Newpoll's preferences

Danger!! This page has an exceedingly high esoteric rating!

June 2003 update

A tale of notional distribution of preferences by the currently most respected polling outfit in the country.

The story so far ....

Each fortnight The Australian publishes Newspoll results from the previous weekend. With Morgan, the only other regular pollster, getting the last election wrong, Newspoll is seen as the most reliable and people hang off the results - to a ridiculous extent.

Newspoll only asks respondents for preferences during election campaigns. At other times they just ask which party will get a '1' next to their name - the primary vote. During campaigns they ask a further question: which major party is likely to get their preferences. As far as I know, the other polling outfits ask people that preference question the whole year round.

Now, as you know, at federal elections you must number every square. This means eventually you must put the Coalition ahead of Labor or vice versa.

more on preferential voting                                             how important were preferences at the last election?

With continuing erosion of the major parties' votes, primary support is becoming a less and less accurate reading of likely election outcomes. Specifically, in today's climate, just looking at primary votes overstates the Coalition's lead because there is a big pile of Green votes out there that will mostly flow to Labor (where they came from in the first place.) Yet this is what the Oz always did (apart from campaigns) until February 2003.

That was when I had this this piece in "Lies & Statistics" in the Financial Review. You should read it before continuing. Table below is a companion piece, with my estimates, from flows at the last election, of two party preferred votes. That is, the last two columns weren't reported in the Oz, just the first six. Hovering over those last two columns gives the Australian's headlines and commentary

See The Australian's headline by hovering your mouse over the Coalition's two party preferred number (second column from right); hovering on Labor's 2pp (right hand column) gives segment of the article - generally the first par plus other snatches.

Based on Newspoll's rounded primary numbers, the ALP is ahead after preferences in the purple results, nine in all, eight between late February and late September. Then came the Bali bombings, and the government led until late April/early March, when Labor took the smallest of leads.

See graph compared with Beazley's performance.           See Newspoll and Morgan graphs.

Now, someone read that AFR column because the very next Newspoll - and all since then - reported in the Oz gave notional two party preferred votes, based on the flow at the last election.

This was good, I thought, all was well with the world. But as the fortnights passed it started to dawn that their notional calculations were based on the non-major party preference flows as an aggregate. It was difficult to verify without being privy to unrounded numbers, so I asked Sol Lebovic and he confirmed this was so.

In May 2003 I included this par in another Lies and Stats AFR column.

"There is a small if to this. ACNielsen measures both primary and two-party preferred, while Newspoll gets only primary support and notionally distributes, as the minor party and independent preferences flowed at the last election. But it does so in aggregate, without differentiating between, say, a Green and One Nation vote. This overstates the Coalition's lead, by (and this is complicated by rounding) two points in about every second poll."

Sol Lebovic emailed me, politely asking for an explanation. What follows is the long-winded one I gave him:

"There are two parts to my reasoning.

1. I reckon that you overstate the Coalition's 2pp by, on average, more than 0.5 percent (I say why in (2.) below). Accepting that you overstate it by 0.5 percent, then, on average, every second poll is incorrectly rounded up rather than down.

Because the two party preferred gap is always an even number, rounding up increases the gap by two points.

2. The overstating comes from the significant difference between the makeup of the non-major party vote today and at the 2001 election. Mainly, One Nation has disappeared and the Greens have jumped up (although they were down again in your last poll).

At the last election, Democrats, Greens, One Nation and “other” preferences split between the Coalition and Labor by these respective amounts: 35.9:64.1, 25.2:74.8, 55.9:44.1 and 51.1:48.9.

In aggregate they flowed 41.7 to Coalition and 58.3 to the ALP; this presumably is what you use.

As I’m not privy to your unrounded numbers, I can only use the rounded ones.

Last week’s poll had those minor votes respectively at 2,6,0 and 8. Calculating the Coalition’s share of preferences (to get its 2pp) in aggregate gives .417 x 16 = 6.7. Adding that to the 47 primary vote gives your rounded figure of 54.

However, distributing them individually gives .359 x 2 + .252 x 6 + .559 x 0 + .511 x 8 = 6.3. Adding that to 47 gives a number that rounds to 53. Of course, I don’t know the unrounded primary Coalition number either. That’s why I’m talking in averages.

The difference, 0.4, is less than the 0.5 I mention, but that’s because the Green vote of 6 is the lowest you give them for months. Usually I believe you are out by a touch more than 0.5; the previous poll was. (Again, working from the rounded numbers)

I know all this seems pedantic, and in the end is only a presentational matter. But the following is true: based on the preference flows in 2001, an election at which the Coalition got 47, Labor got 37, Democrats 2,  Greens 6, One Nation 0 and “others” 8 would flow to 53.3 to 46.7 two party preferred, which is closer to 53 to 47 than 54 to 46.

Phew! Sorry to be so long-winded.

Regards,

Peter "

Anyway, that's where it stands. Time will tell whether there'll be a change in calculation of Newspoll's notional two party preferred votes. In fact, without asking Sol, we can never be sure which method they're using, because none of us have the unrounded numbers. But if they continue using aggregate preference flows they will, about half the time, overstate the Coalition's lead (or understate Labor's) by 2 percent. It's rather trivial, but if they're reporting a government lead of 51 to 49 - same as the last election - when they should be saying 50-50, then the presentational repercussions are obvious.

The end.

Below is what I wrote in February after constructing the table.

What's my point?

My point is that Labor has been ahead for about a third of the Newspolls since the last election, but you'd never know it from the reporting. What's more, as the deservedly best reputed pollster in the business, Newspoll's results are widely quoted. But because they don't distribute preferences (apart from during election campaigns) and The Oz, where they appear, doesn't either, only the primary support is reported and conclusions drawn from there.

Hover your mouse over the two April Coalition two party votes (which are under 50) to see the misleading headlines there. The most egregious report was for the Sept 27-29 survey that had Labor ahead by 1.8 percent after preferences (hover over the 50.9): the government "remains in a dominant position."

See also the ALP's two party lead of over two percent on 5-7 Apr '02 described as "neck and neck". 

The closest the Oz came to contemplating an ALP win was July 19-21 (hover mouse over the 51.4) with "Labor would have been poised to win if a federal election had been held last weekend ". But no equivocation last Tuesday for the late Jan/early Feb '03 poll with the Coalition on 51.1 after preferences: "the Coalition would have won an election held last weekend."

Note: As Newspoll's published numbers are rounded to the nearest integer, there is a theoretical problem showing the derived two party preferred figures to one decimal point. However, as I don't have the original numbers before rounding, to round them again would perpetuate rounding errors.