You may not always be able to calculate the marging of error or confidence intervals, because the population your sample is from is not normally distributed. In such cases you need to pull yourself up (bootstrapping) with a different statistical technique. Here is how.
