ENBIS-15 in Prague6 – 10 September 2015; Prague, Czech Republic Abstract submission: 1 February – 3 July 2015
Best Manager Award (Antje Christensen), Young Statistician Award (Marit Schoonhoven) and Greenfield Challenge8 September 2015, 14:40 – 15:25
Antje Christensen: The Role of Statistics in Decision Making
A crucial overlap between statistics and management is the art and science of decision making. The theory of decision analysis operates with the need for meaningful, reliable information, and for logically correct reasoning. Both can be effectively supported by statistical disciplines: Meaningful information can come from analysis of historical data, design of experiments, and forecasting. Furthermore, statistics can deliver measures on how reliable the information is. While the correct reasoning is mainly based on logic, probabilistic calculations can play a role. In this talk, I will briefly touch on the basics of decision analysis and then show several cases of decisions at very different organisational levels in the parmaceutical industry.
Marit Schoonhoven: Estimation Methods for Statistical Process Control
The classical literature limits performance studies to control charts with known parameters even though, in practice, the distributional parameters are estimated. Our work studies the effect of estimation on control chart properties and tries to improve existing methods that rely on estimated parameters.
In Schoonhoven et al. (2011a), Schoonhoven et al. (2011b), Schoonhoven and Does (2012), Schoonhoven and Does (2013), Zwetsloot et al. (2014a), Zwetsloot et al. (2014b) and Zwetsloot et al. (2014c) different types of control charts are considered, all based on estimated parameters. The effect of estimating parameters is studied for uncontaminated estimation data as well as for data that are contaminated with outliers or step changes in the distributional parameter. The articles propose new methods for parameter estimation. The idea is as follows: use a robust estimator to obtain an initial estimate, screen the data for anomalies and use an efficient estimator to arrive at a final estimate.
To fine-tune the methods even further, we derive correction factors for control chart limits. Correction factors are necessary in order to obtain the desired in-control performance that a practitioner wants to achieve. In the existing literature, there are no correction factors that can be used to obtain the desired ARL in expectation (that is, the average of the ARLs of all possible charts resulting from estimation). We also derive correction factors in order to guarantee that almost all possible estimated control limits result in an in-control ARL of at least a certain value.