A case in point is classification. In taking this up from a constructive standpoint, I remarked that all classifications ought to be logical. But I did not say what I meant by logical, nor did I tell how a logical classification could be secured. The two most prominent errors made in classifying are (1) not making classifications mutually exclusive, (2) not making them cover all the objects or phenomena supposed to be classified. The first error is the less common, for though occurring among all thinkers, it is comparatively infrequent among those who proceed with caution. It is, moreover, more easily discovered than the second. Consider the classification of constructive methods into comparison, observation, and experiment. It is apparent that these methods overlap. We cannot compare without observing, much of our observation involves comparison, when we experiment we must of course observe the results obtained, and the results are usually always compared. All three methods could be classed under observation. It is well to remember, however, that the first classification may be useful—even more so than one strictly logical, and that the nature of a subject will often make impracticable, divisions which do not overlap in some degree. The second error—that of not making a classification cover all the objects or phenomena it is supposed to cover—is not so easy to detect. It is one to which the greatest philosophers have been heir. Some of our Socialist friends say there are but two kinds of people: capitalists and laborers, “the people who live on others and the people who are lived on.” They overlook that class of farmers who own a little piece of land and do their own tilling. Even if they insist that such a class “is rapidly becoming extinct,” the fact remains that it is still with us and must be taken into account. All classifications are made with a certain number of facts in mind, and fortunate is he who happens to have just the right facts. We cannot hold many facts in mind at once, and we often generalize upon thousands of things by taking a supposedly representative dozen. To avoid error all we can do is to keep constantly on the lookout for examples, especially those which apparently will not fit into our generalization. If they go in without straining anything, our classification receives added warrant. But sometimes you will find that where you have three classes a new fact will necessitate a fourth, and that often it will overturn your whole beautiful structure.