Null Hypothesis

A null hypothesis is one in which no difference (or no effect) between two or more variables is anticipated by the researchers. This follows from the tenets of science in which empirical evidence must be found to disprove the null hypothesis before one can claim support for an alternative hypothesis that states there is in fact some reliable difference (or effect) in whatever is being studied. The null hypothesis is typically stated in words to the effect that “A equals B.” The concept of the null hypothesis is a central part of formal hypothesis testing.

An example in survey research would be a split-half experiment that is used to test whether the order of two question sequences within a questionnaire affects the answers given to the items in one of the sequences, for example, in crime surveys where fear of crime and criminal victimization experience are both measured. In this example, a researcher could hypothesize that different levels of fear would be reported if the fear items followed the victimization items, compared to if they preceded the victimization items. Half the respondents would be randomly assigned to receive one order (fear items, then victimization items), and the other half would receive the other order (victimization items, then fear items). The null hypothesis would be that the order of these question sequences makes no difference in the answers given to the fear of crime items. Thus, if the null hypothesis is true, the researcher would not expect to [Page 542]observe any reliable (i.e. statistically significant) difference in levels of fear reported under the two question ordering conditions. If results indicate a statistically reliable difference in fear under the two conditions, then the null hypothesis is rejected and support is accorded to the alternative hypothesis, that is, that fear of crime, as reported in a survey, is affected by whether it precedes or follows victimization questions.

Another way of understanding the null hypothesis in survey research is to think about the crime survey example and the confidence intervals that can be calculated around the fear of crime measures in the two ordering conditions. The null hypothesis would be that the 95% confidence intervals for the fear measures under the two orders (conditions) would overlap and thus not be reliably (significantly) different from each other at the .05 (alpha) level. The alternative hypothesis would be that the confidence intervals would not overlap, and thus the fear measures gathered under one order are reliably different from the same fear measures gathered under the other order.

Rejecting a null hypothesis when it is in fact true is termed a Type I error. Not rejecting a null hypothesis when it is fact false is termed a Type II error.