The Narrowing

Today I thought we might have a look at the grand old wives tale of pollypunditry, “The Narrowing”. – The inexplicable tightening of polls when the election is called.

Not being a great believer in fairy tale determinism myself, preferring reason backed by the copious use of data to derive meaning from polling behaviour and thus rational explanations for voter behaviour, I’m somewhat bemused by the almost religious qualities that some pundits give to The Narrowing.

But before we get into this, it would be good to take a small overview of the general behaviour of polling data first, as this will come into play later on.

To start with, let’s look at the primary vote of the two major parties since April 1996 using every published Newspoll. The vertical black dotted lines are the elections of 1998, 2001 and 2004 respectively.

The obvious thing that stands out here is how noisy these buggers are. In any given 3 or 4 poll period, you can have poll results falling within a 6 point margin. This, of course, is to be expected simply as result of sampling variation – the dreaded margin of error. As a result, these polls are generally too noisy to use for nearly everything.

Instead, let’s look at the monthly average of Newspoll for the Coalition primary over the same period of April 1996 through to September 2007.

This gives the Newspolls the valium treatment and they become far easier to work with. There is still noise in there, but as you can, it’s been seriously reduced.

By using the less noisy monthly averages of Newspoll, it also makes it easier to see some interesting behaviour in the series. Notice how the series is very “peaky”? Public opinion reacts to the events of the day, but these peaks also suggest an overreaction. Tampa/S11 is a good example here. When those two events occurred, the vote reaction was to rally to the government, but that rally also contained an overreaction component – the bandwagon effect which gave the government a very short polling peak before that bandwagon effect washed out of the system, taking the vote down with it. This is polling overshoot and you can see its effects everywhere throughout the polling series of both the Coalition and the ALP. It normally takes a few months for that overshoot to wash out of the system after a big movement.

On the ALP side, the same thing can be seen observed with leadership changes. When Rudd assumed the leadership, the polling overshot by a couple of points into the low 50s as a primary vote before the herd effect, the bandwagon effect, washed out of the system and the ALP primary settled down into its long run level of 47 that we still see to this day.

Now armed with our knowledge of sampling variability and polling overshoot, we can move on to the next issue that must be taken into account when measuring “The Narrowing”. That issue is the polling movement in the months preceding the campaign.

Remembering back to our quick squiz at the governments advertising campaigns in the lead up to the last two elections, we found that the primary vote of the Coalition made large recoveries in the period of the taxpayer funded advertising blitz.

This is important, not only because it can drive party political momentum into the campaign (as it did against Latham) but can also be used to boost any underlying movement to the government (such as was occurring with the Tampa/S11 episode in 2001). These advertising campaigns assisted the government greatly in setting their final pre-campaign period support level from which to launch into the campaign itself. The 1998 election wasn’t renowned for its massive advertising blitz (at least compared to the last 2 elections) although it did have some GST stuff (I think – can anyone clarify?), and the One Nation effect not only clouded any advertising bounce, but generally made that election rather unusual to begin with. So we havent thrown that graph up.

Now keeping all that in your thought orbit – lets move on to The Narrowing.

We’ll look at every election since 1993.The way we’ll do it is simple. First we’ll show the monthly Newspoll primary averages leading into the campaign (to show us momentum) where the second last observation, and sometimes third last observations are the month(s) in the campaign itself, and where the final observation is the election result. So for 1993 we get:

Notice how the long term momentum was with the ALP and against the Coalition, and carried through the campaign period into the election itself? On primary votes there was a superficial narrowing, but nothing that wasn’t already happening anyway. Over the medium term, the ALP primary from November 1992 through to the election result simply oscillated around 44.

Secondly, we’ll also look at TPP, but measured in a slightly different way. We’ll use the primary votes of the Newspoll estimates and then distribute preferences as they occurred at that election. So effectively we are backcasting preference allocations from the 1993 election onto the primary vote estimations taken in the lead up to the 1993 election. This way, we end up with a more realistic and consistent TPP estimate that reflects more accurately the preference allocations of the time. Also, we’ll add in 6 months of Newspoll monthly averages (calculated the same way), then the first single poll before the election was called, and every Newspoll during the campaign until the election result itself. When the election was called and campaign began is marked with a vertical black line. Doing this we get (and this time the ALP is blue, and the Coalition red – my stuff up in the graph – apologies)

This is interesting because the November and December results look like typical polling overshoot before the estimates settled back into a more modest and realistic level that was consistent with the longer term momentum and trend. Again, there was superficially a narrowing – but it wasn’t as a result of the campaign being called, as the 49% ALP TPP going to 51.4% ALP TPP from the beginning of the campaign to the final poll of the campaign is all simply movement within the margin of error of the polling. The true underlying result ‘probably’ did narrow slightly in the campaign – although the Newspoll estimates cannot say definitively one way or another. What we can say is that the movement in the campaign was consistent with the longer term movement before the campaign.

Moving on to 1996:

Not much can be said about this except that there was absolutely, positively no narrowing at all in either the lead up to the campaign, or during the campaign itself. The numbers did not move for 8 months.

Sounds familiar, doesn’t it.

Now for 1998:

On the primaries, there was probably a slight point or two movement away from the ALP , while the Coalition primary was getting battered around by One Nation and was all over the place – eventually ending up on election day at its long run mean over the period. The TPP figures reflect the Coalition primary vote figures. There’s that big overshoot in June 1998 reflecting the One Nation push, and the final result being very close to the mean vote level over the period of February 1998 through to the October 98 election. During the campaign period itself, there was no narrowing. In fact, using just the polls taken during the campaign, they suggest, if anything, the possibility of a slight widening or a point or so. But at the end of the day that’s just noise and we can safely conclude that the polls didn’t narrow over the election campaign,

Moving right along to the 2001 election:

This polling is interesting for many reasons. The ALP from March to November 2001 were experiencing a slow decline and trend away from their primary vote. That is as clear as day and is consistent with the advertising blitz we showed earlier. The Coalition primary however was experiencing a slow trend towards them, interrupted by the Tampa/S11 overshoot in the polling. It becomes much clearer looking at the TPP figures. The August to September movement is highly pronounced, directly followed by a slow recovery back to a more normal position as the S11 issue washed out of the system. Here there was definitely a narrowing, but one easily explainable. If you look at the TPP chart, you can see clearly the longer term trend between March and August away from the ALP and toward the Coalition – forecasting that forward you end up with something pretty close to the election result.

By the election, the vote for each party had recovered to again reflect that earlier movement after it was interrupted by Tampa/S11.This type of behaviour is extremely common in the financial and commodity markets where you have long run movements getting interrupted by exogenous shocks, whereupon they undergo an adjustment process as the effects of that shock wash out of the system, and the series attempts to catch up, or revert, to where it’s long run behaviour would have ordinarily had it situated. This is a classic, almost textbook example of an impulse response function.

Now on to 2004:

 

This is pretty easy to explain as well. The ALP was undergoing a longer term trend away from its primary vote while the Coalition was enjoying a trend to it’s primary vote (helped in large part to the advertising blitz and associated pork combined with Lathams implosion).On TPP terms, the vote was pretty volatile, but with the ALP being various levels above 50. What is crystal clear is that far from the vote narrowing once the campaign was called, it blew out toward the Coalition, following what in hindsight looks like a fairly solid , even if volatile trend toward the Coalition that started 6 months or so earlier.

From the last 5 elections, what can we say about The Narrowing?

In 1993 there may have been a narrowing and in 2001 there was a narrowing, but both times that narrowing was consistent with the longer running trends occurring in the vote over the previous 6-8 months. 1993 is reasonably easy to see, even if there was a bit of polling overshoot in the mix, and even if any narrowing was not statistically significant in the campaign.But in 2001 it gets a little more complicated as the impulse response function of the Tampa/S11 exogenous shock on the polling series occurred during the election campaign period itself. But from an econometric perspective – the narrowing is nothing more than a reversion of the polling series to its long run trend via a typical impulse response function

In 1996 the polls didn’t move during the campaign, or at all for 8 months.In 1998 the polls didn’t move during the campaign, although in the period before the campaign One Nation was causing havoc with the TPP and the Coalition Primary. In 2004 the complete opposite of the Narrowing occurred.

So The Narrowing is 2 from 5 at best, 1 from 5 that can be called certain, and where both of those occasions can be explained by the longer term trends and movement occurring in the polling series over the previous months.

While The Narrowing makes for a simplistic bed time polling story and is to be expected to be used by political parties as a weapon for their own electoral self-interest, when it comes to providing actual explanations of electoral behaviour to a broader public, the dogmatic representation of “The Narrowing” as some divinely inspired inevitability is a poor substitute for polling analysis based on the evidence contained within the polling data.

Those who aspire to understand Newspoll properly would do well to spend a little less time proselytizing with faith based arguments and a little more time using the data they proudly claim as their own to provide the type of cogent, rational and evidence based analysis that would better serve the public interest.

Over at the platinum blog of Australian blogs, Larvatus Prodeo, there is an excellent debate going on over this very narrowing here and here, and well worth a squiz.