Couldn you use odd numbered events for a control sample (to develop the correction model), and even numbered events for the statistical tests?
The approach of splitting data into odd numbered events as control and even numbered events as signal needlessly throws away half of the collected data, at a cost of roughly 300 M$. Splitting data certainly does not help address systematic problems, since the systematic problem will appear in the control region just as in the signal region. The only reason splitting data into odd numbered events and even numbered events may be useful is to avoid sculpting of cuts, but Sleuth already solves this problem by having algorithmically defined regions in the single variable SumPt.
Related Questions
- Couldn you use odd numbered events for a control sample (to develop the correction model), and even numbered events for the statistical tests?
- Does Simetar© contain any statistical tests for validating the stochastic variables in the model?
- What is the required sample size of a data set. Using statistical tests (CHI squared)?
