To see why there is a bias, consider a hypothetical case where firms above a certain (unobserved) quality threshold received VC funding.10 The goal of the empirical analysis then would be to identify that threshold. Conditional on sample size, if the probability of observing firms on each side of the threshold were equal, and if the distance from this threshold were randomly distributed, then the data would provide a significant amount of information to estimate the threshold. Now imagine that the dataset was reduced in size by randomly eliminating observations.