Long tail statistics
IT projects, or rather, high-software-content development and projects are plagued by missed (effort) estimates. And that's OK as the potential uptake due to scalability and transformative efficiency gains can be enormous. Now, just how much are typical overruns?
There is highly credible peer reviewed academic research on the "overrun" of software (development) projects (see Flyvbjerg et al.). Unsurprisingly, this research shows the overrun of software projects is significant; it is the most significant of all project types, it beats the Summer Olympics, nuclear power plant construction, tunnels, etc., by a margin.
In fact the distribution telling you (roughly) the probability of finding a software project with overrun x follows a power law distribution (a long tail distribution). Now, this power law (its scaling exponent) is such that if you would calculate the average overrun of projects you would find that the average is infinite.
In other words, philosophically, if you wait an infinite amount of time and the universe would be filled with software projects then you would find that the most likely (central moment) overrun would be infinite!
Now, software devs know this and that's why they are reluctant to give you estimates as the real lead-time varies so much that it doesn't even allow you to calculate a mean.