Questioning Null-Hypothesis Significance Tests
- Details
- Written by FEB UGM
- Category: News
- Hits: 3026
The use of Null-Hypothesis Significance Tests (NHSTs) in research has become a common thing done by many researchers from various majors, including economics and business. However, criticism has leveled a lot of about the use of NHSTs in research because of some weaknesses. In a public lecture entitled “The Case Against Null-Hypothesis Statistical Significance Tests: Flaws, Alternatives and Action Plans”,(30/04/14) Andreas Schwab, one of the authors of the journal Researchers Should Make Thoughtful Assessments Instead of Null Hypothesis Significance Tests deepen the explanations about it.
Significance in the research is crucial. However, the result of significant estimation in statistics is not necessarily important. This is due to that the broader perspective of formal statistics is very vulnerable so that the end result which is obtained - even though it’s statistically significant - has not certainly proved to be accurate. This is shown by the number of researches using NHTs, which have a significant outcome, will have different results if they are being replicated later. If so, it can be concluded that significant results using NHTs not necessarily credible.
In addition, the causes of NHSTs’ bias is because NHTs also heavily influenced by the amount of data and number of variables. When the researchers increased the number of the data, they could have been obtaining significant results. Whereas, they obtained insignificant results before. The addition of variables are also capable of making the estimation result become significant. Actually, there is a possibility that the added variables are irrelevant to the examined variable. According to Schwab, NHTs are only able to answer whether there is influence between one variable with another, not on how much influence it has. “NHSTs do not answer the questions we are really interested in” he add.
Therefore, there are a number of alternative options that can be performed by researchers, especially in the fields of Economics and Business in testing the significance of the results of the estimation. First, researchers need to adapt to the context of the research hypothesis. Researchers should be able to expose any variables that may have a relationship with the examined dependent variable with the help from previous researches or theories. Researchers also need to use the effect size, which means that the extent of the effect given by one variable to the examined variable. Furthermore, the researchers can use the baseline to compare the new data model. Finally, it is recommended that researchers use a simpler model to support generalizations and capabilities of replicated research later.
Sources: Nadia/FEB