Can We Identify the Research Hypothesis With the Alternative Hypothesis?

EXTENDED ABSTRACT - The research hypothesis is commonly a statement of the relationship between concepts and is derived from a theory or stored knowledge in a certain research field. The statistical hypothesis, on the other hand, is the expression of a parameter that characterizes population distribution, so it does not contain any logical reasons for specific relationships or characteristics within it (Hay 1981, p. 232). In general, the statistical hypothesis is derived from the research hypothesis, but not vice versa.

Citation:

Hyunchul Cho and Shuzo Abe (2005) ,"Can We Identify the Research Hypothesis With the Alternative Hypothesis?", in AP - Asia Pacific Advances in Consumer Research Volume 6, eds. Yong-Uon Ha and Youjae Yi, Duluth, MN : Association for Consumer Research, Pages: 329.

Asia Pacific Advances in Consumer Research Volume 6, 2005      Page 329

CAN WE IDENTIFY THE RESEARCH HYPOTHESIS WITH THE ALTERNATIVE HYPOTHESIS?

Hyunchul Cho, Hanyang University, Republic of Korea

Shuzo Abe, Yokohama National University, Japan

EXTENDED ABSTRACT -

The research hypothesis is commonly a statement of the relationship between concepts and is derived from a theory or stored knowledge in a certain research field. The statistical hypothesis, on the other hand, is the expression of a parameter that characterizes population distribution, so it does not contain any logical reasons for specific relationships or characteristics within it (Hay 1981, p. 232). In general, the statistical hypothesis is derived from the research hypothesis, but not vice versa.

Although the above differences of definitions between the research and statistical hypotheses seem to be clear enough, the two hypotheses are quite often regarded as identical ones. For example, in their textbook of research methodology Aaker, Kumar and Day (1995, p. 470) describe the statistical hypothesis as a form of verbal description.

Likewise, the commonly used statement, like "The research hypothesis is the alternative hypothesis," (e.g., Albright, Winston, and Zappe 1999, p. 438; Lind, Marchal, and Mason 2002, p. 337; Tropper 1998, p. 155) does not distinguish the research hypothesis from the statistical hypothesis.

Then, are the differences between the research and statistical hypotheses so trifling that it matters only in the definitional sphere? For pedagogical convenience or just to avoid the feeling of redundancy, can we treat the two hypotheses interchangeable as in the above examples? Our answer is "no," and in this article we explain why it is so, and propose to make clear distinction between the two categories of hypothesis. This is because overlooking the differences between them may lead to the following four points of conceptual and empirical confusion.

CONFUSION 1: It May Lead to Categorical Exclusion of the Research Hypothesis in Null Form (RHNF). There are two forms of research hypothesis. They are the research hypothesis in alternative form (RHAF) and the research hypothesis in null form (RHNF). The former RHAF is the common type. It includes a verbal assertion of the existence of some testable relationship between concepts. But, occasionally researchers use the latter RHNF, which involves a verbal assertion of the nonexistence of some testable relationship between concepts. This RHNF becomes pivotal when researchers propose a new theory/model that accompanies a different structure from the existing theory/model. If researchers ignore the distinction between the research and statistical hypotheses and try to apply the statistical testing method, then in the process of doing so, they most naturally would see the hypothesis as statistical hypothesis and apply the logic of statistical testing that assumes only RHAF. So, it leads to categorical ruling out of RHNF.

CONFUSION 2: It May Lead to Wrong Interpretation of Test Result for the Research Hypothesis in Null Form (RHNF). Regarding the statistical hypothesis as identical with the research hypothesis would hinder the clear understanding of the empirical testing system. So, even when researchers test RHNF, it makes the researchers less aware that their test is based on the logic of proof, which is weak and incomplete. There are arguments that the statistical testing is good for only RHAF and it should not be used for testing RHNF, however the authors argue that if we pay careful attention to its limitation, we can use statistical testing for RHNF.

CONFUSION 3: The Flow of Steps in the Whole Process of Theory Testing May not be Observed. Generally, the whole process of theory testing can proceed in the following 7 steps: (1) Theory --> (2) Setting up research hypothesis (RHNF/RHAF) --> (3) Setting up statistical hypothesis (Translating into H₀ and H₁) --> (4) Testing statistical hypothesis (by the logic of disproof/proof) --> (5) Testing research hypothesis (RHNF/RHAF is supported/not supported)--> (6) Testing theory (the theory is supported/not supported) --> (7) Interpretation of this empirical test result from some philosophical standpoint. Unfortunately, if the research and statistical hypotheses are not distinguished, then step 2 and 3 would merge into a single step, then researchers would neglect all the factors involved in the translation process between them. Or, if step 4 and 5 are not handled separately, it would give wrong conception that a single statistical test determines the acceptance or rejection of the research hypothesis. Obviously, this hinders systematic conduction of empirical research.

CONFUSION 4: It May Lead to Wrong Conception of Theory Testing. If researchers identify the theory with the statistical hypothesis, then all the factors involved in the process of deductively deriving the research hypothesis from the theory and then translating it into the statistical hypothesis would be neglected from the consideration. It not only aggravates the problems mentioned in the previous section but also introduces a new type of confusion. An example is the wrong conception like "The existing theory is the null hypothesis and the new theory is the alternative hypothesis" (Albright, Winston, and Zappe 1999, p.438). The authors show that even when researchers pit two competitive theories (Platt 1964), unless they can ingeniously design a single statistical test for two competing research hypotheses, they need to test separately each research hypothesis derived from each theory.

Based on the clear distinction between the research hypothesis expressed by the verbal statement and the statistical hypothesis described by the parameter(s), the authors propose a new scheme for classifying the research hypothesis and the statistical hypothesis.

REFERENCES

Aaker, David A., V. Kumar, and George S. Day (1995), Marketing Research, 5th ed., New York: John Wiley & Sons.

Albright, S. Christian, Wayne L. Winston, and Christopher Zappe (1999), Data Analysis and Decision Making with Microsoft Excel, Pacific Grove, CA: Duxbury Press.

Hay, William L. (1981), Statistics, 3^rd ed., New York: Holt, Rinehart and Winston.

Lind, Douglass A., William G. Marchal, and Robert D. Mason (2002), Statistical Techniques in Business & Economics, 11^th ed., New York: McGraw-Hill.

Platt, John R. (1964), "Strong Inference," Science, 16 October, Vol. 146 (No. 3642), 347-353.

Tropper, Richard (1998), The Interpretation of Data: An Introduction to Statistics for the Behavioral Sciences, Pacific Grove, CA: Brooks/Cole.

----------------------------------------