Relatives may tend to die of the same cause. A model is proposed specifying the degree of this association, simultaneously allowing for misclassification of cause of death. These mistakes usually degrade any existing correlation structure, the effects being marked when error rates reach between 20% to 30%. It is then that the search for associations becomes very difficult indeed, and results misleading. This is an example of a wider malaise that can afflict analysis of association, and the developments here apply to two-by-two contingency tables in general.
Tallis, G. M.
"Misclassification, Correlation, and Cause of Death Studies,"
1, Article 6.
Available at: http://digitalcommons.wayne.edu/humbiol/vol74/iss1/6