I can't disagree with the examples provided, and therefore, with the probability assessments. However, I have encountered a distribution where there is one hit for every two misses (a total of 2 hits out of 4 misses), which allows us to estimate the probability, with some assumptions, as 1/2, but closer to 1/3. To avoid delving into esotericism, it would be best to look at the original source.
In general, you have to guess; as soon as there's a series of misses, you know to strike with a powerful unit next. And again, spells that deal direct damage never miss – that's another option.