Posts by Gavin White
Last ←Newer Page 1 2 3 4 5 Older→ First
-
By focusing on the cost of the text-in / website 'poll', people are missing its true idiocy. The 'poll' asked 'who won the debate' (past tense), but people were able to vote right from the BEGINNING. So many did so that the online voting system crashed by the FIRST ad break.
The closing addresses, for example, often have a big influence on perceptions of who wins a debate, but would have had negligible influence on that 'poll', simply because the vast bulk of responses would have been collected earlier. Given that the website crashed early on (meaning that it was receiving a lot of votes), it seems probable that a significant percentage were submitted in the first few moments of the debate.
If it measures anything, it measures who people expected to win, not who actually did. Coupled with all the other problems of text-in / website polls, it has no statistical validity whatsoever.
Would it have been so hard to have the poll afterwards?
-
Hard News: We can do better than this, in reply to Pete George,
Oh I agree - that's disgusting. The saving grace, as others have pointed out, is that it's just an anonymous commenter, but I'd like to see the moderators deal with that sort of crap. The biblical quote about removing the plank from your own eye comes to mind.
-
I couldn't agree more Russell. This 'they all do it' cop out is getting beyond a joke - there just isn't an equivalent to Slater on the left. Any cursory read of the Standard shows that they've attacked Labour plenty of times, and H-Fee (trying to prove that Key had a hand in an action which could have crippled NZ's economy) is clearly different to fishing for evidence of who politicians have been sleeping with.
-
Hard News: Decidedly Undecided, in reply to Sacha,
Like Andrew I'm generally strongly against the reporting of polls to 1dp - it implies a level of precision that we just can't deliver.
There is a case, however, for reporting parties below 5% to 1dp in some circumstances. If we round to 0dp, then parties that poll 4.5% will be shown as over the threshold.
People who aren't familiar with stats might not know that the margin of error shrinks as the survey percentage moves away from 50%. For a sample of n=1000, while the margin of error is +/-3.1% for a 50% figure, if the survey percentage is 4.5%, the margin of error shrinks to +/- 1.3%.
Consequently, if a party polls 4.5%, then we're saying that 19 times out of 20 their actual vote lies between 3.2% and 5.8%. That's a bell curve too - I don't have an appropriate calculator in front of me, but it's important to remember that a lot of those 19 occasions should be a lot closer to 4.5% than either 3.2% or 5.8%.
Compare that with a party that polls 5.49% (which, like 4.5%, rounds to 5%). 19 times out of 20 that party's vote should be between 4.08% and 6.9%. Again, if you bear in mind the bell curve idea, it's more likely that the 5.49% party is over the threshold than that the 4.5% party is.
When you get down to parties polling below 1%, then it's more sensible to report them to 1dp. If a party polls 0.6% (United Future's vote at the last election), then the margin of error in a poll of n=1000 is +/- 0.5%.
If you're interested in the stats on this, and also some early 2014 polling on the Conservatives and the Internet party (before them joining with Mana), I wrote about that here. I had a grump there at the tendency to describe parties as being 'below the margin of error' - as that blog discusses, that's actually impossible.
-
Hard News: Decidedly Undecided, in reply to Andre,
Andre - it's worth remembering that the Political Scientist Blog's conclusions were based on nine surveys. Both Andrew and I have looked at our respective polls over the same period (20+ polls in both cases) and come to the same conclusion - undecideds have increased, but the increase is by no means 'marked'.
In our case it's increased by about 2% on average between mid 2012 and in the first half of 2014. It's an interesting piece of information, but not enough to explain the movements in the party vote.
-
Thanks Russell for the opportunity to be on this show - and everyone else for their feedback. Andrew's certainly been fighting the good fight to date!
One issue we discussed was TV3's reporting of NZ First's poll rating of 4.9% as definitely being below the threshold. That's statistical nonsense, as I've explained on my blog here. If you don't want to go the blog, the long and the short of it is that, if a given party polls 4.9% in a representative survey of n=1000, the probability that they're really on 5.0% or greater is 49.6%.