摘要: SUMMARY Goodman's work on quasi-independence and missing values for contingency tables is unified so as to cover in general incompleteness, here defined. Cyclic computational procedures are provided maximum likelihood fitting of iincomplete data. Methods given determining appropriate degrees freedom, whether by identifyinig separable or semiseparable subtables elimination cell isolates. There discussion how inicompleteness might arise other contingency-table situations, e.g. testing symmetry square tables. But it suggested that use some power-increasingf test be preferable detecting departures from symmetry.