> From: Mike > >In a company with several thousand PCs, many of which are no-name clones and > >few of which are ever turned off (many sitting in labs unattended and unused > >for weeks or months at a timer), I don't think we've ever had a fire... > > Great - but, please understand that your sample size is still very low. > rgds > mike > Perth, Western Australia As a really rough rule of thumb the reliability of a random sample from a normal distribution (ie the type we never see in nature :-)) is p ~ SQRT(1/N) so for a sample of "a few thousand" this is SQRT(1/2000) ~ 0.02 This is of course really really (really ...) rough but what it says is for a sample of a few thousand the real figure worst case is liable to be a few % off what you measure (but of course in reality could be yonks off what you measure due to non-statistical real world variation. Worst case errors occur when the sample is around 50% and are tighter at either end. In this case the sample is 0 (ie ...I don't think we;ve ever had..." so this suggests the result is liable to be fairly indicative. That said, if my office is found burn't out tomorrow morn by a wayward power supply, I won't be surprised!