The Mokken scale has been widely applied in many fields, including education, sociology and psychology. The scalability coefficient is an important index in Mokken scale analysis that is used to assess the homogeneity, hierarchy and consistency among the items in a scale. For a set of items, there are three types of scalability coefficient, Hij for each pair of items, Hi for individual items, and H for the entire scale. However, the scalability coefficient can not be computed with currently available software where missing data are present. This study proposes a method to derive the distribution of the item-pair scalability coefficient. The method computes all possible allocations of missing values and will not remove any missing data and retains the sample size. A real data example with randomly drawn missing observations is used to demonstrate the distribution of the item-pair scalability coefficient with our approach.