Abstract:
|
For an incomplete two-way contingency table including units missing cases, we propose a new identification rule for missing mechanisms using cell probabilities concerning missing data. We firstly introduce a variable-based missing mechanism to clarify the missing mechanism in two-way contingency tables in general selection and pattern mixture models. Next, we suggest two missing ratios to show the non-identifiability between missing-at-random and missing-not-at-random, as well as to characterize the missing mechanism. Simulation studies show that the missing mechanism criterion works well for large samples.
|