N>30, or not! This is the question!


I am very confused about how to select the appropriate statistical method (parametric or not) by considering N number in the dataset. Some references say that if the N number is below 30, parametric tests cannot be used but some others say that it is not necessary to reach 30 subjects to perform a parametric test as long as the dataset shows a normal distribution.

Can you please provide simple explanations and/or suggest some references that supports both points of view?

Any help will be appreciated.

You may start your reply by giving a short answer as "Yes." or "No." (any other short answers are also welcome) to the question below and then add the details.
Is it necessary to have N>30 to perform parametric tests?


Well-Known Member

A parametric test is very general term ...

For example, the two-sample t-test assumes normal distribution for the averages.
If the population data distribute normally, you can use a t-test with very small sample size. (ignore the potential power issue, depend on the required effect size)

In case the data doesn't distribute normally and the sample size is sufficiently large the average will distribute normally per the Central Limit Theorem.

What is sufficiently large? the rule of thumb is 30.
If the distribution is "reasonably" symmetrical, it may be a smaller number. If the data is very skewed it may be larger.
Last edited: