Outlier model

[drphilmarshall]

The main reason for enabling combination of lenses' time delays to infer H0 is to enable a principled outlier model to be made: lenses that have measured time delays that predict H0 to be very different from the ensemble should be inferred to be suffering from some unknown systematic error, and discounted. This can be done probabilistically (see Hogg et al 2010). Urgency is low - this is something that could be addressed in a pull request from a Good Team, even.

[drphilmarshall]

This could also be a good hack project at Astro Hack Week 2016 - I'll pitch it and report back.

[drphilmarshall]

I realized while writing the hack pitch that the way we have the problem set up, in terms of a set of time delay posterior sample ensembles, may make it difficult to analytically marginalize out the outlier probabilities - so to solve this problem we may have to recast it in terms of multivariate Gaussian approximations to the time delay PDFs.