A hypothesis is a proposed statement to explore a possible theory. Many studies in the fields of social sciences, sciences, and mathematics make use of hypothesis testing to prove a theory. Assumptions in a hypothesis help in making predictions. It is presented in the form of null and alternate hypotheses. When a hypothesis is presented negatively (for example, TV advertisements do not affect consumer behavior), it is called a null hypothesis. This article explains the conditions to accept or reject a hypothesis.
Why is it important to reject the null hypothesis?
A null hypothesis is a statement that describes that there is no difference in the assumed characteristics of the population. For example, in a study wherein the impact of the level of education on the efficiency of the employee need to be determined, null (Ho) and alternate (HA) hypothesis would be:
In the above-stated null hypothesis, there is very little chance of a relationship between both the variables (education and employee’s efficiency). When a null hypothesis is accepted, it shows that the study has a lack of evidence in showing any significant connection between the variables. This could be due to problems with the data such as:
- high variability,
- small sample size,
- inappropriate sample and,
- wrong data testing method.
Hence, for efficient, appropriate, and reliable results, it is suggested to reject the null hypothesis.
Conditions for rejecting a null hypothesis
Rejection of the null hypothesis provides sufficient evidence for supporting the perception of the researcher. Thus, a statistician always prefers to reject the null hypothesis. However, there are certain conditions which need to be fulfilled for the required results i.e.
Condition 1: Sample data should be reasonably random
A random sample is the one every person in the sample universe has an equal possibility of being selected for the analysis. Random sampling is necessary for deriving accurate results and rejecting the null hypothesis. This is because when a sample is randomly selected, characteristic traits of each participant in the study are the same, so there is no error in decision making. For example, in the sample hypothesis, instead of collecting data from all employees, the data was collected from only the board members of the company. This hypothesis testing would not provide good results as the sample does not represent all the employees of the company.
Condition 2: Distribution of the sample should be known
A dataset can be of two types: normally distributed or skewed. Normally distributed datasets require application of parametric tests i.e. Z-test, T-test, χ2-test, and F-distribution. On the other hand, skewed dataset uses non-parametric test i.e. Wilcoxon rank sum test, Wilcoxon signed rank test, and Kruskal Wallis test. For reliable hypothesis test result, it is essential that the distribution of the sample be tested.
Condition 3: Value of test statistic should not fall in the rejection region
Test statistic value is compared with critical value when the null hypothesis is true (critical value). If the test statistic is more extreme as compared to the critical value, then the null hypothesis would be rejected.
For example, in the sample hypothesis if the sample size is 50 and the significance level of the study is 5% then the critical value for the given two-tailed test would be 1.960. Hence, null hypothesis would be rejected if,
Condition 4: P-value should be less than the significance of the study
P-value represents the probability that the null hypothesis true. In order to reject the null hypothesis, it is essential that the p-value should be less that the significance or the precision level considered for the study. Hence,
- Reject null hypothesis (H0) if ‘p’ value < statistical significance (0.01/0.05/0.10)
- Accept null hypothesis (H0) if ‘p’ value > statistical significance (0.01/0.05/0.10)
For example, in the sample hypothesis if the considered statistical significance level is 5% and the p-value of the model is 0.12. Hence, the hypothesis of having no significant impact would not be rejected as 0.12 > 0.05.
Important points to note
While making the final decision of the hypothesis, these points should be noted i.e.
- A large sample size i.e. at least greater than 30 should be considered. As per the Central Limit Theorem (CLT) large sample size i.e. at least greater than 30 is considered to be approximately normally distributed.
- For deriving the results either p-value approach or rejection approach could be used. However, the p-value is a more preferable approach.
- Statistical significance should be maintained at a minimum level.
- The choice of the rejection region should be appropriately made by verifying the direction of the alternative hypothesis.
- Investment diversification with momentum hybrid performance - April 14, 2021
- Medium-term momentum analysis of growth, income and value stocks - March 25, 2021
- Conditional CAPM analysis for stock-based investing - February 20, 2021