[ramblenode]

Since so many people here seem to be in the business of peddling cherries, apples, and oranges, I thought I'd throw my own analysis in the mix:

First, the likelihood of earning a PhD shouldn't be viewed as p(graduate) but as p(graduate | admitted) * p(admitted). Once you factor in the high rejection rate of competitive PhD programs, the success rate drops off pretty sharply. Additionally, the applicant pool tends to self-select toward people who at least believe they are minimally qualified because of the time and expense in completing applications and gathering letters of recommendation.

Second, the random error term is much larger in the hypothetical formula for startup success than it is for PhD success--in fact, it's probably much larger than any variable one can control. A consequence of this is that a unit increase of talent/skill/drive will move the needle further toward success in the PhD world than in the startup world. Comparing successful or unsuccessful individuals across worlds tells you very little.

Third, for all the parroting of the "9 in 10 startups fail" statistic, there seems to be almost no work in connecting its relevance. A startup is not a person. A person may found multiple companies in their lifetime. A person only needs to earn a PhD once to be considered a PhD. I could go on, but I think "apples and oranges" is sufficient.