You're being very frequentist (http://www.statisticalengineering.com/frequentists_and_bayesians.htm http://www.statisticalengineering.com/frequentists_and_bayes

You're being very frequentist (http://www.statisticalengineering.com/frequentists_and_bayesians.htm http://www.statisticalengineering.com/frequentists_and_bayes...). You don't have to have a random sample to estimate a probability. In the Bayesian interpretation, you can have a reasonable prior that you update as you collect more data. For example, a meteorologist can look at the clouds and "guesstimate" that there's a 50% chance of rain tomorrow. A frequentist would object and say that you can't take lots of random samples of tomorrow to estimate that probability. The Bayesian is willing to take the meteorologist's guess as a prior, and update it as & when more data becomes available. In this case, I think Sachin's statement is reasonable in the absence of any real data.