(452)
of measurements; and, secondly, by providing means of evaluating the reliabil-
ity of the conclusions drawn from the measurements.
“The present tendency in photogrammetric experimentation is to inves-
tigate individual aspects of a problem separately, by experimenting on single
factors. For instance, a new method of air triangulation would, according to
current tendencies, be tested by applying it to a strip or block of photographs
which was already triangulated by an older method, and then comparing the
results. One factor is thus changed, namely the method of triangulation. There
is, however, no a priori reason to believe that the conclusions thus derived
will remain valid when the circumstances of the experiment are changed (as
may well be the case in practical applications) since there is a large number
of factors affecting the result and which are likely to influence one another.
There is therefore a real need for primary experiments which include a number
of factors, and are designed to give a clear picture of the relevance of these
factors to the main subject of the investigation. This is the basic idea underlying
the so-called factorial schemes of experimentation.
“The main features of the factorial design of experimental investigations
are:
a) Several factors are experimented upon in one experiment; and
b) the variant under consideration (e.g. the differential shrinkage in an
experiment on dimensional stability of film, or a measure of accuracy in re-
search in air triangulation) is measured at every combination of the conditions
of the experiment. (There are many multi-factor schemes where this last con-
dition is relaxed for economy at the minimum loss of information; but we need
not go into this here, to avoid blurring the general approach by details).
“The technique of the Analysis of Variance carries out this procedure very
effectively, as demonstrated in the analysis of the data on film shrinkage in
Section 3.1 of the full text of this communication.
“In fact, experiments planned in accordance with factorial schemes are
the most efficient, in the sense that they give the maximum of information
that can possibly be obtained from a certain number of observations. There is
no great difficulty in analysing experiments thus planned, since suitable
methods of analysis are available in statistical literature. It is necessary, however,
in order to secure correct interpretation of the results of such analysis, to give
careful consideration to all the various stages in which an experiment passes.
“The appendix of the paper contains some suggestions bearing on some
points which have been frequently emphasized by statisticians in somewhat
more general terms, and to others which appear to me to have special import-
ance in our field of research.
“In general, an experiment passes through tive stages: (1) specifying its
purpose; (2) planning; (3) conduct of the work; (4) analysis of the data; and (5)
presentation of the experiment.
“The objective of the experiment must, ot course, be clearly defined.
Whenever practicable, however, an attempt should be made to cover issues of
more general interest. This may be found possible without much additional
effort.
“An important point about the planning is to choose suitable variates.
The variates are the quantities which will be measured or computed and which