6.7 Starting with a Null Hypothesis
Often based on previous theories or observations, a hypothesis is a tentative logical explanation about something that can be tested. As such, stating a hypothesis such as “sunlight repels vampires” is not good, because we cannot find any vampires to test our hypothesis on. On the contrary, stating a hypothesis such as “having a pop-up web advertisement does not increase the number of subscribers” is meaningful because it can be verified with a T-test.
Before we conduct any statistical test, we begin by formulating a null hypothesis. The null hypothesis (H0) represents our starting assumption prior to conducting the test while the alternative hypothesis (H1) states the opposite, and is usually the one you are trying to prove. Examples include:
Example 1:
| Null hypothesis (H0) | Having a pop-up web advertisement does not increase the number of subscribers to LobsterLand.net |
| Alternative hypothesis (H1) | Having a pop-up web advertisement increases the number of subscribers to LobsterLand.net |
Example 2:
| Null hypothesis (H0) | Working from home does not improve job satisfaction |
| Alternative hypothesis (H1) | Working from home improves job satisfaction |
A researcher should always clearly state his or her null hypothesis when presenting research. While the null hypothesis represents the status quo, or the starting assumption, the researcher’s starting assumption is not always immediately clear to everyone else who might read or interpret the results of an experiment. The null hypothesis should be phrased in a way that is easy to understand. For instance, there should be no double negatives such as “dog owners are not more unhappy than cat owners.”
After identifying the null and alternative hypotheses, evidence needs to be collected before you can reject or accept the null hypothesis. The type of statistical test that we use to verify our hypothesis depends on the type of data we have. For instance, if we wanted to test for a relationship between education level and marital status based on a survey, then a chi-squared test of independence would be an appropriate test to use on responses to these two questions: What is the highest level of education you have obtained? What is your marital status?; suppose you had numerical survey data generated from a 5-point scale, then a statistical test such as a t-test would be appropriate.
That said, all statistical tests have assumptions associated with them, making it imperative for the researcher to check the data to ensure none of these assumptions are violated before beginning the analysis and drawing conclusions from the data. We will elaborate on some of the commonly used statistical tests and their assumptions later on in the chapter.
From the results of our statistical tests, we will be able to obtain a p-value. A p-value will enable the researcher to decide how to handle the null hypothesis – whether to reject it, or to fail to reject it. The way to properly interpret the p-value is, “Given that the null hypothesis is true, what is the chance that we could obtain a test statistic this extreme?” Therefore, very low p-values cast doubt upon the null hypothesis.