The trouble is that it is easy to remember successes and forget failures. Last year (which was a very quiet year) there were 18 deadly quakes, which makes for one every 20 days, and if you allow yourself a little latitude you can get quite good results from random signals at that frequency. A number of the events that were counted as fulfilling predictions were not even destructive, only felt.

Also one looks for a correlation between strength of signal and size of event (either in terms of magnitude or casualties). This is claimed for Latur 1993 and Kobe 1995 (on the basis of fatalities) but has been missing since. No big signal preceded Colombia, Izmit or Gujarat so far as I recall.