Asymptotic vs Exact


Im wondering what is the difference between Asymptotic and Exact generated in SPSS and which one that I should use for non-parametric tests (wilcoxon, mann-whitney, kruskall-walis), chi-square and binomial test.

Any suggestion please? Thank you.


New Member

SPSS calculates the significance of the wilcoxon signed-rank, wilcoxon rank-sum, M-W and K-W using the Asymptotic Method by default. This method is considered to be accurate with large samples, but less so with smaller samples. In both large and small samples the Exact test, as its name suggests, is the most accurate way to calculate significance but with large samples the computations can take a while through SPSS. Hence the Monte-Carlo test method can be useful for large samples (as it's quicker), and the Exact test is usually a better option for small samples. To specify which method you want to use to calculate significance, click on the Exact tab in the dialogue box of whichever test you are using.

I always hear that when the sample size is small, asymptotic results may not be valid. But, how small is small? Is there any definition? Thank you.