On page 30: accuracy of chi square aproximation, . It is mentioned that: (However,the statistic is more robust than that and in practice a less conservative approach,such as ensuring that all ei's are greater than 1 and that not more than 20% of them are less than 5, may be taken) . What does this line means?
The requirement of chi-square test is that all expected values should be more than 5, however this is a loose requirement. So even if few expected values (not more than 20%) are less than 5, results of the test would still be valid. There are other tests which can be used when these requirements are not met. See these links https://www.researchgate.net/post/I...at_is_the_appropriate_statistical_test_to_use http://stats.stackexchange.com/ques...st-if-many-cells-have-frequencies-less-than-5
In Ct-3 we combine for all expected values less than 5. But even if few expected values are less than 5 (as long as they're less than 20%), we can carry on with the test. For this exam, just combine to make all expected values greater than or equal to 5.