Detlef Groth, Christiane Scheffler, Michael Hermanussen

• Background: Multiple linear correlations between parameters can be shown in correlation matrices. Correlations can be ranked, but can also be visualized in network graphs. Yet, translating a correlation matrix into a network graph is not trivial. In view of a popular child game, we propose to name this method St. Nicolas House Analysis. Material and methods: We present a new method (St. Nicolas House Analysis) that helps translating correlation matrices into network graphs. The performance of this and other network reconstruction methods was tested in randomly created virtual scale-free networks, networks consisting of bands or hubs, using balanced classification rate and the F1-Score for correctly predicting existing and non-existing edges. Thereafter we analyzed anthropometric data and information on parental education, obtained from an anthropometric survey in 908 Indonesian boys and 808 Indonesian girls. Seven parameters were analyzed: child height standard deviation score (hSDS), child BMI standard deviation scores (BMI_SDS),Background: Multiple linear correlations between parameters can be shown in correlation matrices. Correlations can be ranked, but can also be visualized in network graphs. Yet, translating a correlation matrix into a network graph is not trivial. In view of a popular child game, we propose to name this method St. Nicolas House Analysis. Material and methods: We present a new method (St. Nicolas House Analysis) that helps translating correlation matrices into network graphs. The performance of this and other network reconstruction methods was tested in randomly created virtual scale-free networks, networks consisting of bands or hubs, using balanced classification rate and the F1-Score for correctly predicting existing and non-existing edges. Thereafter we analyzed anthropometric data and information on parental education, obtained from an anthropometric survey in 908 Indonesian boys and 808 Indonesian girls. Seven parameters were analyzed: child height standard deviation score (hSDS), child BMI standard deviation scores (BMI_SDS), mid-upper-arm circumference (MUAC), mean thickness of subscapular and triceps skinfold (mean SF), and elbow breadth; as well as maternal and paternal education (years of schooling). The parameters were considered as the nodes of the network; the edges represent the correlations between the nodes. Results: Performance measures, balanced classification rate and the F1-score, showed that St. Nicolas’ House Analysis was superior to methods using sophisticated correlation value thresholds and methods based on partial correlations for analyzing bands and hubs. We applied this method also in an Indonesia data set. Ranking correlations showed the direct association between parental education and child growth. Conclusion: St. Nicolas House Analysis confirmed that growth of Indonesian school children directly depends on maternal education, with no evidence that this effect is mediated by the state of nutrition.…