|
| | About
preference calculation for October 16 2006 post
Newspoll’s
preference strategy has changed several times and has never been great. Currently
they allocate preferences as they flowed at the last election, in total – not
discriminating between minor parties – which works ok at the moment because the makeup of
those minor parties hasn’t changed substantially since October 2004.
Formula
My two party
preferred calculation is Labor 2pp = Labor primary + 0.61xDemocrats + 0.6xothers for all surveys until end 1994
From January 1995
onwards it is = Labor primary + 0.64xDemocrats + 0.44xOne Nation primary +
0.75xGreens + 0.5xothers
Coalition 2pp is, in
both cases, 100 - Labor's.
I can’t recall my
exact reasoning for some of these numbers, but they are rules of thumb and work well at actual
election results. (At 2004 election - see table 3 here
- Greens and Democrats preferences went 81% and 59% respectively to Labor.
Ideally Newspoll would use this table to distribute preferences individually per
minor party for the current term.) It may seem too detailed for so inherently imprecise a
measurement, and of course when One Nation or the Democrats get low, they simply
become part of 'others'. I welcome any suggestions - eg for the changing
complexion of Democrat preferences. Ideally a formula such as this would take
into account actual preference flows at each election from 1987, but that data
isn't available.
Because these numbers
appear reasonable, and the formula works well at actual election results, I
reckon it is ok for calculating averages over many polls, but I'm open to
arguments. And while it doesn’t
make sense to calculate 2pp to one decimal point from rounded primary vote
numbers, the untrained statistician in me reckons it is ok when averaging over
large sets of numbers.
In any event, the table also includes primary vote
data, which tells a similar story.
Nielsen has the best
preference method – they ask respondents who will get their eventual preference – and
their 2pp was best at the last two elections. But as Newspoll only currently collects primary voting
information - and then apply their rather half-hearted calculations - we may as well do the calculations for them.
Rounding
Rounding,
however, is
an issue. Newspoll only publish rounded numbers, but presumably their two party
preferreds are calculated from the unrounded ones. Take the most recent Newspoll,
three weeks ago. It had 41 to Coalition, 42 to Labor, 6 to Greens and 11 to
others. Their two party preferred was 53 to 47. But my formula only gives a 2pp
rounded to 52 to 48. And actually Newspoll’s strategy,
of distributing all non-major party votes 'in bulk' as per last election (61 to 39 in Labor’s favour) from
Newspoll's rounded numbers, would also round to 52 to 48. So we can deduce that
Newspoll's ALP number was
actually between 42 and 42.5 and/or Coalition was between 40.5 and 41. My
calculations always rely on rounded numbers.
The same inner
statistician mentioned above reckons that the larger the number of polls I’m
averaging, the less this rounding matters.
For many years,
Newspoll never bothered with preferences at all – we might as well have lived
in a first past the post world. They first recognized them during the 1993
campaign. But until early 2003 they only recognized their existence during
election campaigns, and just published primary voting data at other times. Since
2003 they’ve always published 2pp but their method of getting them has chopped
and changed a little. (They have never extracted full preferences from
respondents.)
Crean/Latham
For the first year of
Simon Crean’s leadership, Newspoll didn’t bother with preferences. For the
second Crean year, they did calculate 2pp but (inadvertently) in a way that favoured the
Coalition. From
January 2004 - the whole Latham period bar one survey - they used a method that
overstated Labor support. The average Newspoll
published Latham 2pp was 51.5, rather than my 50.6. This repeated overstating of
Latham's lead no doubt fed into his satisfaction rating.
Newspoll saw how
inaccurate their 2pp was at the 2004 election, and went back to the Crean
method. However, that method - allocation in bulk - works ok today while the makeup of
minor party support stays about the same as at the 2004 election. (Which wasn't the case during Crean's
time.)
| |
|