zero variance for null samples
- t-test shows a large value when the variance of null samples is zero. t-test says that the target samples are louder than the null samples. However, the normal people in LIGO do not care such small thangs rather care about the evident witness. Note that on the other hand, t-test shows that tiny small elevation can be picked up in channels thanks to t-test.
- a witness channel is not always responsible for all the glitches. If that channel is always zero in null samples and non-zero in the target samples, then t-test can pick up the channel, for example, the cWB trigger glitches.
- If someone treats the cWB trigger as a whole, it is expected to be difficult to pick up witness channels because a channel is not a witness of all the glitches.