Hi Roger,
Here is the nulls graph
I will be in Limges through the weekend and out of touch for the most part
I'll be back on Monday.
There are some more results, for which I dont yet have plots.
I redid the eventslide at finer resolution and was surprised to see
that the z-score looses its significance (basically below 2) if I step
just 5 minutes off the predicted exam periods. This is maintained
even if I do things like take subsets of events that exclude the shorter ones
This is a result that is probably telling us something.
I also looked at other subsets
An interesting one is to simply take the first half of events and compare with the second half. The effect is not significant for the last 85 events: z=1.3 vs z=4.4 for the first 85. Hmm. Why so different? The mean diff of these groups is sig at about the 0.01 level, not so huge and I will refine this. My feeling is that it not really a sig result.
We should look again at the diff between events you predicted and the ones third parties predicted. what's noted on the results page as prediction source is need to be used with care, but it does give a rough measure of your input. Here there seems to be something but it's prob not sig. Your 108 solo predictions have event cumulative of z=4.0 The 62 third party's come in at z=1.4. But the (rough) mean diff pval for these is only about 0.1. I'll do a more thorough job when I return.
Here is a blocking fig in JPEG. Best I could do for now. I'll look again next week.
cheers,
Peter