Validity of Capability Estimates Obtained through Structural Equation Modelling in a Multidimensional Setting – An Agent-Based Simulation Exercise

The main aim of this study is to evaluate the performance of structural equation models in the estimation of capabilities in a multidimensional development space. We simulate an artificial society using agent-based modelling techniques. Our artificial world is generated to closely resemble a real-life scenario in Mexico. We use our model to generate data relating to three key dimensions of human development: education, health and social networks. Our model produces information on both generally observable variables such as family background and functionings and generally unobservable values such as capabilities. We then use these data to estimate capabilities using a structural equations model and contrast them with the true capabilities. Our results suggest that the SEM approach provides very good estimates of capabilities. We compare two different specifications of the links among the dimensions and find that the one that accounts for their interdependence produces better results. We conduct a sensitivity analysis which shows that results are relatively stable across different sets of parameter values