76 DONALD C. BARTON 
the effect of single seriously abnormal gradient arrows, or less com- 
monly, pairs of gradient arrows, he attempts to eliminate such errors 
in the first calculation of relative gravity; in the subsequent ad- 
justment of the traverse or traverses, somewhat commonly much 
of the error of closure may be eliminated by a slight change in his 
weighting of those few irregular gradient arrows. The variation of 
the gradient from station to station in many places is not linear and 
the geophysicist has a certain latitude in his subjective estimation 
of the probable variation of gravity between the two stations. The 
error of closure in many traverses can be almost eliminated by taking 
those slightly different alternative choices in the spacing and the run- 
ning of the isogams. The subjective weighting of observations, of 
course, is open to danger of error and in the hands of an inexpert 
geophysicist, leads to much more erroneous results than the auto- 
matic mathematical distribution and adjustment of the error. But 
the writer believes that the expert interpreter can obtain better 
accuracy by exercising subjective judgment in throwing as much of 
the errors of closure as possible into the weak places in lines rather 
than by automatic mathematical distribution of the errors. The 
“probable error” of the determination of relative gravity in those 45 
traverses presumably, therefore, is slightly less than +0.25 mille- 
dyne per 10 kilometers of double traverse and less than +0.35 mille- 
dyne between places ro kilometers apart, airline distance. . 
That ‘probable error,” 0.25 milledyne per 10 kilometers of double 
traverse or 0.35 milledyne per 1o kilometers of airline distance, pre- 
sumably is characteristic of most good torsion balance surveys in 
what a torsion balance operator would call good, fair, and slightly 
poor torsion balance terrane. Those 45 traverses lay in all types of 
terrane in the Gulf Coastal Plain of Texas and Louisiana and a few 
of them were on the llanos of Venezuela and elsewhere than in the 
Gulf Coastal Plain. Less than one-fourth of the traverses were in the 
Coastal Prairies of Texas and Louisiana, which are especially favor- 
able for a low ‘“‘probable error” of observation. The 45 traverses, 
therefore, are fairly characteristic of torsion balance work in areas 
which are commonly regarded as good, fair, or slightly poor for 
torsion balance observations. Mathematically the same grade of 
accuracy could be maintained by use of a sufficiently close net of 
stations to compensate the “‘probable error’ in the individual ob- 
servations, but the number of stations necessary increases as the 
square of the increase in the ‘‘probable error” of the individual ob- 
406 
