Chi-Squared Test

A very useful, robust, simple test which is worth knowing about. This test handles classification data eg. lived v died. As most frequently used the test is based around a 2 X 2 crosstabulation table although one can have any number of rows and columns. Consider the following table:

There are two groups of patients - one treatment and one control. These are then subclassified into those who lived and those who died. In practice you will use a computer to perform the calculations so I will list the logical steps:

1. Invent a null hypothesis that there is no significant difference between the survival rate in the two groups.
2. Calculate how many patients would have lived and died in each group if only chance were at play - the expected values.
3. Subtract the observed rates from the calculated expected values ( called Chi squared ) and assess the significance of the difference using statistical tables.
4. If there is less than a 1:20 chance ( p=0.05 ) the null hypothesis is rejected and a significant difference presumed.