1: \begin{abstract}

2: In this article we study a class of parameters with the so-called `mixed

3: bias property'. For parameters with this property, the bias of the

4: semiparametric efficient one step estimator is equal to the mean of the

5: product of the estimation errors of two nuisance functions. In

6: non-parametric models, parameters with the mixed bias property admit

7: so-called rate doubly robust estimators, i.e. estimators that are consistent

8: and asymptotically normal when one succeeds in estimating both nuisance

9: functions at sufficiently fast rates, with the possibility of trading off

10: slower rates of convergence for the estimator of one of the nuisance

11: functions with faster rates for the estimator of the other nuisance. We show

12: that the class of parameters with the mixed bias property strictly includes

13: two recently studied classes of parameters which, in turn, include many

14: parameters of interest in causal inference. We characterize the form of

15: parameters with the mixed bias property and of their influence functions.

16: Furthermore, we derive two functional moment equations, each being solved at

17: one of the two nuisance functions, as well as, two functional loss

18: functions, each being minimized at one of the two nuisance functions. These

19: loss functions can be used to derive loss based penalized estimators of the

20: nuisance functions.

21: \end{abstract}

22: