Abraham Kaplan

Observation

An observation in science is first of all something done, an act performed by the scientist; only thereby is it something seen, a product of the process in which the scientist is engaged. As process, observation is a part of what Nagel calls "controlled investigation". Scientific observation is deliberate search, carried out with care and forethought, as contrasted with the casual and largely passive perceptions of everyday life. It is this deliberateness and control of the process of observation that is distinctive of science, not merely the use of special instruments (important as they are) – save as this use is itself indicative of forethought and care. Tycho Brahe was one of the greatest of astronomical observers though he had no telescope; Darwin also relied heavily on the naked eye; De Tocqueville was a superb observer without any of the data-gathering devices of contemporary social research.

In behavioral science, observation is less likely to involve special instruments than special circumstances – say, like those of the psychoanalytic-interview situation.
Above all, "observation" means that special care is being taken: the root meaning of the word is not just "to see", but "to watch over". The scientist observes his data with the tireless passion and energy of an anxious mother.

Much of the forethought that goes into scientific observation is directed toward making accessible what otherwise could not be seen, or if seen, would not be noticed. Special care is taken to ensure that the scientist will be able to see what he is looking for if it is there to be seen. Only when this condition is satisfied can negative results have scientific significance; and negative results may be of profound importance, marking – as the Michelson-Morley experiment did – the breakdown of expectations based on received doctrine.

Observation is purposive behavior, directed towards ends that lie beyond the act of observation itself: the aim is to secure materials that will play a part in other phases of inquiry, like the formation and validation of hypotheses. When observation is thought as passive exposure to perception, its instrumentality is left out of account. The scientist becomes a voyeur, finding satisfaction in the unproductive experience of just looking at nature. No doubt there is always some gratification in uncovering secrets, exposing what is hidden; but the scientific motivation is more mature in its demands. In science, observation is a search for what is hidden, not just because it is hidden, but because its exposure will facilitate an intimate, sustained, and productive relationship with the world.

Intersubjectivity

To be able to serve these ends, observation must meet certain conditions. “Repeatability” is often spoken of here as a requirement for scientific acceptability. I believe that this is a mistaken specification, or, at best, a misleading one. Many important scientific observation take place on special occasions whose recurrence is incidental to their scientific significance.

Of particular importance to behavioral science are special events like clinical outbursts, disasters, and war crises, as well as regularly recurrent ones like elections or rain dances; and for other sciences, important observations may be made in connection with eclipses, earthquakes, or the birth of quintuplets. Of course, when such events happen again we can observe them again, but we cannot repeat the observations at will. And the recurrences can be expected to differ in ways relevant to the purposes of the observation - the core of soundness in the misplaced emphasis on the "uniqueness" of the subject-matter of behavioral science. The child asks the magician to "do it again", not to subject him to scientific test, but to enjoy once more the encounter with the miraculous. For the scientist, repetition is a device to improve the quality of observation, but not the only device, and not necessarily the best.

The methodological importance of what is called repeatability is, I think, made more plain by its restatement as intersubjectivity. A scientific observation could have been made by any other observer so situated: nature plays no favorites, but exposes herself promiscuously. The intersubjective becomes the mark of objectivity, for it testifies that the observation is uncontaminated by any factors save those common to all observers.

Whether a distortion common to all men can nevertheless be said to yield something objective is a philosophical question that has no bearing on the conduct of the human enterprise of science. The methodological question is always limited to whether what is reported as an observation can be used in subsequent inquiry even if the particular observer is no longer a part of the context. I ask "Do you see what I see?" to help decide whether what I see is to be explained by self-knowledge or by knowledge of the presumed object.

Error

It is usual to refer in this connection to the so-called "human equation", much better designated as the "human factor". The logical significance of an observation is conditioned by psychological factors that have played a part in determining the outcome of the process. Wishful thinking, for example, has its counterpart in wishful seeing. Certain experiments on extrasensory perception, for instance, were afterwards shown to have yielded apparently positive results because of the direction of the quite honest clerical errors made by recorders hopeful of such a result.
Other studies have revealed the effect of social pressure not only on what we believe but quite literally on what we see.

Observers must be trained to observe scientifically, and the very discipline which they undergo may itself subject them to trained incapacities that will produce distortions in other contexts.
The difficulties in assessing the significance of observations do not stem only from the personal, idiosyncratic failings of the observer, but may be rooted in features intrinsic to the process of observation itself. A century ago Augustus De Morgan, one of the founders of mathematical logic, and more sensitive to the problems of the empirical sciences than most of his successors today, called attention to various ways in which we may confuse properties of our observations with what we suppose we have observed (see W. S. Jevons, The Principles of Science, 1892).

Instead of A causing B, it may be our observations on A that cause B, as is illustrated by the famous Hawthorne experiments, where changes in the productivity of workers under varying conditions were at last understood to have resulted just from the fact that the workers knew they were subjects of investigation. Or, A may produce only our observation of B, which would otherwise occur without being observed, as is illustrated in the apparent increased incidence of psychosis in modern urban life, which may be attributable only to the higher frequency with which it is diagnosed and reported. Or, our observation of A may cause our observation of B, as in the attempt to assess the effect of psychotherapy by using the appraisals made by the patients themselves. Or, our observation of A may be necessary to the observation of B, although in fact it is B that causes A – illustrated in the relation between the manifest and latent content of a dream from  the standpoint of the dreamer. Like all skilled performances, observation is by no means as simple as it looks.

There are several general procedures by which errors of observation are taken into account. These procedures are said to constitute controls of the observation: they are efforts responsive to the effects of the particular context or observer, designed to minimize error in assessing the significance of what has been observed.

(1) First, we may institute procedures to insulate the observation separating it from the factors that would otherwise produce error. The training of observers and the setting up of experimental situations as contexts of the observation are largely insulating devices. Special instruments may be employed, like one-way windows, or the intent if not the fact of observation may in other ways be concealed from human subjects. Astronomical observatories are located where the air is clear, and far from city lights, perhaps even in outer space. Questionnaires are pretested to eliminate ambiguities or unintended implications; and so, endlessly.

(2) Second, we may attempt to cancel error where its elimination is out of question. Observations of a child’s behavior, for example, except in very special circumstances, are inevitably colored by the emotional involvements with the child or those who have the most opportunity to observe him: parents, siblings, teachers, and friends. But the very multiplicity of observers may to some extent cancel out the effect of particular relationships. In general, statistical devices may be employed where there is reason to expect a great number of errors more or less independent of one another, for in that case errors in opposite directions are likely to compensate for each other. An interesting compensatory device for the human factor is reported by Darwin, who tells us that he kept a separate notebook to record observations counter to his theory, lest he overlook or underestimate them.

(3) In most cases, however, errors of observation can neither be prevented nor canceled out. What is still possible is to discount the error, make ourselves aware of its direction, and perhaps even of its extent, and take it into account in our treatment of observational data. In observing the shape of an object we might try to insulate against errors of perspective by viewing it from a point directly above its center; in fact we learn early to make use of the laws of perspective in interpreting what we see from any angle: coins look round as we discount the elliptical shapes they usually in fact present. Reaction times of observers can be measured and corrected for, just as astronomers correct the observed time of, say, eclipses by taking into account the time it takes for light from the event to reach us (this kind of correction was in fact the basis of the first determination of the velocity of light). In general, we standardize instruments and contexts of observation, not in order to eliminate an error but rather to give it a fixed and known value, on the basis of which we can shift at will what we choose to call the "zero point".

What is observed

Setting aside the treatment of error – whether it is prevented, canceled, or discounted – what is it that is observed? A usual statement goes that observations are only of matters of fact, while laws and theories are products of conceptual processes. There is undoubtedly an important distinction here. Facts remain fixed through time in a way in which theories definitely do not (of course as time passes situations change, but it remains a fact that the situation was such-and-such at an earlier time). And a particular set of facts may play a part in a wide variety of laws and theories. For these reasons, as W. S. Jevons (The Principles of Science, 1892) pointed out, errors of fact are more "mischievous" than mistaken theories, often much more difficult to identify and correct. Yet I believe that the difference between facts and theories lies in the ways in which they function in inquiry rather than in the processes by which we arrive at them, in their use rather than their origin. All observation involves theorizing, and – for science, at any rate – perception is impossible without conceptual processes. It is hard to improve on Norwood Hanson’s formulation: “There is more to seeing than meets the eyeball.” (Patterns of Discovery, 1958)

There is an empiricist tradition, from Hume through Mill to Russell, in which a distinction is made between "hard" and "soft" data, according to whether they are purely observational or contain an inferential element. The basis of scientific knowledge is taken to be a "reading" or protocol which does not interpret but merely records a perceptual content. Positivists like Pearson and Mach, and Carnap in his early work, all proposed a reconstruction of knowledge on a phenomenalistic basis (Mach's book on this subject is called The Analysis of Sensations). What we observe are bare shapes, sounds, colors, and textures, which are then organized and interpreted as the familiar objects and events of experience. The contents of observation itself are free from conceptual contamination. Nietzsche’s label for this philosophical doctrine is not, I think, unjust; he called it: “the dogma of the immaculate perception”. (Thus Spoke Zarathustra, 1883-1885). 

The fact is that no human perception is immaculate, certainly no perception of any significance for science. Observation is already cognition, not just material for subsequent knowledge, and the possibility of error is as ever-present in this cognitive process as in the more obviously inferential ones. Seeing is believing because we do not just see something: we see that something is the case. The perception apprehends a significant structure, or rather, a structure which becomes significant in the apprehension, acquiring what Dewey calls a “funded meaning”. We take what we perceive as being of a certain kind, bringing an abstract concept to the perceptual situation and subsuming the concrete given under it.

Because of this tacit predication, the eye with which we see is itself the mind's eye, or it would be indeed unseeing. We feel impelled to put a second metaphorical eye behind the real one only because of a reconstruction according to which first something without meaning is seen, which is then interpreted as having a certain significance – a Kantian intuition on which the faculty of understanding then goes to work. But observation is already the work of understanding: the photoelectric cell does not see anything at all in the same sense as we do. The uninterpreted intuition or bare sensation is not the beginning of perception but the end product of a subsequent analysis, a reconstructed accessory after the fact.

It will not do to say that all this is only a matter of psychology, not of the logic of the cognitive process. For it is surely a logical requirement that what is observed be formulable in propositions, which can serve as premises for subsequent inferences. But the language in which observational propositions are stated is itself inferential in character. As Popper (The Logic of Scientific Discovery, 1959) has urged, there is no purely “phenomenal language” distinguishable from a “theoretical language”, no way to talk about something sensed and not interpreted. Mill (A System of Logic, 1843) argued that "what is needful, in order that the fact supposed to be observed, may safely be received as true . . . is that it be an observation, not an inference." But shortly thereafter he recognized that "we cannot describe a fact without implying more than the fact . . . To describe it is to affirm a connection between it and every other thing which is either denoted or connoted by any of the terms used."

Even if perception itself were immaculate, the perceptual report exposes us to sin, as a necessary consequence of the way in which language works. Benjamin Lee Whorf (Language, Thought and Reality, 1956) has put forward the suggestive thesis that the very structure of a language makes for certain segmentations and interpretations of experience. And there have been some experimental indications that perceptual discriminations, say of colors, are affected by the vocabulary available for labeling the differences.

After the moment of the observer's birth no observation can be undertaken in all innocence. We already know something already, and this knowledge is intimately involved in what we come to know next, whether by observation or in any other way. We see what we expect to see, what we believe we have every reason for seeing, and while this expectancy can make for observational error, it is also responsible for veridical perception.

The Japanese do not hear the difference between l and r because in their language these sounds are not contrastive: the difference makes no difference to the meaning of any Japanese words in which they might occur. This indifference distorts their perception of spoken English, but by the same token allows for easy identifications in spoken Japanese. We do not make proper observations by stripping ourselves of theories – which is impossible, in any case – but rather by making use of the theories appropriate to the observational context. Where special instruments or experimental situations are involved in the observation, it is quite clear that theories must playa part, at least in order to discount the errors that may have been introduced by the observational devices themselves. Chromosomes are so called not because they are themselves such markedly colored bodies, but because of the deep stain they take in the processes employed to make them visible. But theories are also at work even in the simplest and most direct observations.

For we do not observe “everything that is there to be seen”.
An observation is made; it is the product of an active choice, not of a passive exposure. Observing is a goal-directed behavior; an observational report is significant on the basis of a presumed relation to the goal. Color is an index of significant differences of structure and function in various types of both stars and algae but not of mushrooms or of mammals (including humans). To include color among the observational data in all these sorts of cases equally is thereby to make certain hypotheses about the facts, and not merely to report "just the facts".

Data are always data for some hypothesis or other; if, as the etymology suggests, they are what is given, the observer must have hypotheses to be eligible to receive them. In his Theory of Data (1964) Clyde Coombs proposes that the term "data" be used for observations already interpreted in some particular way. I am saying that there are no other sorts of observations, though often the interpretation at work is far from explicit and clear. The profound importance of Coombs's analysis is that no one interpretation is necessitated by what is observed; there are always many ways of mapping behavior into data, as he puts it. But without some mapping or other the process of observation is of no scientific significance. To speak of observations as yielding "facts" is only to point to the objective locus of their status and function as data; but to perform their function they must be, as Coombs says in a Kantian idiom, "in part a product of the mind of the observer".

What has just been said is what I also take to be the burden of Dewey's repeated insistence that the perceptual and conceptual materials of knowledge are instituted always in “functional correlativity” with each other. Theories are as much involved in the determination of fact as facts are in establishing a theory. When a fall of meteorites was reported to the French Academy towards the end of the eighteenth century, the datum  that stones sometimes fall from the sky was dismissed as "a superstition unworthy of these enlightened times". We ourselves, of course, are really enlightened, but we are no less dependent on theory for distinguishing between fact and superstition. The example given is one of a large class of what might be called cryptic data: those which, in a given state of science, are hard to make sense of in the light of the theories current at that time. Not uncommonly it is the cryptic data that provide a point of departure for significant theoretical advance.

Even greater importance in the history of science attaches to what might be called invisible data, those which are recognized as data only conjointly with the acceptance of the theory explaining them. They are not Bacon’s “clandestine instances”, which embody some characteristic only in the most rudimentary and therefore inconspicuous form. On the contrary, invisible data are retrospectively seen to be perfectly obvious and even striking manifestations.

Freud's clinical observations of hysteria in males were dismissed as absurd (the word hysteria itself is from the Greek for womb), as were his accounts of infantile sexuality, since everyone "knew" sexuality begins only at puberty; today both of Freud's observations can easily be made by most wives and all mothers. The "paradigm observer" has been described by Hanson (Patterns of Discovery, 1958) as "not the man who sees and reports what all normal observers see and report, but the man who sees in familiar objects what no one else has seen before”. Intersubjectivity is still called for, but it is achieved only after the fact.

In some cases data are invisible because a scientific dogma makes for a simple refusal even to look at them: none so blind as those who will not see. Galileo's colleagues refused to look through his telescope at the moons of Jupiter which he had discovered – a case which today is often cited (with what justice is debatable) by those interested in telepathy and related matters. Whatever the scientific merit of particular findings, it cannot be denied that in general, what is most responsible for invisible data, is the force of preconceived opinion.

"This it is," Mill (A System of Logic, 1843) has remarked, "which, in all ages, has made the whole race of mankind, and every separate section of it, for the most part unobservant of all facts, however abundant, even when passing under their own eyes, which are contradictory to any first appearance or any received tenet."

It is not our conceptions only which both limit and inform the process of observation. Instruments of observation are also, of enormous significance for the course of inquiry as a whole. The state of knowledge at any particular time is profoundly affected, not only by the techniques of gathering data then current, but also by the technology for doing so.
It was more than the nineteenth century's pride in its technological achievement which impelled Jevons (The Principles of Science, 1892) to proclaim that "the invention of an instrument has usually marked, if it has not made, an epoch.”

I believe that the history of science bears him out to the full. His statement is as true for behavioral science as it is for physics and biology. In 1903, for example, "the experimental training of a rat in a maze led at once to a long series of studies in the evolution of animal intelligence with the maze as the observational instrument" (E. G. Boring in P. Frank, The Validation of Scientific Theories, 1961). Similar importance attaches to such observational devices and situations as, say, free association, communication nets, and opinion polls.