The amalgamation and geometry of two-by-two contingency tables (Q1116235)

Let \(A_ i=(a_ i,b_ i,c_ i,d_ i)\), \(i=1,2,...,n\) be the two by two contingency table corresponding to the i-th of n mutually exclusive subpopulations with \(a_ i\cdot b_ i\cdot c_ i\cdot d_ i\neq 0\), and \(\alpha =\alpha (A)\) be a measure of association. If the n tables are gathered together, the process is called amalgamation obtaining a table \(A=(\sum^{n}_{i=1}a_ i\), \(\sum^{n}_{i=1}b_ i\), \(\sum^{n}_{i=1}c_ i\), \(\sum^{n}_{i=1}d_ i)\). The authors call the inequalities \[ \max_{1\leq i\leq n}\alpha (A_ i)<\alpha (A),\quad \min_{1\leq i\leq n}\alpha (A_ i)>\alpha (A) \] amalgamation paradox. By suitable designs of the sampling experiments it is shown how it can be avoided. Conditions for ``homogeneity'', i.e. \((a_ i+b_ i)/(c_ i+d_ i)=const\), \(i=1,2\), with respect to various measures of association are also investigated. Some proofs have interesting geometrical interpretations.