A couple of days ago I examined my treatment of the Morgan face-to-face poll. For the next few months at least, I plan to leave the residual face-to-face polls within the weekly Bayesian aggregation. However, I will exclude them from the sum-to-zero constraint in the Bayesian model. I think this gives a more accurate estimate of where the actual population voting intention sits.
For completeness, I have now extended the Bayesian aggregation model to cover the primary voting intention polls. I model the primary voting intention polls on much the same basis as I model the two-party preferred polls (see previous paragraph). Each model takes about 5 minutes of run-time on my computer.
The preliminary primary vote charts from the Bayesian model follow. Please note: I am still developing and testing this analytical suite and these results should be treated with a little caution.
The observant among you will note that these four charts do not add to 100 per cent at the end point. I suspect that is the result of rounding errors (and the fact that not all of the input data from the polling houses adds to 100 per cent either).
The other thing worth noting is that while the final TPP result (see below) for a number of polling houses is similar, they way in which the primary votes add to that total is more divergent.