Need help with Sample Size (Percentiles)

[dbajric]

Hello,

I am trying to figure out how many measurements I have to take in order to be xx% confident that the results are within certain margin of error. I was able to find and use a formula that gives sample size required for results that are averages (mean), but we are shifting in using percentiles (e.g. 75th Percentile), and was wondering if there is a different approach in calculating the sample size for this type of data.

Currently, we have about 360 measurements per day, and calculating the 75th percentile from it. This number was chosen randomly, and I'd like to have a more scientific approach in defining these types of measurements.

Any help appreciated,
-D.B.

[fed1]

Draw the empirical cdf, together with its 95% CI. This is done the same way you would do for kaplan meier estimate, without censorship.

Now draw a line across the paper at Pr{T < t] = .75.

where it crosses the upper,lower CI, draw vertical line down to x axis. these are your upper and lower limits for the .75 percentile.

I think SAS will do this automatically, but i dont remember how. Look in proc lifetest documents.