Confucious,
the point about grammar counties / authorities vs those without grammars is that you have to look at the outcomes for ALL children.
Yes, if you take in the top 5% (or less) as superselective grammars do, then you will get good results - so the individual grammar schools get very good results.
a) that doesn't make them good schools - they often have relatively poor value added: they take in selected, bright children, who, unsurprisingly, do well at GCSE and A-level, but the 'progress' is only what you would expect.
b) if you look at the whole county / area and compare all results of all children within that area, essentially the OVERALL results are the same for grammar and non-grammar counties with similar demographics. If the grammar system was genuinely better, you would expect the supposed 'improvement' for the bright children to mean that overall results were better, but this is not the case.
I seem to remember that the precise results depend on particular subgroups - I seem to remember very able children do very slightly better in grammar counties, but this is offset by the fact that middle ability children do worse, but i would have to re-read the study.