Mind the ’credibility’ gapColby Cosh finds out what subsets, modelling assumptions and ’non-probability samples’ have to do with polling these days
