1: \begin{abstract} We consider a novel paradigm for Bayesian testing of hypotheses and Bayesian model comparison. Our

2: alternative to the traditional construction of Bayes factors or posterior probabilities of a model given the data is to

3: consider the hypotheses or models under comparison as components of a mixture model. We therefore replace the original

4: testing problem with an estimation one that focuses on the probability or weight of a given hypothesis within the

5: mixture model. We analyse the sensitivity of the posterior distribution of the weights induced by different priors on

6: these quantities. We stress that a major appeal in using this novel perspective is that generic improper priors are

7: acceptable.  Among other features, this allows for a resolution of the Lindley--Jeffreys paradox.  For example, a

8: reference Beta $\mathcal{B}(a_0,a_0)$ prior on the mixture weights can be used for the common problem of testing two

9: contrasting hypotheses or models.  In this case the sensitivity of the posterior estimates of the weights to the choice

10: of $a_0$ vanishes as the sample size increases. We therefore advocate a default choice of $a_0=0.5$, derived from

11: \cite{rousseau:mengersen:2011}.  Similar results apply for testing more than two alternatives.  Another feature of this

12: easily implemented alternative to the classical Bayesian solution is that the speeds of convergence of the posterior

13: mean of the weight and of the corresponding posterior probability are quite similar.  \end{abstract}

14: