Yesterday we used LOESS regressions to try and track out the size and nature of the Trunbull poll bounce – today we’ll use our Pollytrack methodology to do the same.
Another piece of polling data was released late yesterday, the Essential Report (that you might notice on the Skynooz Agenda program every Monday). This poll, like the latest Nielsen and Newspoll also showed a Turnbull bounce at the expense of the ALP. However, what makes EMC a little different to the other pollsters is their use of a fortnightly rolling average in their vote estimates that they release every week.
This week’s fortnightly average came in on the primary votes as a 44/35 lead to the ALP (The ALP down 1 and the Coalition up 2 since last weeks rolling fortnight average) for a TPP estimate of 57/43 to the ALP (with the ALP down 1/Coalition up 1 from their last poll). This used a sample of 1812 respondents.
To plug these numbers into our Pollytrack methodology to see if there really was a Turnbull bounce, firstly we need to look at last week’s Pollytrack data, the state of affairs before Turnbull took the leadership.
Yet to do this on an apples with apples comparison basis, if we are to calculate the current polling results as a pooled super sample, we have to remove the Morgan F2F poll from today’s measurement (as the current Morgan Face to Face poll in the Pollytrack window was taken before Turnbull became Opposition Leader). We also have to adjust the latest EMC poll to make an estimate of just it’s post Turnbull sample.
To adjust EMC, we can simply double the size of the movements it experienced over the last fortnight while simultaneously halving the sample size. This gives us an estimate for just the post-Turnbull component of a two party preferred of 56/44 washing out from primary votes running 43/37 to the ALP out of a sample of 900 respondents. Plugging this new data into our Pollytrack methodology where we take a polling average weighted by sample size we get the following result:
The ALP primary vote dropped 2.7 points to come in at 41.9. the Coalition primary is up 2.6 points to come in at 39.4 and the ALP TPP dropped 2.2 points to come in at a 54/46 split.
If we chart these changes with the MoE’s attached to each Pollytrack observation since August we get:
Marked on the charts in pink is the possible overlap between last week’s Pollytrack and today’s Pollytrack that could be due to pure sampling error (yes, you arent going mad – it really does look like the pink lines are at a slight angle but they’re actually straight. It’s just one of those weird visual tricks that fool the eye)
We have a problem in attempting to determine the exact probability of Turnbull having a statistically significant bounce in the polls that arises from the nature of our pooled sample. Ordinarily we would expect each poll that makes up our Pollytrack to have a mean of its headline value with standard deviation of X and where the probability distribution of the pure sampling error of each poll is a normal distribution (shaped like a bell curve). This means that if we were to guess a value of what the true measure of public opinion was, the further our chosen value was from a given poll result, the increasingly unlikely it would be to represent the true state of political opinion.
But because pollsters have their own quirks that derive from their quota’s, their population weighting and whatnot, we can’t be sure that the sampling error on our pooled sample has the shape of a normal distribution. So to be conservative here, we’ll make an assumption that the instead of being shaped like a bell curve, it’s actually a uniform shape – meaning that the true value of public opinion has a 95% certainty of being randomly distributed somewhere within the error margin bars on our charts.
If we were to do it this way, which in practice almost certainly reduces the probability of there being a statistically significant Turnbull bounce, we can easily calculate the joint probability of these polls having their real results existing in the polling overlap area (the pink area), ending up with an approximate 93.5% probability of Turnbull experiencing a bounce in public opinion of around 2%. If we were to treat these polls as having a normal probability distribution on their sampling error, the probability of a Turnbull bounce blows out to about 98%+.
So we can be pretty certain that Truffles received a bounce. Even though the polls individually had movements within their margin of error, collectively it is a bit of a different story with there being at least (”at least”, because we are doing this conservatively using a uniform distribution approach to sampling error) a 93.5% chance of the Turnbull movement being real, and being around the 2% mark in TPP terms.
9 Comments
Pardon my cynicism, but I bet if I polled a hundred people in the pub and asked, “Did you know the Liberal Party has a new leader? And if so, do you know his name?” I bet the numbers wouldn’t be very high.
Most people know and care very little about politics. As polling and political tragics we must not fall into the trap of pinning motivations to the public that aren’t really justified.
On these blogs we often criticise the press gallery for operating in a rarefied bubble and assuming the general public gives a stuff for their ruminations. We should be wary we in the blogosphere do not allow ourselves to become the same.
I mean, if the polls have shifted in the last week or so, can we be sure that’s because the Liberal Party has a new leader? Most people wouldn’t know or give a stuff. The shift might be due to any of an infinite number of reasons. The public are fickle when it comes to politics. We aren’t really to know why they change their (current) preferences at any time.
Cuppa, it is true that there are various segments of the public when it comes to viewing the political world. Some are as you say, “tragics”, others are moderately engaged, and some perhaps many, couldn’t give a rats about most of it until election time.
It is also true that bloggers criticise the media from time to time about their self importance without acknowledging our own failings and biases.
However, when it comes to the polling numbers, they are what they are. All you can do is take the results on face value, aggregate them, and track the trends over time.
Essential research is friendlier to the ALP than the others, so I would like to see some Galaxy numbers, which tends to be more favourable to the L/NP, for the tally to be more balanced. Still, it’s an indication, and that’s fair enough for now, and this tracking that Possum is doing is first rate.
In my view, the Turnbull change has unquestionably boosted the L/NP’s stocks, but whether that vote holds up over time, is another matter entirely.
Hi Aristotle. Please don’t get me wrong. I guess I’m saying to the LNP, enjoy your bounce, but don’t be too confident that it’s a Turnbull bounce. The “bouncing” of figures might be due to natural gravity (the ALP coming back toward Earth with a corresponding lift in LNP), or any of any number of variables.
Possibly it’s due to Turnbull, probably even, but not necessarily.
Point taken though, the polls are what they are and we must take them at face value, aggregate and follow trends.
No 4
Cuppa for your thesis to be consistent, you would have to argue that those same apathetic people in the pub would be unlikely to know the name of the Prime Minister.
Yes, your point is fair enough Cuppa – we can’t be sure of the reason for the change. But it is most likely to be as a result of Turnbull. A change in leader always changes the dynamics for better and worse.
Cuppa – every new mid-term opposition leader in the last 23 years has received an immediate poll bounce except Beazley Mk II. Some like Rudd and Howard it worked out for, some like Downer and Latham… well not so much.
It’s a pretty unlikely scenario that Turnbull didnt have something to do with these poll figures, and even though a large chunk of the population is turned off from politics, most still catch the news be it in the papers, on the TV or on the radio.
At the end of the day, the data is the day and it suggests a poll pounce with an extremely high probability. All we can do is really go with the data, because the alternative is pretty much akin to “making stuff up”
Ari – I’d love to see some Galaxy polls with voting intention too. More polls from all I reckon!
So you’re implying that the name of the latest leader of the Liberal Party (the third leader in less than a year of a party not in government) would be as popularly-known as the name of the Prime Minister?
What the …! Now, I know you’re a Liberal, GP, and you’d like to see your stocks raised, but this is pretty desperate wishful thinking. Sounds like chutzpah; do you really believe what you wrote?
Fair enough, Possum at number 7. There’s another thing I was thinking. Assuming there will be some in the community who haven’t yet caught up with the news of the Liberal leadership change, I would not be surprised if there was another small bounce or two to come as these stragglers (for want of a better word) assimilate the news. Not much of a bounce, mind, because those would definitely be of the non-engaged type.
Do past polling patterns bear out this theory: the initial bounce, followed by the smaller bounce from the stragglers?