Understanding Look Elsewhere Effect in Particle Physics

[ChrisVer]

I would like to ask for some guidance concerning this effect. So far I read the article in wiki and what is written about it in the cms page.
As an overall, taking the "definition" from the paper cited by the cms page, "When searching for a new resonance somewhere in a possible mass range, the significance of observing a local excess of events must take into account the probability of observing such an excess anywhere in the range."
However I am not sure I understand how is that possible? obviously in the data someone takes, there will be regions without a significant signal, and maybe regions with a significant signal. For example the m_{ee} for the Z\rightarrow ee will show a peak at ~91 GeV with a width of ~2GeV. Is there any trial factor considered in this Z example? (of course if we say that there isn't, I don't see a reason for such a trial factor to play a role in Z' searches).
One difference is that the LEE appears (let me rephrase it) "when we don't know where our signal is supposed to be". But wasn't that true for all our searches? At first we never knew where a signal was supposed to be, and then we do know..we didn't know that the Higgs was at 126GeV, yet now we do...So I am getting a little confused with when a LEE takes place and when it doesn't.

One additional "problem" I have is that I don't understand how searches (trials) that have not been published affect the data taken from the detector. My problem is that I don't find the connection between the older to the newer analysis... If today I measure 100 events at mass 1TeV, and I don't publish it, how can this lead you in measuring 1000events at 1TeV the next day?

Thanks.
PS. I know this concerns mostly particle physics, but I think the effect itself is a statistical one, that's why I chose this sector to post.

 

[Dale]

Without getting too much into the physics, the look elsewhere effect is just another name for multiple comparisons.

A p value tells you the probability of randomly detecting your signal if there is really no signal. So if your p value is 0.05 then there is only a 5% chance that you "detect" a signal that isn't there. The problem comes if you are looking for a lot of different signals. If each signal is independently p=0.05, and if you look for 14 different signals, then the chance that you randomly detect at least 1 signal is about .5

 

[Stephen Tashi]

These notes have some suggestions: http://www.stat.berkeley.edu/~mgoldman/Section0402.pdf