Be careful with what the article is saying. The data shows it works well with a normal distribution. It cites CLT, which also presumes (although they forget to mention, as usual
) that it must be from independent characteristics.
Since runout is not normal, nor are you likely sampling a population, but rather a continuous process, the caveat in their conclusion is especially meaningful:
If the underlying distribution deviates from normal distribution to a highly skewed distribution, then the accuracy of the 95% confidence statement with sample size obtained from the square root of N plus one rule decreases.
Since runout is not normal, nor are you likely sampling a population, but rather a continuous process, the caveat in their conclusion is especially meaningful:
If the underlying distribution deviates from normal distribution to a highly skewed distribution, then the accuracy of the 95% confidence statement with sample size obtained from the square root of N plus one rule decreases.