Weighted networks: the issue of dichotomization

ABSTRACT The purpose of this methodological paper is to explore one issue in the analysis of complex networks when weights are used to value edges. Often these weights are removed by setting a threshold and considering the links existing only if their value is higher than this. This practice allows then using simpler metrics as provided by many software packages. However, many common network properties, which often lead to specific interpretations, can be highly sensitive to the assumptions and thresholds used. By discussing a case, we show the differences that arise when dichotomizing a weighted network. We conclude that while unweighted networks can provide insights into some structural properties, the operation can be unnecessary and even detrimental for studying many features and processes when valued relational data are available.