The score test for independence in R x C contingency tables with missing data

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• Missing data in contingency tables pose analytical challenges.
• Existing methods like likelihood ratio tests can be computationally intensive.
• Accurate independence testing is crucial for epidemiological and statistical research.

Purpose of the Study:

• To propose a novel score test statistic for R x C contingency tables with missing data.
• To provide a computationally simple and noniterative method for testing independence.
• To extend the score test for ordinal contingency tables and conditional independence.

Main Methods:

• Development of a score test statistic for independence in R x C tables with missing data.
• Approximation of the statistic's distribution to chi-squared under the null hypothesis.
• Extension of the score test for ordinal variables and conditional independence.
• Application of the methods to a subset of data from the Six Cities Study.

Main Results:

• The proposed score test statistic approximates a chi-squared distribution with (R - 1)(C - 1) degrees of freedom.
• The statistic is computationally simpler and noniterative compared to the likelihood ratio statistic.
• Extensions provide natural parallels to the Mantel-Haenszel statistic for conditional independence.

Conclusions:

• The proposed score test statistic is a viable and efficient tool for analyzing R x C contingency tables with missing data.
• The method offers advantages in simplicity and computational ease.
• The extensions facilitate robust analysis of ordinal data and conditional independence.