To further examine membership of top layer clusters in 2012-2016 and 2017, we calculate weighted Jaccard coefficient. Jaccard coefficient refers to a measure of similarity of two sets and is calculated by dividing the size of the set intersection (i.e. number of unique skills that are present in both sets) by the size of their union (i.e. all unique skills in two sets). The coefficient varies between [0, 1], where Jaccard coefficient of 0 means that the two sets have no shared members and 1 means that there is a perfect match between the two sets. We use a weighted version of the Jaccard coefficient as proposed by \citet*{ioffe2010improved} and use the proportion of unique job adverts as weights. As shown in Figure \ref{121574}, the weighted Jaccard coefficient is high, which indicates a high level of stability of the methodology at the top layer.