Should we accept the best split?
t statistic for t test
pooled variance