Urie Bonfenbrenner on scientific method

What is science? What is the aim of science? Pre­diction and control? Pursuit of knowledge and truth? Science pursues knowledge and truth in accordance with a particular set of rules. In broad terms, science seeks the establishment of verifiable general laws--principles that can be tested rigorously in terms of objective, controlled observation.

Steps in the scientific process: How do you establish a verifiable gen­eral law? Before you can engage in verifying, you have to have something to verify. That something is called....a hypothesis: a tentative explana­tion. Taking a tentative explanation and demonstrating whether it is right or wrong--that is the business of science.

Would you say then that science essentially begins with a hypothesis and goes on from there? Anything preceding a hypothesis is not strictly science? Is that right? No. Testing out some ready-made explanation under controlled conditions is actually only the last step in a much longer process, and the easiest. Easiest, because there exists a well-worked-out methodology for testing hypotheses.

Not so for prior stages of the scientific process. For example, where do hypotheses come from? Observation? Data? Theory? Intuition? Of course, some sources of hypotheses are likely to be more rewarding than others. Whatever their source, hypotheses must stand some chance of being consistent with events in the objective world. This is why observa­tions are so often relied upon in developing a hypothesis, for they focus on objective reality from the very beginning, or at least try to do so.

How does one go about observing? You have to be looking for something, something worth looking for. And this business of finding something worth looking at--of asking the right questions--is one of the most difficult aspects of scientific work, and one about which the least is known.

But even if you have a good research question, something important to look for, you are not yet in business as a scientist. Looking is not enough; you also have to see. Most of us look without seeing. Seeing is a skill that has to be learned.

And even if you do see something--something important--your observation has no scientific value unless you can describe what you have seen in such a way that another observer gets an accurate picture enabling him to distinguish the phenomenon from others similar, but not identical to it.

So science involves several different processes--beginning with defi­nition of a problem, observation, and description, and then proceeding to hypothesis formulation and testing.

Defining A Problem This is probably the most difficult of the scien-tist’s tasks. What is difficult is not simply to ask questions, but to ask the right questions--questions that are reasonable and fruitful. Moreover, they must be researchable.

Researchable means that you pose a question that is capable of being answered--or at least progress can be made on it--through observation or experiment--i.e., by obtaining objective data. This is called empirical investigation. Fruitful means that such investigation leads eventually to the formu­lation of hypotheses, some of which are verified and become established as general principles, explanations of the phenomenon in question.

It’s this business of seeing and defining a researchable and fruitful problem that separates the men from the boys (and the women from the girls?), the creative scientist from the hack. It was Goethe who said, "That is most difficult of all though it would appear to be the easiest, to see with your eyes what is lying before them."

What is a Hypothesis? We have said that the scientific process begins with the asking of a question about the nature of objective real­ity. The scientist then tries to obtain an answer to his questions through observing, classifying, and relating what he sees. He tries to arrive at tentative answers. Such tentative answers are called hypotheses.

In the beginning we said that a hypothesis was a tentative explanation. Actually, a tentative explanation is only one form--the highest form--of hypothesis. There are other types as well, each of which has its use in sci­ence. If we seek a general definition, one that encompasses all uses of the term, we can say that a hypothesis is any supposition about a fact.

Forms of Hypotheses

  1. A supposition concerned with only one characteristic or class of events we shall call a single variable hypothesis.
  2. An associative hypothesis is one that stipulates a statistical relation­ship between two variables without explicitly asserting that one of these variables influences the other.
  3. A causal hypothesis asserts that variation in one factor produces variation in the other.

Examples of these types of hypothesis:

  1. Igneous rocks are radioactive.
  2. Rates of congenital malformation are higher among residents of mountain areas than of plains and valleys.
  3. An increase in the amount of natural radiation increases the rate of congenital malformation in the population living in the area.

Each of these three statements is a supposition about a fact, but the propositions differ in kind. To begin with, note how the first statement can be distinguished from the other two. The latter both postulate a relation between two different things, whereas the former deals with only one factor--radioactivity in igneous rocks. Additionally, the second proposition simply posits a statistical relationship between two variables, rate of congenital malformation on the one hand, geographi­cal environment on the other. In contrast, the last proposition explicitly goes beyond a merely statistical association in claiming a cause and effect relationship: natural radiation is presumed to contribute to the development of congenital malformations.

It is true that many associative hypotheses, when proved correct, add weight to an existing causal hypothesis, or suggest a new explanatory principle. But the associative hypothesis may not imply or suggest any explanations at all; it may merely call attention to a phenomenon to be explained. What an associative type hypothesis does is to raise the ques­tion of a possible pattern among the variables observed. If the presence is confirmed, this fact in time may serve one of three purposes: call attention to a new research problem, suggest a new causal hypothesis for investigation, or provide evidence in support or rejection of an already existing one.

We have said that a causal hypothesis differs from a purely associative one in stipulating a cause-and-effect relationship. What do we mean by this term? A cause-and-effect relationship exists when a change in one variable is a necessary and/or sufficient condition to produce a given effect on another variable.

Independent and Dependent Variables Notice first of all that the hypothesis postulates a relation between two factors, one which produces the change, the other one in which the change is produced. The first, the one that produces the change, is called the independent or antecedent variable. The other, the one that is changed, is called the dependent or consequent variable, because it hangs (depends) on the independent variable and follows from it.

Usually, the causal hypothesis is a one-way street. It works in only one direction. For example, changing the amount of natural radiation increases congenital malformation, but a change in the malformation rate, say by medical intervention, cannot have the slightest effect on the radioactivity of the rocks. But there can be causal relationships which operate in both directions. In other words, independent and dependent variables are interchangeable. For this and other reasons, statements of causal hypothesis can be deceptive and may require careful analysis to ascertain which variable is which.

A helpful device in this connection is to conduct a hypothetical experiment in your mind--what the German psychologists call a "gedanken experiment," a "thought experiment." Try changing each of the variables, and see what happens to the other one. If a change in A leads to a change in B, then A is an independent variable and B dependent. Conversely, if A varies with the change in B, it is A which is the dependent variable. If both change we have what is called a reversible causal relationship.

This does not mean, however, that a directional relationship, in which a change in one variable is followed by a change in the other, necessarily implies cause and effect.

Necessary and Sufficient Conditions This brings us to the main feature which distinguishes the causal hypothesis from its purely asso­ciative counterpart: the former always stipulates that one variable actu­ally influences the other. This requirement may take two different forms, specified in the definition by the terms "necessary" or "sufficient." What do each of these terms mean?

Necessary means that the effect cannot occur except under the speci­fied condition. For example, the disease process known as tuberculosis cannot occur in animal or man in the absence of a creature known as the micro-bacterium tuberculosis, the bacterium which we say "causes" the illness. It is a necessary condition.

Although the tubercular bacillus is necessary for the development of the disease, it is quite possible for a person to be a carrier of this bacillus and yet be perfectly healthy, because the human organism develops anti­bodies which keep the bacillus under control. In other words, the bacil­lus is a necessary condition, but not a sufficient one. A sufficient condi­tion is one that produces a given effect: For example, stick a healthy baby with a pin, and it cries. He will also cry if deprived of food. But sticking him with a pin will do the job. It is a sufficient condition. Notice that in actual practice, a sufficient condition may not produce the given effect in every case. An infant will not cry every time it is stuck with a pin: it may have a bad case of laryngitis, the pin may not hit a pain spot, or the baby may be crying already. In other words, something can interfere.

We can distinguish several kinds of such interferences. To begin with, the phenomenon in question can occur only under certain bound­ary conditions. These boundary conditions are of two kinds. First, all necessary conditions for the effect in the dependent variable must be satisfied. Thus a baby cannot cry without a functioning voice box. In other words, a sufficient condition cannot be effective until all necessary conditions are met. But even if all necessary conditions for the effect are satisfied, it may not be able to perceive the stimulus and hence not make the response.

All of these conditions must be met--not for the effect to occur in the dependent variable but for the independent variable to be functional in producing the effect. In other words, we are dealing here with a class of variables which limit not the dependent variable but the relation between the independent and dependent variables. To distin­guish such requirements from what we have called necessary condi-tions--those that apply to the dependent variable itself--we shall refer to them as contingent conditions. Given an independent variable x and a dependent variable y, a contingent condition is one that is not neces­sary to produce a given effect in y but is required for x to produce the given effect in y. Boundary conditions, then, include all necessary and contingent conditions.

We are now in a position to offer a definition of a sufficient condition: within a given set of boundary conditions, x is sufficient condition if a change in x produces a change in y.

Interrelations of the Three Types of Hypotheses A causal hypothesis, then, is one that stipulates one variable is a necessary or suf­ficient condition for another. Note also that the causal hypothesis always implies an associative hypothesis as well, since a necessary or sufficient condition inevitably products a statistical relationship between inde­pendent and dependent variables. Finally, an associative hypothesis assumes that each of its components is an observable phenomenon; in other words, it implies two single variable hypotheses. In short, the three types of hypotheses fall into a nested arrangement, like a set of Russian dolls, with the causal hypothesis containing the associative hypothesis, and the associative containing two single-variable hypotheses.

If one asks whether there exists a still higher order construct encom­passing more than one causal hypothesis, the answer is of course found in the concept of a theory. Although this term has no rigorous definition or use, it usually refers to a body of interrelated hypotheses (e.g., Freud­ian theory, evolutionary theory, etc.).

The hierarchical structure of the causal hypothesis dictates the steps to be followed in proving a cause-and-effect relationship. The investiga­tor begins by specifying a procedure for observing each of the separate variables in the investigation. By demonstrating that such observations can be made, he in effect confirms the single-variable hypothesis for each factor. He then proceeds to demonstrate a statistical association between two of the variables under circumstances which require such an association to be the product of a cause-and-effect relationship.

The Logic of Verification We now come to the final phase in the scientific process: testing a hypothesis--proving that it is true or false. How do we do this? Given our definition of a cause and effect relation­ship, the task of verification becomes that of demonstrating that a given condition is necessary or sufficient. As we shall see, the proof begins somewhat differently for the two cases, but ends with a similar exacting requirement.

Let us begin with the case of a necessary condition. Since necessary means that the dependent variable (B) cannot occur without the inde­pendent variable (A), one must show first of all that all possible instances of (B) are preceded by (A). The situation with a sufficient condition involves a different requirement. Here variation in one factor must pro­duce a given effect on the other. In other words, a change in A must be followed by the specified effect in B. Now let us suppose that one of the two, one or the other of the above requirements is fulfilled. Does this mean that a necessary condition has in fact been demonstrated in the one case and a sufficient condition in the other? Unfortunately not. If that were all there was to it, the work of the scientist would be much easier than it is, and much less interesting.

The inadequacy of either of the above demonstrations is brought home by the tale of a man who obviously understood exactly what he was doing. This chap discovered that every time he drank a highball of scotch and ginger ale, he became drunk. The next time he tried gin and ginger ale and got the same effect. Then he experimented with rye and ginger ale and you know what happened. What to do? "Aha," he said, "I know what does it." And the next time he eliminated the ginger ale.

This story, which all too often has some truth to it, brings home the point that it is not enough to show the required kind of relationship between the independent variable A the dependent variable B to demon­strate a necessary or sufficient condition. One must also establish that no other factor X is functioning as an independent variable. For if it is, then the results are confounded; that is, one cannot separate the effect of A from the effect of X. And if X is actually responsible for the entire effect, then A is neither necessary nor sufficient.

Demonstrating that a given effect is produced by A and not by some other factor can turn out to be a fairly complicated task, especially if people in general, and scientists in particular, are sure they already know what the necessary or sufficient condition is. Verification does not take place in a vacuum; it occurs in the minds of men. The human mind is not always capable of seeing objectively, let alone thinking objectively. Both perception and thought can be distorted by conviction and desire. Nor are scientists immune to such distortions. Since one does not have to be a social scientist to regard himself or be regarded by others as an authority on problems of human behavior, the researcher in this sphere frequently finds himself confronted with a task of psychological as well as logical persuasion. Nevertheless, it is logic that lies at the heart of the matter.

VIP (Very Important Point) 1.

A causal hypothesis is not proved so long as an alternative hypothesis can be offered to explain the same finding. In other words, no causal hypothesis can be regarded as confirmed until all plausible alternative hypotheses have been eliminated. This requirement points up a critical limitation of any purely associative hypothesis. The demonstration that a particular kind of association exists of course does not eliminate the possibility that this association is the product of some third factor.

VIP 2.

The demonstration that a particular kind of association between two variables exists cannot, by itself, prove that one of these variables is a necessary or sufficient condition for the other. Putting the matter more generally--in establishing a necessary condi­tion, all positive instances of the dependent variable must be associated with presence of the independent variable. There can be no exceptions. So the results for an associative hypothesis can be useful after all. They can be used to disprove a hypothesis about a necessary condition.

VIP 3.

Results for a hypothesis of association can disprove a hypothesis of necessity if there is at least one positive case of the dependent variable that is negative for the independent variable.

Can results for a hypothesis of association disprove a hypothesis of sufficient condition as well?

VIP 4.

A negative result for a hypothesis of association cannot, by itself, disprove a hypothesis of sufficiency, since it does not eliminate the possibility that an association in fact exists but is counteracted by some third factor. Since results for a purely associative hypothesis can neither prove nor disprove that a variable is a sufficient condition, do they have any utility at all in investigating hypotheses of sufficiency? Of course, the associa­tion can show that you are on the right track.

VIP 5.

A positive result for an associative hypothesis increases the prob­ability that the implied causal hypothesis can be sustained: a negative result decreases that probability, eliminating it completely in the case of a hypothesis of necessity.

In view of the above principle, tests of associative hypotheses are very useful in the early stages of research, not only to pin down the indepen­dent variables that are to become the foci of further investigation, but also tentatively to identify or eliminate other factors that may have to be controlled in order to establish the hypothesis in question. Remember: a hypothesis is not proved until no alternative hypothesis can be given to explain the observed findings. Does that mean that you have to rule out any factor that happens to be associated with the independent variable of the hypothesis? No. So what alternative hypotheses do you have to rule out? When must a variable be considered as a potentially confound­ing factor?

VIP 6.

In testing a hypothesis, a variable must be considered as a potentially confounding factor when it is associated with the independent variable of the hypothesis and scientific grounds exist for believing that the vari­able in question could influence the dependent variable.

And then there is the fundamental principle for verifying causal hypotheses:

VIP 7.

To establish a cause-and-effect relationship, one demonstrates that variable x has an effect on variable y by allowing x to vary, holding con­stant other sources of variation for y, and then showing that y varies in a specified fashion as a function of the variation in x.

The Problem of Levels of Analysis

The scientist is free to choose the level of analysis at which he wishes to work (chemical, biological, psychological, sociological, etc.). His independent and dependent variables may be at the same or at different levels. The only restriction upon him is that he may not claim conclusions beyond the levels at which he has worked, although his results may suggest new problems and hypotheses at these other levels. At whatever levels the causal hypothesis is couched, the logic of proof remains the same. We may summarize this logic by offering a paradox: the process of proof is actually one of disproof. Scientific truth is established by default.

Some Questions to Learn From

In discussing the structure of hypotheses we focused attention on the distinction between independent and dependent variables and between necessary and sufficient conditions. If the explanation of these differ­ences seemed clear to you this is fortunate but perhaps deceptive. For simple as these distinctions appear in theory, they can be elusive in practice. When confronted with an actual research problem, even the experienced investigator can become confused about which variable is independent and which dependent, about whether a given hypothesis is one of necessity or sufficiency or both. Indeed, this is one of the ways in which progress in science occurs: what earlier seemed simple and self-evident becomes confused and problematic.

So do not be discouraged if you become confused. Stay loose. It may be a sign of progress! To help you in this paradoxical fashion, we provide below a set of ques­tions which, if they serve their proposed purpose, will appear simple at first, turn out to be confusing, but lead eventually or ultimately to a clearer understanding of hypotheses and their constituent elements.

In short, you are asked to wrestle with these questions as a necessary preparation for this course and your future work in anthropology or any other science.

  1. Can a dependent variable be a necessary condition? A sufficient condition?
  2. Can an independent variable be a necessary condition? A suffi­cient condition?
  3. Can a condition be necessary without being sufficient? Give an example.
  4. Can a condition be sufficient and not be necessary? Give an exam­ple.
  5. Can a condition be both necessary and sufficient? Give an exam­ple.
  6. Can there be more than one necessary condition for a given effect? Give an example.
  7. Can there be more than one sufficient condition for a given effect? Give an example.


from Ralph Bolton, adapted from Urie Bronfenbrenner lecture notes