Simple linear approximations to the likelihood equation for combining evidence in multiple 2\(\times 2\) tables: A critique of conventional procedures (Q1087275): Difference between revisions

For \(k=1,...,K\), the variates \(X_ k\) and \(Y_ k\) are binomial with incidence probabilities \(p_ k\) and \(q_ k\) respectively. Several observations are taken on each of the variates. Define \(\beta_ k\) as \(\log (p_ k(1-q_ k)/q_ k(1-p_ k))\). Define the following three models: \(M_ 0: \beta_ k=0\) for all k; \(M_ C: \beta_ k=\beta\) for all k; \(M_ S: \beta_ k's\) are arbitrary. Various procedures have been proposed to estimate \(\beta\) in model \(M_ C\), to test \(M_ 0\) against \(M_ C\), and to test \(M_ 0\) against \(M_ S\). Linear approximations to the log-likelihood functions in the three models are developed, and are used to study the properties of the various tests and estimators.