The war against bias: experimental design for big data
Henry Wynn (London School of Economics, UK)
The talk first reviews work (by others) on optimal experimental design for “big data”. This ranges from methods arising from the social and medical sciences, particularly in causal modelling, to recent work which tries to extract an optimum design from a loosely structured data set of covariates and also the literature on optimal design to guard against bias. The authors draw on some of this work but take a more game-theoretic approach. The idea is that the causal modelling operation, run by an notional “Alice”, needs a shield protecting against bias built by a notional “Bob”. The two operation can act harmoniously when the joint operation is a over product space but, even when not, a Nash equilibrium may be achievable, which balances the two objectives.
Joint work with Elena Pesce and Eva Riccomagno.