The affinary coefficient
For measuring the similarity between histograms and between modal symbolic variables an affinary measure related to the Hellinger distance may be used. This approach is described in this paper. Consider a symbolic data matrix \(\underline{X}=(\xi_{kj})_{N\times p}\) for \(N\) data units \(k\) and \(p\) symbolic variables \(Y_j\). Assume that \(Y_j\) is a categorical, modal variable with \(m_j\) modalities \(\{1,\dots,m_j\}\) and that the entry \(\xi_{kj}=(n_{kj1},\dots,n_{kjm_j})\) of \(\underline{X}\) returns the numbers \(n_{kj\nu}\) of individuals which share the category \(\nu\) for the variable \(Y_j\). Each row \(x_k=(\xi_{k1},\dots,\xi_{kp})\) describes an individual modal object with \(p\) variables. With this notation, for two given units \(k,k'\in E\), the affinary coefficient related to the frequency distributions \(\xi_{kj},\xi_{k'j}\) is given by \[ \text{aff} (\xi_{kj},\xi_{k'j}) \colon= \sum_{l=1}^{m_j}\sqrt{(n_{kjl}/n_{kj\cdot})\cdot(n_{k'jl}/n_{k'j\cdot})}, \] where \(n_{kj\cdot}\colon=\sum_{l=1}^{m_j}n_{kjl}\) is the number of individuals belonging to the unit \(k\) for which the variable \(Y_j\) is observed. Note that \(0\leq\text{aff} (\xi_{kj},\xi_{k'j})\leq 1\) with value 1 if \(\xi_{kj},\xi_{k'j}\) are proportional, and value 0 if they are orthogonal. For the entire collection see [Zbl 1039.62501].
This page was built for publication: The affinary coefficient
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2756637)
