Glossary items: click on the icon to go directly to the glossary entry
You should use a two-way Chi-square test when:
- Dealing with the frequencies classified according to two sets of categories in a contingency table
- Each item sampled contributes to only one cell in the contingency table
- Looking for a difference between an observed frequency distribution of a single sample and an expected frequency distribution based on no association between the two variables
Do not use this test when:
- Using descriptive statistics ,
percentages
or anything other than frequencies
- Expected values include small values. As a rule of thumb: If any expected value is zero or more than one fifth of the expected values are less than five, the chi-square test becomes unreliable. (Note that's expected values: it does not matter what the observed values are)