A class is defined arbitrarily, by identifying a minimal number of characteristics required for inclusion in the class. Recognizing a scientifically useful classification, however, requires inductive insight. Ideally, only one or a few criteria specify a class, but members of the class also share many other attributes. For example, one accomplishes little by classifying dogs according to whether or not they have a scar on their ear. In contrast, classifying dogs as alive or dead (e.g., based on presence/absence of heartbeat) permits a wealth of generally successful predictions about individual dogs. Much insight can be gained by examining these ancillary characteristics. These aspects need not be universal among the class to be informative. It is sufficient that the classification, although based on different criteria, enhances our ability to predict occurrence of these typical features.
Classes are subjectively chosen, but they are defined according to objective criteria. If the criteria involve presence or absence of an attribute (e.g., use of chlorophyll), definition is usually straightforward. If the criteria involve a variable, however, the definition is more obviously subjective in its specification of position (or range of positions) along a continuum of potential values.
A classification scheme can be counterproductive [Oliver, 1991], if it imposes a perspective on the data that limits our perception. A useful classification can become counterproductive, when new data are shoved into it even though they don’t fit.
Classifications evolve to regain utility, when exceptions and anomalous examples are found. Often these exceptions can be explained by a more restrictive and complex class definition. Frequently, the smaller class exhibits greater commonality of other characteristics than was observed within the larger class. For example, to some early astronomers all celestial objects were stars. Those who subdivided this class into ‘wandering stars’ (planets and comets) and ‘fixed stars’ would have been shocked at the immense variety that later generations would discover within these classes.
Each scientist applies personal standards in evaluating the scope and size of a classification. The ‘splitters’ favor subdivision into small subclasses, to achieve more accurate predictive ability. The ‘lumpers’ prefer generalizations that encompass a large portion of the population with reasonable but not perfect predictive accuracy. In every field of science, battles between lumpers and splitters are waged. For many years the splitters dominate a field, creating finer and finer classifications of every variant that is found. Then for a while the lumpers convince the community that the pendulum has swung too far and that much larger classes, though imperfect, are more worthwhile.
A class can even be useful though it has no members whatsoever. An ideal class exhibits behavior that is physically simple and therefore amenable to mathematical modeling. Even if actual individual objects fail to match exactly the defining characteristics of the ideal class, they may be similar enough for the mathematical relationships to apply. Wilson [1952] gives several familiar examples of an ideal class: physicists often model rigid bodies, frictionless surfaces, and incompressible fluids, and chemists employ the concepts of ideal gases, pure compounds, and adiabatic processes.
Coincidence
Classifications, like all explanations, seek meaningful associations and correlations. Sometimes, however, they are misled by coincidence.
“A large number of incorrect conclusions are drawn because the possibility of chance occurrences is not fully considered. This usually arises through lack of proper controls and insufficient repetitions. There is the story of the research worker in nu-
