Network Equipment Availability and Significant Figures looking for assistance in solving what I thought would be straightforward question but am struggling with. 
I monitor large numbers of Network gear including switches, routers and firewalls for availability. I report monthly from a network monitoring system database on combined uptime availability using 5 significant figures, for example 99.742% 
I am concerned that this may overstate the precision based on our polling frequency (we check most devices every two minutes but some longer, the simple average is 169 sec.)
Is there a formula or methodology I can use to determine polling frequency to report 5 significant figures? Similarly what is the number of significant figures I could legitimately report using the current polling?
Thanks for your time.
 A: The typical rule for significant figures when multiplying or dividing is to use the lesser of the numbers of sig figs of the arguments. In this case, you have infinitely many sig figs because each measurement is 0 or 1 and the divisor is an exact count of observations. So, sig figs don't seem to help.
Besides, sig figs tend to underrepresent the uncertainty involved in measurement. Statistics is concerned with sampling error, which typically implies a lot more uncertainty than would be expressed by rounding an estimate to the correct number of sig figs. In other words, merely rounding your availability number (based on, say, the number of measurements) would likely suggest your estimate is a lot more precise than it really is, because you don't know the true long-run availability; you only have a sample.
A common tool in statistics for expressing uncertainty is a confidence interval. A 95% confidence interval (CI) for some unknown quantity (e.g., availability) is an interval (e.g., [.991, .995]) such that on average, 95% of such intervals will contain the true value. So instead of reporting one number, you could report a CI. CIs can be reported as a range or as a midpoint plus a margin of error (e.g., .993 ± .002).
In simple cases, calculating a CI for a proportion is easy. Accounting for the fact that your proportions are grouped into devices would take a hierarchical model or an appropriate bootstrapping scheme. For the complication of computing new CIs as you accumulate data, with the goal of stopping data collection once you have enough precision, see:
Jennison, C., & Turnbull, B. W. (1989). Interim analyses: The repeated confidence interval approach. Journal of the Royal Statistical Society, 51, 305–361. Retrieved from http://www.jstor.org/stable/2345448
