Frequency of gravitational wave detection

As you may have guessed, this question is of great interest to the LIGO team. Simultaneously with the publication of the paper you mentioned announcing the discovery, the LIGO team submitted a number of companion papers with further details about the discovery, and predictions. One of these addresses your question:

The Rate of Binary Black Hole Mergers Inferred from Advanced LIGO Observations Surrounding GW150914

Their event rate estimation method considers both GW150914, and another significantly weaker (and less statistically significant) event. They consider a number of models for how the event rate might depend on system properties, and ask what the observations of GW150914 and the other candidate event imply for the overall rate. The results vary from model to model, but they chose models they felt could roughly bracket astrophysically plausible behavior. As summarized in their abstract:

Considering only GW150914, assuming that all BBHs in the universe have the same masses and spins as this event, imposing a false alarm threshold of 1 per 100 years, and assuming that the BBH merger rate is constant in the comoving frame, we infer a 90% credible range of $2-53 , mathrm{Gpc}^{-3} ,
mathrm{yr}^{-1}$ (comoving frame). Incorporating all triggers that pass the search threshold while accounting for the uncertainty in the astrophysical origin of each trigger, we estimate a higher rate, ranging from $6-400 ,
mathrm{Gpc}^{-3} , mathrm{yr}^{-1}$ depending on assumptions about the BBH mass distribution. All together, our various rate estimates fall in the conservative range $2-400 , mathrm{Gpc}^{-3} , mathrm{yr}^{-1}$.

Note that the paper is submitted, not published, i.e., still under peer review. Speaking as someone with expertise in such calculations, some aspects of the method look fishy to me, so I think it's worth checking back on the article in a few weeks for revisions. It doesn't take fancy methodology to see that the order of magnitude here (a few to ~100 per cubic gigaparsec per year) is in the right ballpark. But the paper presents a methodology that could make more detailed and more precise estimates and predictions as data accumulate, so it's important to make sure the methodology is sound.