If one observation xk is removed from n observations: New n = n-1. New sum = old sum - xk. New sum of squares = old sum of squares - $xk^2$. New mean = new sum / (n-1). New variance = new sum of $\frac{squares}{n-1}$ - (new mean)^2. This is similar to the correction technique but changes n as well. Always recalculate both sums before computing the new variance.