I've used the household transportation survey and cross-check against other sources to estimate bicycle trips and mileage, then overall injury and fatality stats to piece together a very approximate probability of accident per mile and per trip. There are tabulations of accidents summarized by most of the factors that typically describe a commute, and by general categories of rider (age, sex, etc) by which you may arrive at a little better approximation.

You'll need to be willing to do a bit of research, and then to apply sound statistical reasoning. And be forewarned that much of the data is incomplete or dubious - one needs to evaluate and assign confidence intervals to the any calculations.

Or, alternatively, the use of good judgement based on knowledge of the conditions and danger. But I prefer math.