"Confidence Intervals" with *very* small samples.

94 Views Asked by At

I am collecting performance data for an application, I gather results each version.

Given that the measurement is imprecise, I'd like to calculate a confidence interval for the performance data.

Issue 1:
With multiple measurements I can apply a CI with a t distribution, as recommended for sample sizes less than 30. However my sample size is actually 2-4, is that too small?

Issue 2:
Many builds only run the performance tests once. How could I estimate error for that? I mean, technically there's an infinite error. But that is less than useful.

Honestly, I can't remember much from my statistics days. I'd simply like to graph an upper and lower bound, and the average.