I would like to develop an indicator. I was able to research 33 values from a population of about 1000 objects. It is now easy to calculate the mean value and a confidence interval. Is this a legitimate procedure? Especially, plotted, the distribution of my values is rather U-shaped; can I assume a normal distribution at all? If not, how should I adjust my approach (e.g. should I first try to approximate the "true" -then presumably somehow quadratic- function from my data)? Is there a more appropriate method to build an indicator based on the available data?