Dean Knight has a post on how rounding party vote figures to 0 or 1 decimal point can lead to the wrong allocation of seats. He queried the Colmar Brunton seat projections as he (and I) got different putting their results into the Electoral Commission calculator.
They use the raw numbers (with no rounding) in their calculations, so their projections are accurate.
It poses a small issue for me. I always prefer to calculate the seat projections myself using the official calculator, rather than trust what the media report. However as I only get the rounded poll results, then the projections may differ by a seat for some parties.
In future I will use the Colmar Brunton projections, as they have shown to Dean they definitely do it the right way, but will still calculate my own for other polls.
One reason I do my own calculations is I don’t assume the electorate seats will be the same as in 2005, but instead go off any public electorate polls to feed into assumptions on electorate seats won. Hence am projectiing Maori Party wins six seats, not four seats as Marae/Digipoll has shown them ahead in six.