QuoteOriginally posted by: PaulI've been trying to model the behaviour of the polls. It's tricky because there's not much of a pattern. I started with the obvious ODE/chemical reaction thing, people going from one state to another in Remain/Leave/Don't Know. In a discrete-time setting that's just transition matrices again. Unlike the General Election it's easier because there isn't the nonlinearity caused by votes being lumped by seats, it's a straight winner takes all. But that makes it harder because then I have no edge over anyone else trying to model this!I'm exploring the idea of there being Pro EU News and Anti EU News, and those explaining changes in the polling. PAre you trying to predict the polls or the voting outcome?Is there really such as thing as "Pro EU News"? Isn't it more accurate to say that a given bit of news is pro-EU for some and anti-EU for others. For example, if a bit of news says Brexit will hurt London cabbies by preventing anti-Uber regulations (I'm making that up!), that's pro-EU for taxi drivers, anti-EU for riders who want to see more competition in taxi-like services, and neutral for people that have no opinion on taxis vs. Uber. There may be some news that is nearly universally on one side (e.g., Brexit will make British men go bald) although even that might engender anti-EU support among chrome-dome-loving women.The other issue is what if pro-EU news has a greater effect on motivating anti-EU voters to turn out to vote than it has the effect of converting undecided voters into the pro-EU camp and getting them to vote. There's modeling the fear versus persuasion effects on all the conditional probabilities of a person's opinion, their willingness to voice that opinion in a poll, and their willingness to take action on voting day.