[prof_hobart]

If you were using the primaries as a way of predicting the overall winner, the critical word would be "random".

The statistical idea that a small sample can accurately represent the whole is based on the assumption that the small number are largely indistinguishable from that whole. It's like taking a sample of

In this case, the people voting in primaries are a self-selecting group - people who are both relatively actively engaged with politics and members of one of the two main parties.

It's not unreasonable to think that the wider population (even the wider population of people who normally vote at an election) could behave vastly differently.

For instance, if you had a candidate who was so divisive that they were either loved or hated, then it wouldn't be overly surprising if the vast majority of their supporters had signed up to vote in the primaries, with little support outside of that core. Compare this to a more middle of the road candidate - they might struggle to get as many people out to vote in a primary, but might do far better in the wider population.

That theory is very relevant in the UK at the moment, with the leader of the Labour Party (Jeremy Corbyn) being very popular amongst grassroots party members, but with his enemies saying he could never win a general election as this level of support would never be reflected outside of party supporters.