eom 
C 
) 
CO ct — 
- 
  
  
   
which are the first differences of the column E. When these are worked out as in 
Table 3 it becomes obviously wrong to conclude that “departures from the height 
profile should not exceed a maximum of about 15 metres”. 
  
  
  
  
  
  
Table 3 
1—2 243 13-4 4-5 
Sttipi \— 2 — 4 0 +20 
2. |-t16 — 2+ 3 +. 8 
3 |—4 — 1 . : 
249. 11485" 4 
5 |——12 — 9 | +17 — 7 
6 | +20 —14 | + 9 —12 
2.222. An error of a more subtle type occurred in a recent publication dealing 
with air triangulation (5). 
To provide evidence to support the use of a reseau ruled on the camera's 
register glass, the root mean square error of evaluating want of correspondence 
y-parallax) was estimated from a number of non-reseau and reseau air photo- 
graphs; and the two estimates were compared. The former was 0.032 millimetre, 
and the latter was 0.016 millimetre. This would indeed be a considerable impro- 
vement requiring no statistical test to establish its significance. These mean square 
errors were not of course observed quantities. They were derived from the residual 
want of correspondence after relative orientation. The fact was overlooked that 
relative orientation was not calculated in the same way. The two methods used 
were shown elsewhere (6) to differ enough to account for the larger proportion of 
the apparent improvement. The contribution of the reseau was shown to be a 
twenty per cent reduction of error instead of a fifty per cent reduction. This 
improvement required a statistical test to establish its significance. 
3. 
Such shortcomings as I have pointed out in the preceding Section 2 may 
appear at first sight to be merely unfortunate errors of analysis, or defects of 
presentation, but have little bearing on the general approach to the design of the 
experiments. The link will, however, become much clearer when they are looked 
at as limitations to the usefulness of the experiments; for it is on the design of an 
experiment that the amount of reliable information that can possibly be obtained 
from it will ultimately depend. It is therefore essential that the experiment be so 
designed to ensure sound and unambiguous conclusions. This can only be achieved 
by working out, in advance, the details of the procedure of gathering the data and 
the methods of their analysis, as well as the interpretation of all possible results. 
Now, if these requirements were effectively covered, no such errors could be 
incurred. 
3.1. The power of the modern approach to experimentation may be demons- 
trated by a discussion of the data on the long term shrinkage of air films referred 
to in. 2.1. 
13