Estimation of confidence intervals for Multinomial proportions of sparse contingency tables using Bayesian methods

Multinomial distribution, widely used in applications with discrete data, witnessed varieties of competing intervals from frequentist to Bayesian methods, still prove to be interesting in the case of zero counts or sparse contingency tables. The methods commonly recommended in both approaches are considered based on its influence of zero counts, polarizing cell counts, and aberrations. The inference based on comparative study shows that Bayesian approach, with an appropriate prior could be a good choice in dealing with a sparse data set without any imputation for zero values.