Thursday, July 6, 2017

How Often Are Federal Newspolls Released?

A humble little subject, but I just thought I'd put up a resource piece about how often federal Newspolls are released and have been released over time.

The timing of federal Newspolls is frequently a subject of discussion, much of it clueless or biased.  One fairly prominent claim links the switch to federal Newspolls in the early 1990s with the frequent turnover of major party leaders, although there is actually no evidence that this is true at all.  On social media, Newspolls are eagerly awaited and considered "due" every second Sunday, mostly by one-eyed Labor supporters.  If Newspoll fails to appear this is claimed to be evidence that it is being "hidden" because the results are bad for the Coalition.  If this were actually the case, Newspoll would skew towards the Coalition compared to other polls (it doesn't), Newspoll would have gone AWOL during obvious Coalition low points like the Hockey budget, the Prince Philip knighthood and even the collapse of Malcolm Turnbull's "utegate" attack on Kevin Rudd (it didn't), and Newspoll conspiracy cranks would be able to post reliable advance predictions of when Newspoll would come out (they don't.)  But those tweeting these nonsense never let the facts get in the way of their inane barracking.

This week I saw a new strain of the viral dumbness that is Newspoll truthism - a claim that Newspoll was becoming less frequent in order to string out the time it would take for the Coalition to lose 30 consecutive Newspolls on Malcolm Turnbull's watch.  (Turnbull has lost 14 in nine months, while Abbott's 30 spanned sixteen months, making Newspolls 17% more widely spaced so far during Turnbull's losing streak.)



Newspoll scheduling is also significant because of theories that certain politicians attempt to pump the Newspoll cycle by making announcements or attacks (on other parties or, more often, their own) at a time when they are most likely to influence voters who are being polled by Newspoll.  It's hard to tell if this is purely confirmation bias (people notice such cases when they match the pattern and ignore them when they don't) or if there's actually something in it.  At a time when Kevin Rudd was being widely accused of this, Scott Steel found that media mentions of Rudd didn't increase significantly during Newspoll cycles.  Beyond that I'm not aware of any attempt to measure the issue.

Below is a table showing the intervals between Newspolls in weeks in every year since Newspoll started, with the start and end dates for polling in that year indicated.  In each case the date used is the last date in the poll sample period (up to late 2012) or the date of release (after that).  The intensifications of polling prior to elections are shown in yellow and an asterisk is used where the interval was less than a week:


The table has been fairly hastily hand-compiled and I do not guarantee it to be absolutely error-free, but there are some broad details I want to stress:

* Newspoll became "fortnightly" in 1992, around the same time Paul Keating became Prime Minister.

* Newspoll takes a long break over summer, usually ceasing in early or sometimes mid-December and returning between mid-January and mid-February.

* Newspoll is usually more often seen on average in election years than in non-election years.  From 1992 to 2016 the average was 25.8 Newspolls per election year versus 23.3 per non-election year.  There tends to be a tight packing of polls before the election and a long break after it.

* Newspoll has become slightly less frequent in more recent years, dropping from an average annual poll tally in the mid-20s per year to around 20.

* Although described as a "fortnightly" poll, Newspoll has always varied the schedule from time to time.  Examples of reasons for not following the two-week schedule (excluding around election time) have included:

- coinciding with major events such as leadership spills or Budgets
- coinciding with openings of parliamentary sessions
- avoiding clashing with Fairfax's former pollster, Nielsen, which came out monthly
- avoiding polling on long weekends and public holidays
- and, I suspect, contractual obligations

Every year in Newspoll's "fortnightly" history saw at least one disruption to the "fortnightly" schedule that was not caused by elections.  The average from 1992 to 2014 was 2.7 such disruptions per year, with only one disruption each in 1993, 2000, 2004, 2005, 2007 and 2009 and as many as seven in 1997 and 2003.

In 2015, the Newspoll brand was transferred from being run by its own company to being administered by Galaxy, which uses a different combination of methods (online polling and landline robopolling) and typically a somewhat larger sample size.  Around this time it became known that Newspoll was polling not fortnightly, but six times per quarter (presumably excluding the summer break), which was always going to mean the odd three-week jump now and then.  From this year's output so far, with six gaps of three weeks (assuming Newspoll comes out next week) and two of two weeks, it might be surmised that the schedule required by News Ltd has become even less frequent.  This would be consistent with a general wind-back in polling early in this government's term across the media, with both Ipsos and ReachTEL also being less required so far by their usual clients (Fairfax and Seven respectively.)

So no, Newspoll isn't strictly fortnightly any more, but in a sense, it never really was.  It isn't "monthly" either; rather it currently usually comes out between every two and every three weeks.

One final piece of the puzzle: Newspoll sometimes comes out on Sunday nights (with print release in Monday's Australian) and sometimes on Monday nights (print on Tuesday).  The main factor influencing this is whether it is a parliamentary sitting week - a Sunday release is much more likely if it is.

1 comment: