Came here to say this, and in general any dataset that has a chance of having incomplete data is going to tends towards this same result.

Maybe the problem is less pronounced for "successful" companies, but given my past experience looking at the data, I wouldn't take this at face value.