On the Theory
of Measurement
CIS343
The Numerological Fallacy
In a book on statistical analysis in the social sciences, the following definition of measurement was given:
“Measurement is the process of assigning
numbers according to a rule.”
Clearly
this is not sufficient for measurement.
For example, suppose I assign numbers in the range of 0 to 9 according
to this rule:
Assign to each person the last digit of their social
security number.
Absolutely
no measurement has occurred.
Measurement does not begin with numbers, but with a concept. We will refer to the idea that measurement
is simply the assignment of numbers according to a rule The
Numerological Fallacy.
The Priority of the Concept
What
is missing in the numerological fallacy is an understanding of this: In order
to have measurement, one must have a prior concept of what is to be
measured. Therefore, the
conceptualization precedes the design of a method of measurement. As we proceed to develop measurement
techniques, we use the underlying concept to test the applicability of the
proposed method. If, for example, our
proposed method for measuring weight tells us that the jockey who rode the
winning horse in the Kentucky Derby weighs more than most lineman for the
Philadelphia Eagles, we become highly suspicious.
Difficulties
As
we seek a measurement method for a concept we encounter two problems:
w It may be easy to find a
measure in some contexts, but difficult in others.
w We may be fuzzy in our
understanding of the concept we wish to measure.
We
look at each of these in turn.
Differing Contexts
Distance
is a concept about which there is no controversy. We begin with the basic method of employing a measuring object
(ruler, tape measure, etc.) that has been calibrated with an agreed upon
standard. We will refer to this as our base method. For short distances we can lay down this object and read the
measurement. For greater distances we
use mathematics to devise methods of measurement that can be calibrated to this
base method.
Problems
arise, however, when we begin to talk about distance in space. Some distances can be calibrated back to our
base method, but others cannot. We can,
for example, send radio signals to the moon and use the time that elapses
before the signal returns to calculate distance. Since this same method can be used on earth, it can be calibrated
back to our base method. But, what
about measuring distances to astronomical objects that are too far away for
this to work? At some point astronomers
use differences in what is called the red shift to measure distance. For more distance objects they employ a
diagram called the Hertzsprung-Russell diagram and use positions on this
diagram as a measure of distance.
A
legitimate question, at this point, is this: are we still measuring the same
concept? Unraveling this question
requires philosophical analysis. The
point is clear, however, that as the context of application changes, a concept
that was initially well understood may become murky.
Fuzzy Concepts
Other
concepts are fuzzy from the very outset.
Sports offers many examples. In
the ranking of NCAA football teams a coaches poll will differ from a
sportswriters poll which will differ from “computerized” rankings. Another example. Who is the better quarterback - the one who has the best quarterback
rating or the one who leads his team to Super Bowl victory?
Even
as we accept that some concepts are simply not amenable to clear cut
measurement, there are principles we can use in our search for a good
measure. In most cases we seek a scalar
measure, one whereby any quantity we obtain is either less than, equal to, or greater
than some other quantity. Underlying
scalar quantities is the notion of difference,
which is slightly more general. Here we
can apply the rubric of philosopher Gilbert Ryle: “A difference in order to be
a difference must make a difference.”
In other words, if we want to say that quantity A is greater than
quantity B, there should be a practical implication. I.e., it may be that something else stands in direct (or inverse)
proportion to the original measurement.
As a simple example, in the grocery store we will expect to pay more for
2 gallons of milk than for one gallon.
The
idea, then, if the original concept is fuzzy, to determine whether it can be
measured in a practical sense, we need to find a practical implication to
relate it to. We will refer to this as
the Difference Principle.
The Application Context
This
discussion of measurement arises from our desire to make more precise the
concept of locality. In applying the
Difference Principle to this, we will need to find some measure of locality
which can be related to an operating system performance difference. In other words, if process A has a greater
degree of locality than process B we want to be able to infer that in some
specific context process A will perform better than process B. Process A may execute more quickly than
process B, or it may have a better hit ratio in cache memory, or it may
generate less page faults. We will see
that finding that ideal measure of locality is not a simple task.
The
problem of finding a measure of locality and its appropriate application
context is explored further in “Measuring Locality” note and in the “Process
Characteristics” project. We may find
that different contexts require different methods of measurement, to be
applicable.