This article is from the source 'guardian' and was first published or seen on . It last changed over 40 days ago and won't be checked again for changes.

You can find the current article at its original source at http://www.theguardian.com/world/the-swing/2013/aug/28/poll-averaging-model-shows-coalition-in-lead

The article has changed 2 times. There is an RSS feed of changes available.

Version 0 Version 1
Election 2013: Poll averaging model shows Coalition firmly in the lead Election 2013: Poll averaging model shows Coalition firmly in the lead
(21 days later)
The latest update to my poll averaging model puts the Coalition on 53.4% to Labor's 46.6%, in two-party preferred (TPP) terms. Labor's TPP level is unchanged over the past week.The latest update to my poll averaging model puts the Coalition on 53.4% to Labor's 46.6%, in two-party preferred (TPP) terms. Labor's TPP level is unchanged over the past week.
For Labor supporters, that's the extent of any good news here: the decline in Labor TPP vote share appears to have stopped.For Labor supporters, that's the extent of any good news here: the decline in Labor TPP vote share appears to have stopped.
Rudd has been prime minister (again) for about 60 days. Polls taken immediately after his return to the leadership suggested that Labor's standing in the polls jumped a little over six points, with another half-point improvement over the next 10 days.Rudd has been prime minister (again) for about 60 days. Polls taken immediately after his return to the leadership suggested that Labor's standing in the polls jumped a little over six points, with another half-point improvement over the next 10 days.
A slow decline followed, which accelerated in late July. Between the last few days of July and mid-August, Labor appears to have lost at least half a point of TPP vote share per week.A slow decline followed, which accelerated in late July. Between the last few days of July and mid-August, Labor appears to have lost at least half a point of TPP vote share per week.
This period accounts for most of the loss we've seen in Labor's vote share since the post-Gillard peak in early July; Labor is now polling 2.5 points below where it was on the weekend of 6-7 July.This period accounts for most of the loss we've seen in Labor's vote share since the post-Gillard peak in early July; Labor is now polling 2.5 points below where it was on the weekend of 6-7 July.
About the best that might be said for Labor's standing in the polls is that the rate of Labor's decline has appeared to slow, and has perhaps even halted.About the best that might be said for Labor's standing in the polls is that the rate of Labor's decline has appeared to slow, and has perhaps even halted.
The data points to a possible levelling of the parties' TPP trajectories, perhaps stabilising at current levels, as whatever remaining undecided voters settle on a particular vote choice.The data points to a possible levelling of the parties' TPP trajectories, perhaps stabilising at current levels, as whatever remaining undecided voters settle on a particular vote choice.
Alert readers have asked one or two questions about my model. For instance, if Newspoll's results this week suggested an improvement for Labor, then why hasn't the model estimate also tracked back in Labor's direction?Alert readers have asked one or two questions about my model. For instance, if Newspoll's results this week suggested an improvement for Labor, then why hasn't the model estimate also tracked back in Labor's direction?
Newspoll is just one of several polls in the field this past weekend. The balance of the polling — taking into account the sample sizes and track records of each pollsters — suggested a very slight decline in Labor's TPP position, and so a slight boost for the Coalition.Newspoll is just one of several polls in the field this past weekend. The balance of the polling — taking into account the sample sizes and track records of each pollsters — suggested a very slight decline in Labor's TPP position, and so a slight boost for the Coalition.
I've also been asked why my poll average is on the more bullish side for the Coalition, the more pessimistic side for Labor. I have two responses.I've also been asked why my poll average is on the more bullish side for the Coalition, the more pessimistic side for Labor. I have two responses.
First, my model reconstructs the entire trajectory of ALP TPP voting intention from six months prior to the 2010 election right up through the most recent polls. I do this to exploit the fact that on election day 2010 we knew that Labor was on 50.12% TPP (the actual 2010 election result). This provides a fixed reference point for the model. I estimate the trajectory of ALP TPP vote shares subject to the constraint that it run through that point on that day; this constraint also lets me estimate "house effects" or "bias" terms for each pollster. Without a fixed constraint we have no external reality to anchor the model to, eg is Labor doing great and the polls are all biased in favour of the Coalition, or the opposite, or is the truth somewhere in between? A known reference point resolves this indeterminacy.First, my model reconstructs the entire trajectory of ALP TPP voting intention from six months prior to the 2010 election right up through the most recent polls. I do this to exploit the fact that on election day 2010 we knew that Labor was on 50.12% TPP (the actual 2010 election result). This provides a fixed reference point for the model. I estimate the trajectory of ALP TPP vote shares subject to the constraint that it run through that point on that day; this constraint also lets me estimate "house effects" or "bias" terms for each pollster. Without a fixed constraint we have no external reality to anchor the model to, eg is Labor doing great and the polls are all biased in favour of the Coalition, or the opposite, or is the truth somewhere in between? A known reference point resolves this indeterminacy.
Second, the pollsters collectively underestimated Labor TPP in 2010 (and in previous elections too, by the way). As I've detailed in a previous post,
the Australian polling industry does not appear to collectively underestimate Labor TPP, at least not based on the evidence from recent elections.		Second, the pollsters collectively underestimated Labor TPP in 2010 (and in previous elections too, by the way). As I've detailed in a previous post,
the Australian polling industry does not appear to collectively underestimate Labor TPP, at least not based on the evidence from recent elections.
Accordingly, my model presents an "average" that appears to tilt in the Coalition's direction, but this is because every pollster tilted towards Labor in 2010.Accordingly, my model presents an "average" that appears to tilt in the Coalition's direction, but this is because every pollster tilted towards Labor in 2010.
Could my bias corrections be wrong? Perhaps the pollsters' performances in 2010 were "one offs" and we shouldn't unduly rely on the 2010 experience when handicapping the pollsters in 2013.Could my bias corrections be wrong? Perhaps the pollsters' performances in 2010 were "one offs" and we shouldn't unduly rely on the 2010 experience when handicapping the pollsters in 2013.
Fair enough. And so I fit some alternative models to the data as well that I report below. I also stress that my model simply doesn't equate "house bias" with performance in 2010. In my model, bias parameters are "free" parameters; the only hardwired constraint in my preferred model is that the recovered ALP TPP trajectory runs through the 2010 election result.Fair enough. And so I fit some alternative models to the data as well that I report below. I also stress that my model simply doesn't equate "house bias" with performance in 2010. In my model, bias parameters are "free" parameters; the only hardwired constraint in my preferred model is that the recovered ALP TPP trajectory runs through the 2010 election result.
An alternative model is to set the bias parameters to the difference between the pollster's final pre-election poll in 2010 and the actual result. This is equivalent to "norming" the pollsters relative to Newspoll, who almost hit Labor's 50.12% TPP result in 2010 exactly, with a 50.2% TPP estimate.An alternative model is to set the bias parameters to the difference between the pollster's final pre-election poll in 2010 and the actual result. This is equivalent to "norming" the pollsters relative to Newspoll, who almost hit Labor's 50.12% TPP result in 2010 exactly, with a 50.2% TPP estimate.
Yet another alternative is to assume the polling industry's bias away. In this model, the anchoring constraint is that the estimated house effects sum to zero, ie there is variation across pollsters, but on average, the industry "gets it right".Yet another alternative is to assume the polling industry's bias away. In this model, the anchoring constraint is that the estimated house effects sum to zero, ie there is variation across pollsters, but on average, the industry "gets it right".
Results from these three models appear in the table below:Results from these three models appear in the table below:
Three different models, the same result: a convincing win for the Coalition.Three different models, the same result: a convincing win for the Coalition.
Our editors' picks for the day's top news and commentary delivered to your inbox each morning.Our editors' picks for the day's top news and commentary delivered to your inbox each morning.
Our editors' picks for the day's top news and commentary delivered to your inbox each morning. Enter your email address to subscribe.
Our editors' picks for the day's top news and commentary delivered to your inbox every weekday.