Indices of Dependence Between Types in Multivariate Point Patterns (Q4956052)

A multivariate point pattern is a spatial pattern of points which belong to one of a finite number of distinct types [\textit{D.R. Cox} and \textit{P.A.W. Lewis}, Proc. 6th Berkeley Sympos. Math. Statist. Probab., Unif. Calif. 1970, 3, 401-448 (1972; Zbl 0267.60057)]. To investigate dependence between different types of points the usual approach begins by estimating ``cross-type'' versions of the standard summary functions \(G\) and \(F\). The purpose of this paper is to pursue an alternative. In their previous paper, Stat. Neerl. 50, No. 3, 344-361 (1996; Zbl 0898.62118), the authors introduced a new summary function \(J\) for univariate (single-type) point patterns \(X\) as \[ J(t)=(1-G(t))/(1-F(t)) \] defined for all \(t\geq 0\) with \(F(t)\neq 1\). Here the ``empty space function'' \(F\) is the distribution function of the distance from an arbitrary fixed point \(0\) to the nearest point of the pattern \(X\), and the ``nearest neighbour distance function'' \(G\) is the distribution function of the distance from typical points of \(X\) to the nearest other points of \(X\). The \(J\)-function is an index of spatial interaction, identically equal to \(1\) for a Poisson process, and generally takes values less than \(1\) for clustered patterns and greater than \(1\) for ordered patterns. An appealing property is that the superposition \(X=X_{1}\cup X_{2}\) of two independent point processes \(X_{1}, X_{2}\) has \(J\)-function \[ (1)\qquad J(t)=(\lambda_{1})/(\lambda_{1}+\lambda_{2})J_{1}(t) + (\lambda_{2})/(\lambda_{1}+\lambda_{2})J_{2}(t), \] where \(J_{1}, J_{2}\) are the \(J\)-function of \(X_{1}, X_{2}\), respectively, and \(\lambda_{1}, \lambda_{2}\) are their intensities. A similar statement holds for the superposition of \(m\) independent point processes. In the present paper these ideas are extended to multivariate point patterns. Let \(X_{i}\) be a process of type \(i\) points and let \(X=X_{1}\cup\dots\cup X_{m}\) be the process of all points regardless of type. Three approaches are proposed which correspond to investigating three different forms of independence between types. First, they may compare the left and right hand side of (1) or its analogue for \(m\) types. Second, they may construct an ``inter-type'' \(J\)-function \(J_{ij}\) for each pair of types \(i\) and \(j\). Third, they may construct a function \(J_{i.}\) for each \(i\) summarizing the dependence of \(X\) on \(X_{i}\). The \(J\)-functions for a wide variety of stochastic models are calculated and applications to bivariate and trivariate point pattern data are contained.