On Measuring Confounding Bias in Mixed Multidimensional Data

Currently, we witness a significantly increased understanding, that disaggregation of data is the key to better-informed decision-making and enactment. The UN 2030 Agenda for Sustainable Development prominently stresses the importance of disaggregated data for achieving its goals. We argue that the key benefit of data disaggregation is the increased potential to understand confounding effects. Unfortunately, in today’s tool landscape, it lacks systematic support for understanding confounders. Therefore, in this paper, we contribute as follows. We integrate a means of confounder adjustment, i.e. multiple linear regression adjustment, as a new feature into the data analysis tool GrandReport, which works in mixed mode, i.e. for both numerical and categorical data in parallel. Next, we conduct experiments on an extended air pollutants dataset from several cities. We utilize the tool GrandReportl to investigate the correlation between standardized air quality indices and \({\text {CO}}_2\) levels before and after confounder adjustment in regard with additional city data. We argue that the experiments provide evidence for the usefulness and usability of the suggested approach.