Null Hypothesis

The null hypothesis is an essential component of HYPOTHESISTESTING in social research. It is relevant when quantitative measures of social activities have been made and when hypotheses derived from theories are to be tested. The theoretical (or research) hypothesis defines a certain pattern to be found in the data, and the statistical analysis is designed to evaluate the extent to which the evidence from the collected measurements supports the existence of that pattern. The null hypothesis is the hypothesis that the pattern of data found has occurred simply by chance.

Take an example of a RANDOM SAMPLING of participants drawn from the human POPULATION, which is randomly divided into two groups. One group may receive a treatment, such as a small dose of alcohol, whereas the other receives a placebo (to conceal from the participants which group they are in). Their performance on a driving task (or other related skill) may then be measured. The research hypothesis will be that the two groups will have different levels of performance. The null hypothesis is that the alcohol has no effect and any resulting difference between the two groups is due to random chance. If the null hypothesis is true, we expect the mean performance of the two groups to be the same. However, any random assignment of participants into two groups almost always will produce two groups whose measured performance differs by some amount, even when both groups are treated exactly the same. However, the larger the difference in the means, the less probable is such an outcome. Hypothesis testing relies on estimating the probability (or chance) of obtaining the results we measure.

If it is assumed that the null hypothesis is true (in this example, that the alcohol has no effect), it is possible to calculate the chance of getting any pattern of data that may result from the measurement of two randomly allocated groups of participants. Usually, this is done by using this assumption (that the null hypothesis is true) to calculate (or otherwise estimate) the sampling distribution of a statistic (such as T ratio or chi-square). This distribution is then used to compare with the sample statistic calculated from the measurements taken.

In principle, even if there is no effect of the treatment (so the null hypothesis is true), any mathematically possible pattern in the data may be found by chance alone, however unlikely. The smaller the probability of finding a given pattern of results by chance, the more we consider that we have evidence that supports the research hypothesis. Some statistical methods set a conventional probability level, below which the procedure allows the decision to be made that the null hypothesis may be rejected and the research hypothesis accepted. Other methods consider that there is no hard-and-fast decision rule that can be universally applied, and that the evidence obtained with one set of measurements can be fully evaluated only in the context of other research that has investigated similar problems in similar ways (see, e.g., MacDonald, 2002).

...