choice of methods of adjustment and it is therefore perhaps legitimate to raise the
question here. Again, at the risk of riding a hobby horse or flogging a dead one,
it would seem of value to look into just what we are supposed to be doing when
we «adjust» a block of triangulation, whether terrestrial or aerial. The word
«adjust» arose from the very special circumstances attaching to a geodetic trian
gulation. Unlike every other statistical problem of estimation that one is likely
to meet, that of «adjusting» a geodetic triangulation has the peculiarity that, for
every unknown there is one and only one «observation». We are referring here
to a figurai adjustment, of course. The unknowns are the angles to be found and
the single observations are the values accepted for each angle after the station
«adjustment». Apart from questions of economy, there are good statistical reasons
why there is no advantage in including in the figurai adjustment all the measures
actually made at each station, but whether this be so or not, it has the result of
giving us one «observation» for each unknown. This, in its turn, has the effect
of making the residuals between the finally calculated values of the angles and
the «observed» values equal in each case to the unique correction that must be
applied to an «observed» value to give the final value. From this circumstance,
which is very rare, if not unique, in statistics, arises the wholly incorrect and
misleading notion that observations are «adjusted» to give final values. One would
have thought that the temptation to use this word would have been resisted by
surveyors who are always very careful to ensure that field observations are never,
under any circumstances, tampered with. One has only to consider more common
problems to see how midleading the expression is. For example, what person,
having measured a length several times, would use the word «adjustment» for the
process of accepting the arithmetic mean as the best result, for it is clear that
nothing had been adjusted, the observations, as is proper, having remained as
they were when they were made. But if observations are not adjusted what
exactly is done? Quite simply, the observations, redundant and therefore in
practice discordant, are used to estimate the unknown or unknowns: the problem
is one of «estimation» and not of «adjustment». If the question were merely a
matter of words it would be idle to pursue or even to raise the subject; but it is
doubtful if anything can be merely a matter or words, and it is certainly not the
case here. The work «adjustment» at once conjures up in the mind’s eye the
problem of stretching, contracting and rotating the members of a redundant steel
framework so as to allow the structure to be assembled; and it is clear that, in
such a case, the more elements that can be manipulated the easier will the problem
become and the smaller the «corrections» to the original (incorrect) dimensions.
Such an analogy has no harmful effect on the calculation of a geodetic triangu
lation, because the unknowns are not at the choice of the computer, but in other
applications, such as in aerial triangulation adjustment, it is very easy to reach
the completely false conclusion that the smaller the corrections to the observations,
i. e. the smaller the residuals between the final and «observed» values of any
unknown, the better will be the answer; and, of course, these residuals can be
made indefinitely small by increasing indefinitely the number of unknowns to be
found. To take a very simple example, suppose we make two very accurate
measures of a length and correct these measures for temperature changes. Except
by chance we will obtain two different results and, whatever method of estimation
we propose, we will end up with two finite residuals. Suppose however we regard