RecpCts is a redrawn plot showing the excluded
events in gray
Recpblockstudy are plots that show the effect of
blocking on the aggregate Z score.
The study looks at 2 recipes:
Recipe 2 (intra reg blocking) [could also say intra-node
blocking. I'm still looking for the right terms to communicate these things
clearly and succinctly]
Recipe 5 (inter reg blocking)
For each recipe we look at a range of blockings from 1 sec
to 1 hour. (shortest blocking for recipe 2 is 1 minute)
The study also looks at 2 weightings: event and
For the event weighting the event Zscores are combined as
a Stouffer Z
For the df weighting, the event chisqr for the
recipe is summed to give a total chisqr for the event set. This is then
converted to a Zscore.
Bty, it's helpful to recall that recipe 5 at 1-sec = our
Recipe 2 at 1-sec is the device variance
There is much info in these plots so it takes a while to
It's good to keep a broad view and remember that scores
less than 1.7 are insignificant.
Thus, the recipe 2 plots are telling us we see nothing
interesting happening with this recipe, over the whole set.
There is an interesting issue that I will develop in a few
coming plots: the recipe 2 events at 15 min resolution are slightly
significant if you look only at those events where you actually used this
recipe. When you turned later to using other recipes for events, the events
recipe2/15-minutes stopped having a sizeable effect size. In other words, up to
about event 39, recipe 2/15-min shows an effect. After that it doesn't.
Intereseting coincidence. You can see this if you dig into the plots recipe 2 at
15min and recipe 5 at 1-second, but I'll make it clear in some separate plots. A
start is plot "flip 1-2"