Supplemental Material for Causal Relationships in Longitudinal Observational Data: An Integrative Modeling Approach

An Integrative Modelling Approach".Here, we report the outline of the simulated scenarios used to test the causal discovery properties of the integrative approach alongside findings of conventional multiple regression.All codes, including the complete simulations scripts and the application to the Millenium Cohort Study, are available at osf.io/pmw6c/.The algorithm is specified in the function bellow and follow the steps:1 -train a model in part of the sample using the XGBoost algorithm 2 -use the model to predict the outcome for each individual in the remaining of the sample (a cross-validation procedure has to be added when analyzing empirical data) 3 -calculate the prediction errors for each individual prediction defined as the absolute difference between predicted and observed scores gcxboost<-function (df,predictors_cols,outcome_col){ d<-as.matrix(df)bst<-xgboost(data=d[1:n_train,predictors_cols],label =d[1:n_train,outcome_col],max.depth=6,eta=0.3,nrounds=100 ,objective="reg:squarederror") pred<-predict(bst,d[(n_train+1):(n_train+n_test),predictors_cols])