I tinkered with the numbers a little to see how different the results would be.
In the main post of the results of the Wentworth By-election first survey I used the standard weighting I use all the time: age, gender, and reported 2016 vote. The sample of 2016 Liberal Party voters was considerably less than it should have been – 37.5% of the sample rather than 60 odd, which required weighting close to the maximum limit of what is permissible to get it to balanced.
However, that still left the Liberal vote considerably lower than it’s likely to be. So, while I do not like doing this as it is a statistical no-no, I did a quick crunch, boosting the Liberal weighting by 30% to get Sharma’s first preference vote share on par with a recent Reachtel poll of the Liberal vote in the mid 30s. Note: I only boosted Sharma’s voters, not all 2016 Liberal voters, as that would have just boosted everyone and made little difference.
This is roughly on par with the recent Reachtel poll. This makes is a closer race, but still a Phelps win… unless Phelps comes third, in which case it’s 50/50.Just for fun, I also did a simulation doubling the Heath vote, getting it closer to where I think it’s likely to end up. I did this fairly crudely by assigning all the undecideds to Heath, and doing the same 30% boost to the weighting as I had done for Sharma. Less close.
What we can guestimate from here is that Sharma cannot afford to get anything less than 35% primary, and needs Phelps’ support to fall, thus ensuring it’s a Labor v Liberal 2PP fight… if he is to have any chance at all.
Inferred two candidate/party preferred methodology explainer
It is very difficult to ask who people are going to preference at the best of times, as most people do not understand preferencing. In a three horse race where one of those three had not formally declared at the time we put the poll in the field, it’s impossible. So what I do is utilise all the data points I have (80 from the first survey in the Wentworth By -election study) to infer how they are likely to preference.
Sometimes it’s pretty easy, for example a Science party voter commented, in response to three different questions:
- Not inclined to give my preference to Liberals.
- Like Kerryn Phelps a lot.
- But I will give my second preference to Kerryn Phelps.
From this it is pretty easy to infer a preference flow to Phelps, and a preference to Tim Murray over Dave Sharma.
It’s not usually that clear. The Liberal leaning Green voter is usually the tough one to pick, but Wentworth has a few: like this senior citizen, who is very consistent in his commitment to the Greens, doesn’t mention Phelps at all, doesn’t take any of the open invitations to talk about the Libs, until this, towards the end:
- … most likely a return a tight contest will finish with a liberal victory.
Now, whether he was wishing, or predicting, is debatable. I’ll grant you that. There are a number of other indicators (where he lives, higher awareness of Sharma than Phelps or Murray, friends are voting Liberal, the repeated references to “voting on values” which is phrasing favoured in conservative politics, and a veiled reference to hoping the leadership spill will reflect badly on both major parties) that I took into account before putting him in the Sharma preference box.
We will ask in Survey 2 for a preference indication, but so far this inferred method – while labourious – has served well and probably slightly more accurate or predictive than the reported preferences.