We use a technique called time sliding to generate more background data to understand our detector noise better, and thus set better upper limits on the false alarm rate. So even with a 3 month observation time we can claim 5+ sigma.
If so, would it be fair to say (as a simplification) that instead of increasing the confidence in GW events by collecting more of them, you're doing so by holding the number of collected events almost constant and increasing the confidence in the characteristics of noise the detector is reporting?
Yes, exactly that. We don't have the luxury of many events, so to increase our confidence we generate more background. One of the problems is what to do with data that contains more than one event. In the second detection paper, one of the plots contains the background with and without the other event taken into account. This is because during the time sliding process you will necessarily slide the time series from one detector across a gravitational wave signal in the other, producing false background noise.
There's no real standard way of dealing with this, so we show both cases - but both claim 5+ sigma.
