Incomplete Two-Dimensional Contingency Tables
Yvonne M. M. Bishop, Stephen E. Fienberg · Biometrics · 1969
This paper examines the estimation of expected cell counts in two-dimensional contingency tables when some of the cells are known a priori to be empty. A log-linear model, corresponding to the model of independence or homogeneity for 'complete' tables, is fitted to the cell counts. The resulting cell estimates can always be computed by an iterative proportional fitting method, although some tables can be fitted directly. The direct method of fitting is applied to a set of medical data. It is pointed out that the iterative proportional fitting method can readily be extended to tables of higher dimensionality.