by SCIAG »
21 Oct 2016 12:59
21 Oct 2016 12:59
Throw Whoscored's player ratings in the bin but the statistics are reliable.
I'll also defend polling data. It needs to be treated properly, but if you're using a suitable model that accounts for house effects, historical trends, uncertainty, and voter distribution then you should be able to predict election results fairly easily. I think Ashcroft is the only one who attempts to do things at a constituency level though, and that makes General Election polling very difficult. 2015 was also a very weird election because it came after the first coalition government since WWII, so there weren't as many data to draw on when modelling.