TerryM--re

Error in Sheppard's correction

Your additions to Sheppard's correction had serious problems and I've deleted them. Notice that the estimate of the variance of a normal distribution is made larger, not smaller, by rounding, because the rounding error is negatively correlated with the observation. For the uniform distribution what you suggest would make sense, but Sheppard's correction is biased in favor of other sorts of distributions. Michael Hardy (talk) 21:57, 16 June 2016 (UTC)