1: \begin{abstract}

2: \vspace{-10pt}

3: We formalize the notion of nesting probabilistic programming queries

4:  and investigate the resulting statistical implications.  We demonstrate

5: that while query nesting allows the definition of models which could not 

6: otherwise be expressed, such as those involving agents reasoning about other agents,

7: existing systems take approaches which lead to

8: inconsistent estimates.

9: %changes are required to the approaches employed by existing systems to ensure consistent estimation.  

10: We show how to correct this by delineating possible ways one might want to nest 

11: queries and asserting the respective conditions required for convergence.

12: %To aid in the development of inference engines that satisfy these conditioning,

13: We further introduce a new \emph{online} nested Monte Carlo estimator that 

14: makes it substantially easier to ensure these conditions are met, thereby

15: providing a simple framework for designing statistically correct inference

16: engines.  We prove the correctness of this online estimator and show that, when using the

17: recommended setup, its asymptotic variance is always better than that of the

18: equivalent fixed estimator, while its bias is always within a factor of two.

19: %Our results both reaffirm

20: %the need for query nesting and highlight potential pitfalls

21: %associated with doing so.

22: \end{abstract}