90 



DESIGN AND USE OF INSTRUMENTS 



which the following results were obtained. 

 On the tachometer dial he found that the 

 small i^-inch numbers marking every 100 

 rpm from 4 to 35 could be dropped, leaving 

 only the seven larger numbers which marked 

 each 500 rpm. Accuracy was unaffected. 

 A similar elimination of seven small |-inch 

 numbers from the manifold pressure dial, 

 leaving just the seven large ones, left per- 

 formance unchanged in tungsten light and 

 improved reading accuracy in ultra-violet 

 illimiination. No loss attended the drop- 

 ping of several numbers from the climb and 

 dive indicator, even though the subjects in 

 the experiment expressed preference for the 

 fully numl)ered scale. When, as in one in- 

 stance, results significantly favored the more 

 completely numbered dial (see Part D of 

 Fig. 10), they occurred in this direction only 

 for an exposure of 1.5 seconds and not for 

 0.75 seconds where, indeed, the simpler num- 

 bering had a significant advantage under 

 ultra-violet illumination. This difference of 

 result for these two different exposure times 

 could be interpreted as meaning that the 

 amount of usable scale numbering may vary 

 with the kind of reading being made; that 

 for reading "at a glance," infrequent num- 

 bering with easily comprehended "round" 

 values may be desirable, while for more care- 

 ful and leisurely reading, frequent numbering 

 may be an aid. A test of this conjecture 

 with longer exposure times or in a "read at 

 your own pace" situation is indicated. 



No research appears to have been done on 

 numbering arrangements for instruments 

 used imder low illumination. 



Non-uniform Scaling. Instrument de- 

 signers are sometimes forced to use scales 

 with non-uniform graduations. Some in- 

 struments are mechanized in a way that 

 pointer movement is non-linear. Othei's re- 

 quire readings in such widely different parts 

 of the scale range that non-linear scaling is 

 one answer to the problem of getting the 

 entire scale into the available space. No 

 data are yet available, however, on how 

 reading accuracy is affected by scale non- 



linearity, or on the problem of how gradu- 

 ation and numbering changes should proceed 

 along such scales if reading errors are to be 

 minimized. 



Broken or divided scales. The need for a 

 long range of scale values commonly suggests 

 the use of broken scales — that is, of a coarse 

 scale coupled with one or more fine scales. 

 This has been mechanized in a variety of 

 ways: as a large dial with a small insert dial, 

 as a pair of large dials, as a single dial with 

 several pointers. But no matter what form 

 the broken scale has taken, gross inaccuracies 

 in making certain readings have always re- 

 sulted. In field artillery these errors have 

 taken the form of the much publicized "100 

 mil error" (9, 88). In aircraft operation 

 they have occurred with serious frequency 

 in altimeter reading as errors of 1,000 feet or 

 10,000 feet. At best, the broken scale is 

 difficult to read because the operator has to 

 combine several separate readings to get his 

 final value, but the particular difficulty 

 which produces large, systematic errors arises 

 when the coarse scale registers just below but 

 neai" a value which would imply a full circuit 

 of the fine scale. On a coarse and fine scale 

 combination, for example, where the fine 

 scale registers units and tens up to 100 and 

 the coarse scale registers hundreds, a setting 

 of 397 is often read as 497 because the coarse 

 scale pointer is read, not to the next lowest, 

 but to the nearest, number, which is 4. 

 Fitts and Jones (27) found this type of error 

 with reference to altimeter reading to be the 

 most common of all instrument reading 

 errors reported by pilots. Although specific 

 data on the use of Navy twin dial instru- 

 ments are lacking, it would appear that the^^^ 

 common design with two dials which rotate 

 in opposite directions past a common index 

 mark would be subject to various reversal 

 errors as well as to the errors just described. 



Fcjllowing the Fitts-Jones study, Grether 

 (31, 32) conducted a pencil-and-paper study 

 of alternative ways of representing the 

 broken scale altitude data. The multiple- 

 pointer method now in use was very poor, 



