Estimation for the distribution of two-dimensional discrete scan statistics (Q861535): Difference between revisions

Let \(X_{ij}\) be i.i.d. nonnegative integer random r.v.s, \[ \nu_{ts}=\nu_{ts}=\sum_{i=t}^{t+m_1-1}\sum_{j=s}^{s+m_2-1} X_{ij}. \] Then the two-dimensional discrete scan statistics is \[ S=S(m_1,m_2,N_1,N_2)=\max_{0\leq t\leq N_1-m_1; 0\leq s\leq N_2-m_2} \nu_{ts}. \] The authors propose a new method for approximate calculation of \(P(S\leq n)\) for any fixed \(n\). It is based on an approximate representation of \(P(S\leq n)\) through the probabilities of the type \(P(S(m_1,m_2,a_1,a_2)\leq n)\) where \(a_i=2m_i\) or \(3m_i\). Such probabilities are estimated via simulations. Numerical estimates for errors in these approximations are given. The method is compared to alternative ones on numerical examples. It's generalizations to ``non-rectangular'' windows (regions of summation for \(\nu_{ts}\)) are discussed.