The dynamics of optimization: mixing random and risk-based inspections

Msc Graduation project with a TPM research team

You join a research team with one other student and 6 TPM-professors. The team will do mixed-methods research on the optimum between risk-based inspections and random inspection by the Netherlands Food and Consumer Product Safety Authority (NVWA). NVWA asked us to develop a strategy for effective law enforcement. Two elements are part of this strategy. A first element is risk-based enforcement: using analytics to target the companies that put our society to risk by violating laws. A second element simply entails random inspection visits to companies to find out whether they violate laws.  Random surveillance improves the efficiency of risk-based surveillance (exploration-exploitation tradeoff), may increase detection of blind spots and emerging risks, and in particular, is (usually) necessary to obtain unbiased data. Such unbiased data is indispensable for compliance monitoring, for building and updating (machine learning) risk models, and for assessing the prevalence of risks.

So risk-based and randomized inspections have both their advantages and disadvantages. How to find an optimum? And what analytical methods are suitable to find such an optimum? To answer these questions the research will conduct:

• Literature research and empirical research to bring the answer to the question further than “it depends”. It depends on the sector to which the laws and inspection apply (i.e. technological system, size, complexity of laws, etc.). It also depends on the dynamics of a sector. In other words: we want to understand the sector as an emergent system: we want to learn about the main causalities of factors and how they change. The research will result in a framework that contains the main criteria for methods tackling the trade off between risk-based and random approache
• Literature research and expert interviews that result in a categorization and assessment of those methods.  Methods may include a) analytic methods, b) methods based on stochastic (e.g. Monte Carlo) simulation, c) agent-based modeling approaches, d) gaming, etcetera. The assessment will be informed by the criteria as found in the first part of the research.

Feels good? 
Contact Haiko van der Voort (h.g.vandervoort@tudelft.nl