1: \begin{abstract}

2: I discuss an issue arising in analyzing data from astronomical

3: surveys:  accounting for measurement uncertainties in the properties of

4: individual sources detected in a survey when making inferences about

5: the entire population of sources. Source uncertainties require the

6: analyst to introduce unknown ``incidental'' parameters

7: for each source.  The number of parameters thus grows with the size of

8: the sample, and standard theorems guaranteeing asymptotic convergence of

9: maximum likelihood estimates fail in such settings.  

10: From the Bayesian point of view, the missing

11: ingredient in such analyses is accounting for the volume 

12: in the incidental parameter space via marginalization.  I

13: use simple simulations, motivated by modeling the

14: distribution of trans-Neptunian objects surveyed in the outer solar

15: system, to study the effects of source uncertainties on inferences.

16: The simulations show that current non-Bayesian methods for handling

17: source uncertainties (ignoring them, or using an ad hoc incidental

18: parameter integration) produce incorrect inferences, with errors that

19: grow more severe with increasing sample size.  In contrast, accounting

20: for source uncertainty via marginalization leads to sound

21: inferences for any sample size.

22: \end{abstract}

23: