The paper addresses the use of several performance indexes in the analysis of data acquired in contexts such as social robotics (SR) experiments. The main claim is that the analysis using several correlation/distance indexes can be misleading and data sparsity can reveal the presence of abnormal data (and hence have the potential to influence and conclusions). In general, SR experiments embed multiple sensors and tend to produce huge amounts of data both during short and long time intervals. Furthermore, the social nature often means that observations must be spread in time, making them prone to errors, e.g., perception errors caused by memory fails. In particular, the paper reviews the use of the Gini sparsity index to detect bias, e.g., potentially induced by human factors and causing perception problems. By analyzing sparsity variations one can obtain, for example, models for the degradation of human perception over time, which can be relevant information for SR programming. These ideas are illustrated with simulation experiments, with synthetic data, which (i) allow a simple construction of the experiment and, (ii) to avoid any potential bias that real experiments often, inadvertently, induce in the data. Moreover, despite the social robotics flavour the ideas are applicable to other domains.