Note on Selection Bias in Observational Estimates of Algorithmic Progress