Hypotheses are assumptions about empirical (observable) phenomena. They are formulated as empirical (experience-related) statements; thus they can be either true or false – i.e., they are testable. This implies that hypotheses are tentative: their validity (truth or falseness) is subject to empirical test.
In this general form, the definition also corresponds to the conventional, everyday meaning of “hypothesis.” In the social sciences – thus including social science-oriented communication sciences – the concept is more narrowly circumscribed: scientific hypotheses are empirically testable universal (nomological) propositions about causal relationships. They are stated in terms of either “if–then” or “the more–the more (or less)” propositions.
Example: “People learn from news media which are the most important problems facing their country”.
- This statement is an assumption which can be empirically tested.
- The statement can be translated into an “if–then” proposition: “If an issue is prominently discussed in the news media, then people will regard it as an important problem.” In this case, it can also be translated into a “the more–the more” proposition: “The more an issue is discussed in the news media, the more people will regard it as an important problem.”
- The statement applies not merely to a single media user, but to all media users; it is not limited in terms of time and place and is thus universal, and not singular.
- “Media use can lead to violent behavior.” Because of the “can,” this statement cannot be false and is thus not empirically testable (“If media are used, this may, or may not, lead to violent behavior”).
- “John likes to watch television.” This statement is testable, but it is singular and not universal.
- “Diversity of opinion is to be guaranteed.” This is a normative statement, not a proposition.
- “Everyone has the opportunity to get information from the mass media.” This statement does not refer to a causal relationship (if–then proposition) but describes an existing state.
- “The world wide web is a mass medium.” This statement is a definition and thus a denotation; unlike a hypothesis, it cannot be falsified (Definition). A corresponding hypothesis might read: “The world wide web is generally perceived as a mass medium.”
- “If television had not been invented, people would be more peaceful.” The if-component of this proposition is counterfactual and cannot be reproduced artificially. Therefore, the proposition cannot be tested empirically (many everyday claims about media effects have that quality).
The Relationship Of Hypotheses And Theories
In the realm of epistemology, the concept of “hypothesis” has a different meaning than in the context of social science methodology. In epistemological discourse, any testable proposition, according to the “general form” above, is called a hypothesis. That makes “hypothesis” a logically superordinate concept relative to “theory,” and indeed it is often used as a synonym, as, for instance, in Popper’s The logic of scientific discovery.
In social science discourse, the term “hypothesis” is reserved for specific statements rendered directly testable by virtue of measurements. A “theory” in this context is a system of (few) basic propositions and a large number of logical derivations. The basic propositions of a theory as well as the derivations must correspond to the fundamental form of social science hypotheses: they are universal propositions about causal relationships. This is a prerequisite for “backward” inference, from the results of empirical testing, via the derivations, to the fundamental propositions – i.e., the core – of the theory. A theory will, in principle, allow an unlimited number of specific hypotheses to be derived (deduced). Obviously, they cannot all be specified in advance. Therefore, specific hypotheses are spelled out only in the context of empirical studies.
The decision about the empirical validity of a derived hypothesis is tantamount to a decision about the validity (truth or falsehood) of the underlying theory in its current form; it is thereby confirmed, modified, or rejected. If a hypothesis derived from the basic propositions of the theory is confirmed, the theory is said to be corroborated. If the results of an empirical study contradict the hypotheses derived from the theory, it ought to be rejected. In practice, however, theories tend to be more robust than their derived hypotheses. Thus, if empirical results contradict the derived hypotheses, it is common practice to examine or discuss first the possibility of flawed derivations or faulty measurements, i.e., to examine whether logical or technical deficiencies might have occurred in conducting the test. If such deficiencies can be largely excluded, it still remains possible to modify the theory by adding limiting conditions which could explain the deviant observations. These added conditions, however, necessarily reduce the explanatory power of a theory since they restrict its range of applicability. If a theory cannot be “saved” by adding more and more limiting conditions, it will ultimately have to be rejected.
The decision about the validity of a theory is based on hypotheses derived from the theory. This principle is referred to as the hypothetico-deductive method.
According to Popper’s Logic of scientific discovery (2006), the validity of universal propositions (general statements, including hypotheses) can never be proved (verified). To do this would require an examination of all conceivable objects to which the universal statement refers. This is logically impossible, if only because theories also refer to future observations.
While it is true that hypotheses are not verifiable, they are regarded as refuted (falsified) if specific observations contradict a hypothesis. Hypotheses always rule out some possible observations. The content, or power, of a hypothesis is based on the extent to which conceivable observations are ruled out. A hypothesis is considered falsifiable if a basic statement can be formed which asserts the negation of the hypothesis. That is to say that a basic statement postulates the existence of at least one observable instance, anywhere or at any time (universally), that contradicts the hypothesis. Hypotheses, as universal propositions, refer to infinitely many phenomena and situations which can, in principle, never be tested in their entirety; but when a universal proposition – that is, a universal postulate of existence – can be verified, the hypothesis is falsified: “Whenever it is found that something exists here or there, a strictly existential statement may thereby be verified, or a universal one falsified” (Popper 2006, 49). Popper’s most famous example is this: the universal statement “All swans are white” is falsified if the universal existential statement “There exist (somewhere and sometime) non-white swans” can be empirically verified. Scientific research consists of the earnest search for observations which would falsify the hypotheses.
In this manner, the problem of falsification of a general hypothesis has been converted into the problem of verification of a falsifying existential statement. The question now becomes whether the hypothesis should be considered refuted by a single contradictory observation, or whether instead the validity of that observation should be cast in doubt. Since any observation is fallible, we should be trying to replicate it. A theory is considered to be falsified only “if a low-level empirical hypothesis which describes such an effect is proposed and corroborated (Popper 2006, 66).
Only universal existential statements can be replicated. Consequently, the confirmation of fundamental statements can be replicated, potential mistakes can be excluded with sufficient certainty, and theories thus falsified. Yet it will never be possible to confirm a sufficient number of existential statements (with errors safely excluded) to permit a universal statement, and thus a hypothesis, to be verified.
Observations, or empirical studies, can produce results of two kinds: the observations correspond or contradict the hypothesis. The principle of replication calls for repeated, independent observations in order to rule out errors with sufficient certainty. If the replications are in correspondence to the hypothesis, the degree of corroboration is rising. If the replications are in contradiction to the hypothesis, the degree of falsification is rising.
Testing Social Science Hypotheses
The hypotheses derived from a theory are causal assumptions. Actually, causality is a rather demanding concept about temporally immediate cause-and-effect processes. If the assumed causal processes are in fact operating, they will induce corresponding relationships between variables in social science data (e.g., correlations or differences in means) that can be analyzed with the aid of statistical procedures.
In the physical sciences, hypotheses are statements that refer to individual instances. In the social sciences, by contrast, hypotheses are statements about quantities of cases. To illustrate, consider the agenda-setting hypothesis: “When an issue is prominently discussed in the media, there is a strong probability that it will be regarded as important by media users.” To falsify that hypothesis, it would not be sufficient to find one person who is using the media yet does not regard a media-emphasized issue as important. Statements about quantities of cases usually take the form of probability statements, and they can be tested only in terms of empirical frequency distributions. Social science hypotheses are, therefore, tested on the basis of samples.
Potential samples, however, differ in terms of their composition. The set of all objects to which a hypothesis is to apply, called the population or sampling universe, determines the composition of the sample. The safest principle for generating a sample that represents the composition of the universe is a random sample. It implies that each element of the sampling universe has the same chance of becoming part of the sample. However, by using a random sample we no longer obtain exact measurements, but merely estimates of the assumed causal effects, with a specifiable degree of imprecision. Possible errors due to random deviations are represented and tested by means of statistical hypotheses.
By means of statistical hypotheses we can test whether observed relationships between two or more variables in the sample data might simply be the result of chance, i.e., of the random sampling procedures by which the sample was generated. This possibility, i.e., that no causal relationship exists in the population, is expressed statistically in terms of the so-called null hypothesis H0. If the relationship in the data is very clear (significant), in a statistical sense, H0 is rejected, as that would represent a very unlikely case; instead, the (set of ) statistical alternative hypothesis H1 – i.e., the substantive hypothesis that a causal relationship does exist – is maintained.
Regardless of whether or not H0 or H1 is accepted, an erroneous decision can never be ruled out with complete certainty. The two statistical hypotheses imply two possible types of error:
Type I error: H0 is true, but H1 is accepted erroneously;
Type II error: H1 is true, but H0 is accepted erroneously.
In the case of type II error, we draw no new inferences from the data. In the case of type I error, we are definitely drawing erroneous conclusions, as the apparent relationship observed in the data does not really exist in the population but is merely a result of chance fluctuations. The goal thus is to keep the probability of type I error (called α) as small as possible. When the probability of type I error remains below an acceptable level (in the social sciences commonly set at 5 percent), the statistical null hypothesis (H0) is considered to be falsified, and the statistical alternative hypothesis (H1) will be maintained. In that case, a statistically supported substantive causal relationship is assumed to exist.
Thus far, this article has been concerned with the question of how hypotheses in the social or communication sciences can be evaluated by means of empirical data. In epistemological terms, this is sometimes called the logic of scientific discovery. The following is concerned with what is sometimes called the psychology of scientific discovery. In terms of process, a hypothesis has to be devised before it is tested. It is the testing, however, that is decisive, because only the strict rules of hypothesis testing lead to scientific knowledge. By contrast, the generation of hypotheses is not subject to such rules. Science is productive as long as the generation of hypotheses is free. This is where Feyerabend’s “Anything goes” (Feyerabend 1993) is appropriate. It is of no relevance whether the idea for a theory or hypothesis arises from a researcher’s subjective impressions, or an inspirational dream, or inductively as a result of analyzing data.
Although there are no rules for the generation of hypotheses, some recommendations may be offered. A researcher need not wait for ingenious intuitions or inspiring dreams; it is certainly permissible to derive hypotheses from the analysis of empirical observations. Indeed, the results of empirical studies can and should be exploited for the generation of hypotheses. One of the most famous communications researchers, Paul Lazarsfeld, in his landmark study of “people’s choice,” tested and refuted the thesis of direct persuasion, and then proceeded to develop his theory of the “two-step flow of communication” inductively, on the basis of his empirical results.
- Chalmers, A. F. (1999). What is this thing called science? New York: McGraw-Hill.
- Feyerabend, P. (1993). Against method: Outline of an anarchistic theory of knowledge. New York: W. W. Norton. (Original work published 1975).
- Gerring, J. (2001). Social science methodology: A criterial framework. Cambridge: Cambridge University Press.
- Kerlinger, F. N., & Lee, H. B. (1999). Foundations of behavioral research. Belmont, CA: Wadsworth.
- Lazarsfeld, P. F. (1968). People’s choice: How the voter makes up his mind in a presidential campaign. New York: Columbia University Press.
- Popper, K. (2006). The logic of scientific discovery. London: Routledge. (Original work published 1959).