MAR. 4, 1923 CRITTENDEN: MEASUREMENT OF LIGHT 83 
Because of the difficulty of obtaining reliable and reproducible 
results, the practice early developed of preserving values on certain 
steps by the calibration of color filters, such as blue glasses, which, 
interposed on one side of the photometer, would equalize the color 
of two different lamps. Another procedure has been to make many 
observations to determine values for secondary or derived standards 
of the different kinds of lamps to be measured. 
The extreme case of the latter procedure is perhaps that ene ed 
by the British National Physical Laboratory where six sets of electric 
standards were carefully calibrated by equality-of-brightness obser- 
vations covering the range from the color of the pentane flame stand- 
ard up to a low-efficiency vacuum tungsten lamp. This was very 
carefully done by a number of experienced observers and affords a 
good example of the essential uncertainty of the procedure. The 
divergence between different observers, of course, increased on the 
successive steps, and a difference of nearly 3 per cent in the final 
values developed between extreme observers. Measurements were 
also made in which the same observers jumped directly from the 
lowest to the highest efficiency within this range. The average 
candlepower of the higher efficiency lamps by the two methods agreed 
within 0.3 per cent, but it is significant that in the direct single step 
with larger color difference, the observers were grouped more closely 
than on the final results obtained step by step. The total range of: 
the results and the average deviation from the mean in the direct 
measurement are only six-tenths as great as in the step-by-step or 
cascade measurements. When it is considered that the range of 
efficiencies covered is only from 2 to 6.5 lumens per watt, whereas 
present day gas-filled tungsten lamps fall considerable above 20 
lumens per watt on the same scale, it is evident that the results are 
not satisfactory for precise measurements in which an accuracy bet- 
ter than 1 per cent is expected. 
At the Bureau of Standards a somewhat different course was fol- 
lowed in that instead of establishing different sets of standards for 
different efficiencies, values for a group of tungsten lamps were de- 
termined over a considerable range of efficiencies and the variation 
expressed in curves and equations. Furthermore, this work devel- 
oped the fact that for all the ordinary sizes of vacuum tungsten lamps 
the same equations would apply. In other words, the variation in 
candlepower and efficiency as a vacuum tungsten lamp is changed 
in voltage is a definite property, presumably because it represents 
in a fairly simple way the properties of radiating tungsten at given 
