the standard error of the systematic errors can be determined
from the special law of error propagation. In general the
standard errors can be determined from the product of the
standard error of unit weight and the square root of a
weight number, which also can be directly determined from
the normal equations. For a determination of the standard
error of arbitrary linear functions of the parameters, for
instance the combined corrections, the general law of error
propagation generally has to be used. Attention must then
be paid to possible algebraical correlation between the
parameters, expressed by the correlation numbers. The
weight and correlation numbers are usually summarized
in a matrix, frequently called the matrix of weight coeffi
cients.
With the aid of the standard errors of the parameters
and of functions thereof, it is possible to judge the signi
ficance of the differences between the parameters and of
functions thereof on one hand and the ideal values on the
other. In this way tolerances for the systematic errors and
functions thereof can be established.
Because the standard error of unit weight in connection
with calibrations can be used as an indication of the basic
accuracy of the device under calibration, the magnitude
of the standard error of unit weight can be used for a judge
ment whether or not the obtained accuracy is in agreement
with some prescribed value to be used as a characteristic
of the device in question. According to the resolution No.
II: 6 from the Lisbon Congress, quoted above, such values
shall be furnished by the manufacturers in order that the
instrument users shall be able to judge the quality of the
actual device under calibration and in particular decide
when repair or adjustment must be made.
2.3. The Concept of Weight
Above, all measurements have been assumed to be of
equal weight, i.e. of the same basic quality. If, for some
reason, different quality of the measurements in the same
set is to be expected, weights can be introduced and used
in the computations in a well known manner. When
weights are introduced before adjustment computations
they are generally called a priori weights. Weights can also
be determined from the results of the adjustment and then
have the character of a posteriori weights.
Examples of a priori weights due to physical circum
stances are weights of image coordinates due to the fact
that the photographic image for evident reasons never is a
mathematical plane. If such coordinates are measured
orthogonally, the lacking flatness will have different
influence in the center of the image in comparison with the
edges. Practical experiments have also indicated that this
assumption is justified and that there can be considerable
differences in the geometrical quality of image coordinates
depending on the location of the points. There is probably
a weight function for each camera and photograph and the
weights could be expressed as functions of the radii. This
is a question that should in particular be noted in connec
tion with calibrations of cameras and photographs under
operational conditions.
A posteriori weights are to be referred to the quality of
results of adjustment procedures and can be determined
with the laws of error propagation.
The theoretical definition of weights of measurements
usually refers to the quality of the measurements as deter
mined from redundancies. Ordinarily weights are defined
as inversely proportional to the variances. Because, in
principle, variances can refer to the concepts of precision or
accuracy respectively, weights can also be referred to either
of the two concepts mentioned, depending upon how the
variances were determined.
In the case of precision, the variance of the mean can be
arbitrarily decreased through repetitions only.
In the case of accuracy, the residual variance cannot be
decreased in the same simple manner because the variance
will always be composed of another type of error popula
tion, containing also systematic errors of such a magnitude,
however, that they cannot be individually determined.
Consequently there is reason to distinguish between
precision weights on one hand and accuracy weights on the
other. The precision weights refer to the squares of the
standard deviations (of a single measurement or of func
tions of such measurements, e.g. the mean of repeated mea
surements), i.e. precision variances. The accuracy weights
refer as a principle to the squares of the standard errors of
unit weight of measurements after the adjustment or
to the squares of standard errors of functions, i.e. accuracy
variances.
2.4. Summary
The principles, summarized above, were in more detail
sent to all national societies of photogrammetry, organiza
tions, and individuals for critical examination. A number
of answers were received, the majority of which agreed
with the proposals. Only in a few cases some objections
were noticed, mainly of formal nature.
Those who did not answer were regarded as being in
agreement, according to usual principles and especially
emphasized in a letter, sent to all addresses in good time
before the dead line, Oct. 1, 1966.
Hence, the following may briefly be summarized as the
opinion of the national societies of photogrammetry.
1. Distinction should be made between the two basic
concepts and terms of precision and accuracy according to
the definitions in Ref. 2: 1, quoted from statistics.
In brief the following definition, quoted from Eisenhart,
Ref. 2: 2, can be used:
Precision refers to closeness together,