A long time ago I learned that big projects are more likely to fail. One empirical study defined failure as "no line of code used in production". All the studies I know are decades old. In the hope we've learned something since then: What is the relationship between project size and failure now? (Say, the past decade?)
To answer my question:
P. 4 of the Standish report's CHAOS Report 2014 is a good answer. Alas the chance of failure seems broadly unchanged.