81
Operationalism had a strong influence on the human sciences (in particular, psychology) through the work of early behaviorists such as Burrhus Frederic Skinner
(1945), and Edwin Garrigues Boring (1923), who found it to be consistent with the
prevailing empiricist attitudes of the time; the operationalist doctrine obviated the
need to define measured properties independently of the manner in which they were
assessed. Stanley Smith Stevens, who was Boring’s student, famously asserted that
“measurement, in the broadest sense, is defined as the assignment of numerals to
objects or events according to rules” (Stevens, 1946: p. 677; see also Green, 1992;
Leahey, 1980; McGrane, 2015).
One obvious reason for the attractiveness of operationalism to early psychologists was the difficulty they faced in precisely defining psychosocial properties,
which are not observable by traditional criteria, and seemed dangerously metaphysical in contrast to the positivist and behaviorist zeitgeist of the time: claims such as
Boring’s (1923: p. 35) that “intelligence is what the tests test” neatly sidestepped the
issue. Notable in the idea of operationalism is that “it is meaningless to ask whether
something is ‘really’ being measured […] there is neither a need nor a place for
postulating properties which are prior to the measurement operation” (Borsboom,
2005: p. 93). For example, according to an operationalist perspective, any form of
knowledge or skill (e.g., reading comprehension ability) could be measured simply
by assembling a set of questions that are judged (by whatever criteria) to be relevant; the property of reading comprehension ability could then be defined in terms
of the answers a student gives to the set of questions. Stevens’ definition of measurement is consistent with operationalism, insofar as the only given necessary condition for measurement is the presence of a rule (or operation) for numerical
assignment. Stevens was indeed quite clear that this could be any rule other than
“random assignment, for randomness amounts in effect to a nonrule” (Stevens,
1975: p. 47).
But despite the attractive simplicity of operationalist approaches to thinking
about measurement, difficulties quickly become apparent. Measurement results,
like many forms of information, are prized largely because of their transferability;
by defining what is measured by a given instrument as just the property that interacted here and now with the measuring instrument, it is not obvious on what basis
the information obtained could be used in reference to anything other than the current conditions.
Indeed, a coherent operationalist stance would hold that purely empirical means
cannot demonstrate that different instruments measure the same property, given that
any such demonstration unavoidably includes an inferential component. This has
the consequence that each unique set of operations must be associated with a distinct concept; thus, for example, the outcome of the application of an alcohol thermometer and the outcome of the application of a mercury thermometer cannot refer
to the same property, nor could two distinct tests of reading comprehension ability
be claimed to measure the same property (Borsboom, 2005; Green, 2001). This is
clearly inconsistent with common thinking and discourse; for example, most educators would immediately recognize that students’ knowledge, skills, and abilities are
not equivalent to their score on a particular test, and that such an identification
4.2 Characterizing measurement
Operationalism had a strong influence on the human sciences (in particular, psychology) through the work of early behaviorists such as Burrhus Frederic Skinner
(1945), and Edwin Garrigues Boring (1923), who found it to be consistent with the
prevailing empiricist attitudes of the time; the operationalist doctrine obviated the
need to define measured properties independently of the manner in which they were
assessed. Stanley Smith Stevens, who was Boring’s student, famously asserted that
“measurement, in the broadest sense, is defined as the assignment of numerals to
objects or events according to rules” (Stevens, 1946: p. 677; see also Green, 1992;
Leahey, 1980; McGrane, 2015).
One obvious reason for the attractiveness of operationalism to early psychologists was the difficulty they faced in precisely defining psychosocial properties,
which are not observable by traditional criteria, and seemed dangerously metaphysical in contrast to the positivist and behaviorist zeitgeist of the time: claims such as
Boring’s (1923: p. 35) that “intelligence is what the tests test” neatly sidestepped the
issue. Notable in the idea of operationalism is that “it is meaningless to ask whether
something is ‘really’ being measured […] there is neither a need nor a place for
postulating properties which are prior to the measurement operation” (Borsboom,
2005: p. 93). For example, according to an operationalist perspective, any form of
knowledge or skill (e.g., reading comprehension ability) could be measured simply
by assembling a set of questions that are judged (by whatever criteria) to be relevant; the property of reading comprehension ability could then be defined in terms
of the answers a student gives to the set of questions. Stevens’ definition of measurement is consistent with operationalism, insofar as the only given necessary condition for measurement is the presence of a rule (or operation) for numerical
assignment. Stevens was indeed quite clear that this could be any rule other than
“random assignment, for randomness amounts in effect to a nonrule” (Stevens,
1975: p. 47).
But despite the attractive simplicity of operationalist approaches to thinking
about measurement, difficulties quickly become apparent. Measurement results,
like many forms of information, are prized largely because of their transferability;
by defining what is measured by a given instrument as just the property that interacted here and now with the measuring instrument, it is not obvious on what basis
the information obtained could be used in reference to anything other than the current conditions.
Indeed, a coherent operationalist stance would hold that purely empirical means
cannot demonstrate that different instruments measure the same property, given that
any such demonstration unavoidably includes an inferential component. This has
the consequence that each unique set of operations must be associated with a distinct concept; thus, for example, the outcome of the application of an alcohol thermometer and the outcome of the application of a mercury thermometer cannot refer
to the same property, nor could two distinct tests of reading comprehension ability
be claimed to measure the same property (Borsboom, 2005; Green, 2001). This is
clearly inconsistent with common thinking and discourse; for example, most educators would immediately recognize that students’ knowledge, skills, and abilities are
not equivalent to their score on a particular test, and that such an identification
4.2 Characterizing measurement