Here's a look at PRN26's atomic clock behavior for three days prior to it going out. Erratic clock behavior is one of the likely causes of an outage and is typical of an aging clock. According to Richard Langley's GPS status page, this is a Rubidium atomic clock and was turned on in July of 1992 - which makes it almost 20 years old - 10 years beyond its design life! This is the second oldest operating clock in the entire GPS constellation. (The oldest clock belongs to PRN 32, turned on in December of 1990)
Ideally, this pink line will be kept as close to zero as possible. The amount of error in shown in this plot is incorporated into your GPS receiver and can be seen as errors in your position determination. The US Air Force adjusts the satellite's clock values to keep this line as close to zero as possible. A typical adjustment can be seen about 12 hours into Feb 07. Normally the clock values move around a bit, but about 6 hours into Feb 08 the clock jumped dramatically. Each dot on the line represents 15 minutes, and clock errors of several meters (1 meter = 3.34 nanoseconds) in that short of a time are unusual. You can see that the clock is settling down between jumps but keeps having erratic episodes until, around 1800 hours on Feb 09, it starts changing by 1 meter every 15 minutes. Finally, the Air Force took it off the air at 1917 UTC.
There's always a balance between keeping a satellite on the air (more satellites in view means a better position determination for you) and turning off a satellite that is harming your position. So, as the Air Force watches each 15 minute update, they have to decide to keep it on, or turn it off. Good call on this satellite and let's hope PRN 26 is healthy again soon.