Dear people commissioning polls in preferential elections
(like the Wellington mayoral race),
please take a lead from Australian pollsters, who have gotten used to polling in elections which use preferential voting, and ask people which of the 'serious' candidates they will rank higher. Don't just ask people who their first preference is.
If you do this, your poll may actually have a chance of predicting something useful.
Also, don't only 'poll' your readers!
Regards,
Graeme
Ooops
Word to your mother!
talking about counting in preferential systems - the current reporting in the press of the labour leader result Cunnliffe got:
He received 18.2 per cent of the caucus vote, which was worth 40 per cent of the total.
He received 27.11 per cent of the membership vote, which was also worth 40 per cent.
And he got 15.6 per cent of the affiliated union vote, which made up 20 per cent of the total.
is quote arse-backwards - based on this the middle phrase should be more like:
“he won 67% of the membership vote which counts as 40% resulting in 27.11% of the total vote”
(rejigg the other two the same way)
Someone doesn't really understand this stuff at all
About half the population don't have adequate numeracy skills (such as being able to check their payslip) - I think the media reflects that in their failure to cope with anything that requires a grasp of numbers.
I have a hard time reading that Dom Post article and I’m a statistician! Did the online version lose all its formatting, or was it published that way? Yeeech.
Yes, the two requirements for polls to pass muster are:
(1) get a representative sample from the right people, where you can calculate the probability of selection for each person. Self selected polls need not apply. Reader polls need not apply.
(2) report the predicted outcome using the right metric to get there. If preferential voting is used then you have to collect the right information from informants so you can replicate the procedure used in the counting.
The Dom Post seems to fail on both counts.
What sort of nonsense (or mischief) is going on when we see
…had support from 27 per cent of the 635 readers
and
Of those surveyed, 275 were eligible to vote in the Wellington City Council elections
in the same article?
Hebe, in reply to
What sort of nonsense (or mischief) is going on when we see
…had support from 27 per cent of the 635 readers
and
Of those surveyed, 275 were eligible to vote in the Wellington City Council elections
in the same article?
Fails on multiple counts.
izogi, in reply to
talking about counting in preferential systems - the current reporting in the press of the labour leader result...
Trying to report preferential results as a flat percentage seems very dodgy to me in general, given how systems like STV will stop counting for each candidate once the candidate has crossed a known threshold.
eg. The Herald reported that David Cunliffe won the Labour Leadership with 51.15%, which isn't strictly true. David Cunliffe had won automatically with his 51.15% by the first round because 50% was a mathematically unassailable threshold for three candidates, which is probably why the Labour Party website reports it as a final result from round one, but it's not very representative of the final tally after round 2 which shows that after Shane Jones was knocked out, David Cunliffe had far more preferential support (albeit after the weightings) over his closest rival than a mere 51%.
Given the Dominion Post's considerable support for their preferred candidate John Morrison - which extends to not even reporting this fiasco - it's difficult to see their use of statistics as anything other than the worst form of propaganda-driven push-polling.
izogi, in reply to
The DomPost has reported it now, albeit late.
On Sunday the WCC Watch blog (kind'a biased in its own way) speculated that upon questioning by the DomPost, John Morrison simply denied that anything actually happened. Not that this is a good excuse to avoid investigating further if it were deemed newsworthy. There was hardly a lack of witnesses.
Thanks for this posting, Graeme.
As you imply, to get a more accurate idea of which of the serious candidates is preferred by a majority of likely voters, two-candidate preferred (TCP or 2CP) polling is required.
The questions could be something like these—
Q. If the [name of territorial authority] mayoral election was held today, to which candidate would you probably give your first preference vote? If not sure, which candidate are you currently leaning toward?
Q. If your first preference candidate were to be eliminated during the count, to which other candidate (your second preference choice) would you wish your vote to be transferred? (If the respondent can’t answer, no matter.)
Q. If your second preference candidate were to be eliminated during the count, to which other candidate (your third preference choice) would you wish your vote to be transferred? (If the respondent can’t answer, no matter.)
NB. Respondents who answer ‘don’t know’ to the first question are not included in the results.
Given that this type of polling would be a first for New Zealand; that we’re talking local elections here, and that, consequently, most people approached “on the fly” would have trouble giving more than two preferences, little would be served by pushing on and asking a fourth question.
Once the pollsters have gathered the voting intention data, they then perform a notional count, through as many iterations as are necessary, to determine the last two candidates standing (even if one candidate has attained an absolute majority of votes before the other most-preferred candidate has been identified). The total of votes given for each of the two remaining candidates is then expressed as a percentage of the total of votes remaining in the count, to produce the 2CP result.
For people who knew what they were doing, a (manual) hand-count would not be too onerous – say, 750 people polled; up to just three preferences given? Easy. However, for the record, there is at least one computer program available (ahem).
The poll result would be reported in two parts: (1) the first preferences given for each of the candidates – say, 750 in total; (2) the 2CP result based on the lesser number of people who expressed a preference for one or other of the two remaining candidates (that number being the total of votes remaining in the count).
There’s no reason why it couldn’t be done.
Moz, in reply to
The questions could be something like these—
In Oz even in state and federal elections it's normally done as "two party preferred" or three preferences are polled when the race is close. I've never seen a report that tried to extract actual voter preferences in such a laboured way, Steve. If that's what they want they'd almost certainly go with "number your preferences". And I suspect the reason they don't is that a great majority of the respondents don't have any idea, they're just going to vote for one of the major parties. Hence the NSW donkey vote being blown out recently because the first box on the paper was "Liberal Democratic Party", 5% of voters saw a party with "Liberal" in the name and ticked the box. So now we have a reactionary Catholic senator...
Polling at all when most of the candidates have very little profile and there's no real party involvement is going to be hugely problematic, and a lot will depend on what information the poll gives the respondents about candidates, and how long they have to choose. It'd almost be more useful to regard it as a survey of social attitudes - is "Swetha Vapanyanisitanpynata" getting fewer votes than "John Smith" because she's: 1: Labour 2: Indian 3: female or 4: incompetant?
Steve Todd, in reply to
Thanks for this, Moz.
I'm not exactly how sure how it would be done, but we have to come up with some way of polling the voters in respect of mayoral elections conducted by preferential voting that gives more meaningful information as to who the leading candidates are than did the Dom Post poll that prompted Graeme to put up his posting.
In mayoral elections, the great majority of candidates are independents. Therefore, it seems to me we have to determine the two-candidate preferred situation. I came up with how I thought it could be done, but I'm no pollster, so I'm really just floundering.
In the Wellington mayoral race, there are certainly four candidates, in my view, deserving of serious attention. I take your point that most respondents wouldn't "have any idea", which is why I suggested no more than three questions.
Hopefully, a polling company can step up and give it a go, doing as Graeme suggested, take heed of how the Australians do it, but I'm not holding my breath.