MRP is simply a cleverer model for turning voting intention data into seats.
If you have incorrect polling data, it will cleverly turn that into incorrect seats with more precision.
I had a clever model in #GE17 and look how that turned out! /1
Had the polls been right, my seat forecast would have been near perfect as I demonstrated in my post mortem.
marriott-stats.com/nigels-blog/uk… /2
marriott-stats.com/nigels-blog/uk…
/3
After all, 5 out of the last 7 elections have seen at least one party experienced a polling error of 4 points or more.
marriott-stats.com/nigels-blog/uk…
/end