THE ESTIMATOR for the duration of completed unemployment spells used by Clark and Summers is upwardly biased by at least half a month. They implicitly assume unemployment spells ending during a month continue until the last day of that month when a more reasonable assumption would be that they end in the middle of the month. In a world in which cohort escape rates were constant, this would result in a bias of about half a month. In the real world, where escape rates decline with duration, the bias is larger. Correcting the bias in the completed spell distribution would lower the estimates of the proportion of unemployment due to spells greater than or equal to a given length.
