar units evolved at the 

 National Laboratories in France, Germany, and the United 

 States. A set of 14 tungsten filament sub-standards with a 

 specific consumption of about 1*5 watts per candle was pre- 

 pared and sent to the Bureau of Standards, Washington. 

 A similar set was prepared and sent to the Physikalisch- 

 Technische Reichsanstalt, Berlin, and the Laboratoire Central 

 d'Electricite, Paris. The lamps were tested both before 

 leaving and after their return to England. At the same 

 time a set of lamps was prepared at the Bureau of Standards 

 and sent over here for test. Ordinary 100-Wt metal 

 filament glow-lamps were used in these comparisons, and 

 altogether about 18 months elapsed between the initial and 

 the final comparisons. During this period some of the lamps 

 appeared to change somewhat, and differences to the amount 

 of about 0*5 per cent, were observed. In order to eliminate, 

 •as far as possible, the effect of these changes, it was assumed 

 that when comparisons between any two laboratories were 

 made over a short period of time no intermediate change in 



