Basic Introduction to Non-Parametric Tests
Parametric tests are among the statistical tests based on distribution assumptions. In scientific analysis, it is basically checked whether the relationships between demographic factors and variables are statistically significant.
Parametric tests are generally used in undergraduate statistics courses. An assumption must always be adhered to. We all remember the concepts of T-test, Z-test, ANOVA etc. However, in real-world problems, the data is not always normally distributed or does not fit the assumptions of parametric tests i.e. independence or homoscedasticity. In such cases, we use non-parametric tests.
I did my BA in Economics and we used to do almost everything based on assumptions. In my master’s thesis, I had to be quite involved with non-parametric tests. The table I used in this article, of course, does not contain the details. However, it can be a guide for you to understand the concept.
Non-parametric tests are tests that make no assumptions. They can be used easily in cases where parametric tests cannot be applied and allow analysis even in a very problematic data set.
When should we use it?
- When data do not follow necessary assumptions such as normality.
- When the sample size is too small. Because the smaller the size of the data, the harder it will be to follow the assumptions.
- Data are nominal or sequential. For example, surveys such as “Strongly disagree, Disagree, Neutral, Agree, Strongly agree”.
- The data is sorted. For example, an ordered list of products.
- If the data contains outliers.
My conclusion from my experience is that it is computationally expensive for large sample. But it will help to analyze various samples with minimal assumptions. Choose wisely!
