# Hypothesis testing by random sampling

I'm sure this has been asked before, if so please point me to the relevant answers. Anyway, I have about 26000 phone subscription entries in a database. A number of these might contain wrong information that I can't verify automatically. I have no time to manually check each one of them. What I'd like to know is how many of these I should randomly sample and manually check in order to be 95% certain that there are at most 1% entries with errors? Preferrably I'd like to have a formula where I can tweak the numbers. Is this possible? I am grateful for any pointers, links, and answers. Thanks in advance!

-

You can't say a priori how many subscriptions you'll have to check in order to be $95\%$ certain that there are at most $1\%$ errors, since this depends on the error rate. If the error rate is close to $1\%$, you'll need to check a lot more to distinguish it from $1\%$ than if it's $0\%$.
I figured I can use the $3/n$ formula. If I don't find any defects while checking 300 samples, it would mean I am $95\%$ certain of the $1\%$ error rate, right? –  Daniel Lidström Apr 20 '12 at 12:09