The above figure includes the results of significance testing that Dataset Explorer performs:

Significance testing is designed to indicate whether the reliability of the statistics is 95% or greater, based on p-value.
Dataset Explorer calculates the significance result using either t-test or chi-squared statistics to determine the p-value:
■ For continuous variables (for example, subject weight or age), a t-test compares the observed values in the two subsets.
tranSMART uses the following Java method to calculate the t-test statistic:
■ For categorical values (for example, diagnoses), a chi-squared test compares the counts in the two subsets.
tranSMART uses the following Java method to calculate the chi-squared statistic:
If there is not enough data to calculate a test, Dataset Explorer displays a message indicating the lack of data. Also, significance test results are not displayed in the following circumstances:
■ If two identical subsets are defined. In this case, the significance test results are not meaningful.
■ If all subjects in the first subset have one set of values for the categorical value, and all subjects in the second subset have other categorical values. For example, suppose you set Subset 1 to contain only males and Subset 2 to contain only females. Also, suppose that Subset 1 has 15 subjects and Subset 2 has 20. If you then try to show statistics by gender, a table like the following would result:
|
|
Subset 1 |
Subset 2 |
|
Female |
0 |
20 |
|
Male |
15 |
0 |
In this case, the chi-squared function doesn’t return meaningful results.