Hello Ruddy,

The usual guideline is that expected cell frequencies should be at least 5 for the observed chi-square statistic to follow the theoretical chi-square distribution with accuracy. So first, check f(e) values rather than f(o) values for your data set (multiply row total n by column total n for that cell, then divide by grand total, N).

For example, the 2 x 2 table:

15 3

20 25

has one cell with observed frequency of 3, but that cell's expected frequency is 8 (= 18 * 28 / 63)

A downside of having generally low cell frequencies is that the statistical power of the chi-square (or any of the related tests, such as Fisher's exact test or likelihood ratio) will be low. SPSS and other packages do offer an "exact" computation of 1- and 2-sided tests for 2x2 contingency tables, which should be accurate regardless of f(e) sizes.

Good luck with your work!

Hello Ruddy,

The usual guideline is that expected cell frequencies should be at least 5 for the observed chi-square statistic to follow the theoretical chi-square distribution with accuracy. So first, check f(e) values rather than f(o) values for your data set (multiply row total n by column total n for that cell, then divide by grand total, N).

For example, the 2 x 2 table:

15 3

20 25

has one cell with observed frequency of 3, but that cell's expected frequency is 8 (= 18 * 28 / 63)

A downside of having generally low cell frequencies is that the statistical power of the chi-square (or any of the related tests, such as Fisher's exact test or likelihood ratio) will be low. SPSS and other packages do offer an "exact" computation of 1- and 2-sided tests for 2x2 contingency tables, which should be accurate regardless of f(e) sizes.

Good luck with your work!

Indeed the first step is calculating the expected values. If there is cells

Luckily there is also a python implementation of the fisher exact : https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.fisher_exact.html