Saturday, May 7, 2016

ReachTEL 50-50

We have a new ReachTEL poll. This poll has mixed it up methodologically. For the first time in the ReachTEL series, undecideds in this poll have been nudged to express a leaning-towards preference. And for the first time, preferences for the estimated two-party-preferred (TPP) vote share were allocated by respondents, rather than using the preference flows from the previous election.

I am not convinced about the value of using stated preferences in calculating a TPP estimate. Because many voters follow how-to-vote cards, using the preference flows from the previous election can be a better guide to actual preference flows at the next election, than asking respondents about their preference intention.

On the back of the envelop, if I had used my standard preference flows from the 2013 election, the TPP result is 51.8 to 48.2 in the Coalition's favour. My standard 2013 preference flows to the Coalition are 0.1697 from the Greens and 0.533 from others (from Antony Green).  These figures are not too different from my Bayesian model for ReachTEL which seeks to find the "most likely" preference flow estimate to explain each pollsters published TPP results given their published primary vote results.

This degree of change in ReachTEL's practice suggests I should treat this as the first poll in a new series for the purposes of data aggregation. For today, I have used the reported 50-50, without treating these numbers as a new Series. I will look at this again when the next ReachTEL poll comes out.

For the record. If I had subbed in my adjusted 51.8 to 48.2 in the Coalition's favour, the aggregation would have been.

Also, well worth reading: a piece by John Cassidy in the New Yorker, which looks at why so many data analysts got the US Republican nomination wrong.

No comments:

Post a Comment