Nova Acta Leopoldina Band 110 Nummer 377

Fig. 5 Sequentially randomized trial 7. Big Business Large scale data analysis has tremendous potential for monitoring complex systems and for predicting outcomes as structures evolve without major interventions. The knowledge obtained can help prevent problems and plan business logistics and marketing efficiently. State of the art methods allow to exploit past (longitudinal) information ever better for the prediction of what will happen under the current structure. An increasing number of predictors brings too, however, a higher chance of involving false positives. These are factors which appear associ- ated with outcome (after adjusting for other factors in a model for the dataset) but are not re- produced as added value predictors when new data are presented. Methods to correct for the inflation of p-values after multiple testing have been heavily studied and allow now for im- proved detection of prognostic values or a better selection of contributors to a predictive model (MOERKERKE et al. 2006). Even more valuable is the ability to detect causal effects which suggest interventions which may in turn be evaluated and lead to insight into optimal action. Dynamic treatments allow in particular to intervene without delay and thus prevent unwanted evolutions and optimize out- comes. Recognizing sources of weakness and strength in a system and knowing where to best intervene and how is a potential gold mine. It should then perhaps not come as a surprise that big companies get interested in statistics. In their 2009 press release on the occasion of the merger of IBM with SPSS a well established data analysis package, the following message was launched (http://www-03.ibm.com/ press/us/en/pressrelease/27936.wss): “Predictive analytics can help clients move beyond the ‘sense and respond’ mode, which can leave blind spots for strategic information in today’s fast paced environment – to ‘predict and act’for improved business outcomes.” With reference Causal Inference: Sense and Sensitivity versus Prior and Prejudice Nova Acta Leopoldina NF 110, Nr. 377, 47–64 (2011) 61