As it is described in a previous post, when risk management is applied to the whole project plan in a quantitative manner, phase duration becomes a statistical distribution rather than a deterministic figure. It also allows having an overall impact estimation for individual risk or for the combined occurrence of more than one risk.
This approach makes the project plan more realistic but also more complex and difficult to report and understand. Which of the two approaches should we use? Well, again, it depends on the specific project.
Some of the results of such analysis might be: probability of finishing the project on time and / or within budget and contingency reserve estimated to ensure the required level of confidence
One of the most known tools for this process is the Monte Carlo simulation and decision tree analysis.
Notes taken during my study of "Practice standard for Project Risk Management" from PMI