A. 3,
A.5.
28
plan which can be modified to suit the purpose of the experiment within
the available facilities and time. Due consideration must be given to (1)
the sources of variation that are desired to be included in the experiment
as controlled factors; (2) the levels at which these factors are to be con-
trolled; (3) the approximate levels of the uncontrollable but determinable
factors; (4) a practicable process of randomization or of balancing to
eliminate possible bias of results due to the uncontrollable and undetermi-
nable factors; and (5) the number of replications deemed appropriate.
The simple factorial scheme may however be found to involve too many
observations. This difficulty can usually be overcome in the following
Ways:
(1). By applying the process of confounding, which is an ingeneous way
of reducing the number of observations without impairing much the
scope of the work.
(2). By decreasing the number of replications.
(3). By reducing the number of levels of all or some of the factors.
A.24. Checking the Design.
The final design of the experiment should be carefully inspected before
starting the observations. Should there be any doubt as regards the effi-
ciency of the design, a statistician had better been consulted before pro-
ceeding any further, since no more information can be obtained from an
experiment than it is capable of giving in virtue of its design.
CONDUCT OF THE EXPERIMENT.
Since photogrammetric experiments are not normally carried out by the
people who design them it is essential that:
(1). The design shall be translated into detailed instructions covering all
the procedures of gathering the data and of subsequent calculations.
(2). Under no circumstances shall these instructions be deviated from
without consulting the planners who should make the necessary ad-
justments in the design of the experiment.
A.31. Rejection of Observations.
A criterion for the rejection of outlying observations should be included.
It should be noted in the preparation of this criterion that statistical
treatment of data allow for the appearance of outlying observations.
ANALYSIS OF THE DATA.
To each design there corresponds a method of analysis.
PRESENTATION OF THE EXPERIMENT.
A.51. Presentation of the Experimental Data.
(1). Data are better given in numerical tables than on graphs, to allow
checking and further analysis by the reader. Graphs may of course be
used to emphasize important features.
(2). It is good practice to avoid the use of terminology that might mislead
the uninitiated. In particular, the following would better be avoided:
(a) the use of ‘mean error’ and ‘mean square error’ as short for ‘root
mean square error’; (b) the use of ‘probable error’, specially when the
number of observations from which it is derived is small; (c) attaching
+ to the root mean square error.
(3). Enough significant figures should be retained in order that errors of