20.08.2013 Views

Suppes and Zinnes - basic measurement theory.pdf - Ted Sider

Suppes and Zinnes - basic measurement theory.pdf - Ted Sider

Suppes and Zinnes - basic measurement theory.pdf - Ted Sider

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

22 BASIC MEASUREMENT THEORY<br />

The answer to the second problem, the uniqueness problem, for pointer<br />

<strong>measurement</strong> has been the source of some confusion. The uniqueness<br />

of the readings is determined by the uniqueness of the corresponding<br />

fundamental or derived numerical assignment, not, as might appear, by<br />

the method of calibrating the pointer instrwnent. The fact that the pointer<br />

instrument for measuring mass gives a ratio scale is not because of the<br />

equal spacing of the divisions on the dial; it would be quite possible to<br />

use a nonlinear spring <strong>and</strong> have the divisions of the dial unequally spaced<br />

without altering the scale type, On the other h<strong>and</strong>, neither is it because<br />

of the fact that two points (e.g., 32 kg <strong>and</strong> 212 kg) are generally fixed or<br />

determined by the caiibration procedure. Suppose, for example, the<br />

fundamental <strong>measurement</strong> of mass yielded only an ordinal scale <strong>and</strong> the<br />

extension of the spring were monotonically related to mass. Then,<br />

although precisely the same calibration procedure could be carried out,<br />

that is, two points could be fixed <strong>and</strong> the dial divided into equally spaced<br />

divisions, the resulting readings of the instrument wonld nevertheless be<br />

only on an ordinal scale. Thus the scale type of a pointer instrument<br />

derives directly from the scale type of the corresponding fundamental or<br />

derived numerical assignment. As a corollary to this, it follows that<br />

merely inspecting the divisions on the dial of a pointer instrument, or even<br />

observing the calibration pfocedures of the instrument, does not enable one<br />

to infer the scale type of the <strong>measurement</strong>s obtained from the instrument.<br />

3. EXAMPLES OF FUNDAMENTAL MEASUREMENT<br />

In this section some concrete examples of different empirical systems<br />

are given, <strong>and</strong> in each case solutions to the representation <strong>and</strong> uniqueness<br />

problems are exhibited. The empirical systems themselves are not<br />

necessarily of interest. Many are too restrictive in one way or another to<br />

have direct application to significant psychological situations. These<br />

systems, however, permit bringing into focus some of the essential aspects<br />

of <strong>measurement</strong>.<br />

The proofs establishing representation <strong>and</strong> uniqueness theorems are<br />

generally long <strong>and</strong> involved. For most empirical systems, therefore, we<br />

shall have to content ourselves with simply stating the results. This means<br />

that for each empirical system described a full numerical system is stated,<br />

<strong>and</strong> then, more particularly, at least one subsystem which is a homomorphic<br />

image of the empirical system is given. The uniqueness problem is<br />

answered by giving the relationship between any two subsystems (of the full<br />

system) that are homomorphic images of the empirical system, Two<br />

proofs are given in Sec. 3.1 <strong>and</strong> 3.2. The first proof (Sec. 3.1) is relatively<br />

. ,-Ji'<br />

EXAMPLES OF FUNDAMENTAL MEASUREMENT<br />

simple <strong>and</strong> it will serve to introduce some of the necessary terms <strong>and</strong> ideas<br />

that are generally encountered. The second proof (Sec. 3.2) is considerably<br />

more difficult so that it will give some notion of the complexity of the<br />

representation problem.<br />

There is, as was indicated previously (Sec. 1.1), some degree of arbitrariness<br />

in the selection of a full numerical system for'a given empirical<br />

system. In each case other full numerical systems could have been chosen<br />

<strong>and</strong> a different representation theorem established. Thus it should be<br />

emphasized that, although the representation problem is solved for each<br />

empirical system by exhibiting one homomorphic numerical system having<br />

certain reasonably natural <strong>and</strong> simple properties, the existence of other,<br />

perhaps equally desirable, nwnerical systems is not ruled out.<br />

3.1 Quasi-Series<br />

Nominal <strong>and</strong> classificatory scales are somewhat trivial examples of<br />

<strong>measurement</strong>s so that we shall consider first an empirical system leading<br />

to an ordinal scale. The empirical system to be described is one that might,<br />

for example, be applicable to a lifted weight experiment in which subjects<br />

are given the weights in pairs <strong>and</strong> are instructed to indicate whether one<br />

weight of each pair seems heavier or whether they seem equally heavy.<br />

Some useful definitions are as follows. A relational system 'lf = (A, R)<br />

consisting of a set A <strong>and</strong> a single binary relation R is called a binary<br />

system, A binary system (A, I) is called a classificatory system if <strong>and</strong> only<br />

if I is an equivalence relation on A, that is, if <strong>and</strong> only if I has the following<br />

properties:<br />

1. Reflexive: if a E A, then ala;<br />

2, Symmetric: if a, bE A <strong>and</strong> alb, then bla;<br />

3. Transitive: if a, b, c E A <strong>and</strong> if alb, blc, then ale,<br />

where a E A means that a is a member of the set A. A common example<br />

of an equivalence relation is the identity relation =. In psychological<br />

contexts it is convenient to think of the indifference relation or the "seems<br />

alike" relation as an equivalence relation, although there are many cases<br />

in which the transitivity property fails to hold,<br />

We can now define a quasi-series (Hempel, 1952).<br />

Definition L The relational system 'lf = (A,/, P) is a quasi-series if \<br />

<strong>and</strong> only if<br />

L (A, I) is a classificatory system;<br />

2. Pis a binary, transitive relation.<br />

3, If a, bE A, then exactly oneofthefollowingholds: aPb, bPa, alb,

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!