Petra,

Your opening up a can of worms with that question, but thats okay. Its better to know ahead about what is expected then to be surprised.

In the past I had always been posting predictions with 10 day windows. Seeing that most events happened within 3 - 7 days of the signal I started taking a chance and reducing the 10 day window down to 8 days in order to decrease the probability of getting a hit by chance. On one of my predictions back in January 1996, I went with an 8 day window and when the 8 days went by, I went and increased the window another 2 days. Alan didn't go along with this and treated the first 8 day window as a single prediction and the subsequent 2 day extension as an additional prediction. Both missed. So instead of getting 1 miss I got 2 misses. This doesn't bode well when you are trying to get to the statistically significant category and caused me to become significantly frustrated. For quite awhile after that episode, I made a half-assed (excuse my french) attempts with subsequent predictions causing additional misses. It was like, okay, what the heck, it isn't going to get anywhere, so why try and be right on.

So based on the above, you can see that once a prediction is made, it isn't acceptable to the scientific community to make any changes to the prediction once it has been made. Personnally I disagree with this as a methodology may receive additional information after the prediction is posted that can have a significant impact on what was posted. For instance, based on initial indications, you think that you have an M6.0 event and later indications indicate an M6.9 event. In the event of this kind of scenario, do we want the public to think that a less serious quake is going to occur? But then again, we are only at a stage of trying to validate a methodology instead of trying to warn the public. But then again, I think that both should work hand in hand where the public would know that it is a possibility that a large quake may occur much like weather reports. Heck, the weather reports are never right.

As to the reasons for my misses, I sometimes have a problem with weeding/separating out multiple events close in space and time. This causes me to think that I have a larger quake getting ready to occur when instead its two smaller quakes. For instance, the very first miss that I had under Alan, I had posted a prediction for an M4.0 - M5.5 quake within 100 miles back in January 1995. Within a couple of days twin M3.8's occurred of the Santa Monica coast. In a January 1996 prediction, when Alan changed my extension into a second prediction, we were experiencing a several thousand quake swarm at Mammoth. So, its things like this that play havoc with my methodology and hurt any chance of becoming statistically significant (or at least until I'm able to figure out a way to separate out individual smaller quakes). There may be others ways around this as well in how a prediction could be posted and evaluated to cover these kinds of scenarious, but I need to get together with a statistician to come up with a way to do this.

Hope I've answered your questions.

Dennis