Thus with large n, we can hedge against the risk of model misspecification, while maintaining high statistical power if the model were actually a good fit to the data. This theoretical insight underpinning our methodology can be traced back to the works of L. Le Cam and E. Lehmann, among others.