A 3% Labour lead in London would indeed be a considerable surprise. Perhaps as the focus gradually shifts away from Brexit, it will lead to the Conservatives doing better than expected in Remainy/traditional Tory areas - I would like to think it was Brexit rather than pro-Corbyn enthusiasm that had the greater impact on voting behaviour last year.
The poll using a different sampling method, supposedly to better reflect voters that are harder to reach and less politically engaged.
Some the regional samples look odd compared to other polls. For example, the Scottish sample has Labour at 16%, and the SNP on 46%.
subsamples have never been credible.
Telephone polls have always been credited with reaching the people who arent engaged. But they were the worst in tge referendum and only survation use them now
Sadly, even Survation seem to have given up on the telephone. They used to do both telephone and online polls but I think all their polls since July 2017 have been online. Mid June - mid July 2017 they produced two telephone and two online polls. The mean voting intentions were:
The only genuine use for subsamples is to determine whether an unusual poll is more likely to be a rogue/outlier or the start of a new trend. And even then it's a far, far less reliable method than simply waiting for a couple more polls to confirm or deny the change.