Monday, March 14, 2011

Polls Apart

There was a very interesting set of polling numbers released on the weekend, faithfully leaked to the Twitterverse by @ghostwhovotes.  They can be seen here:

At first glance, the numbers seem to be a little encouraging for Labor.  Primary vote has risen from 23% in Jan-Feb to 26% in Mar 2010 (down from 39% at the last election).

Think about that, just for a second.  Even on these improved numbers, one out of every three people who voted Labor in 2007 will not be doing so this time round.

The strange thing is that the Coalition has picked up 4% as well, moving from 46% to 50%, up from 37% in 2007.

That's an impressive number too - a full half of the population marking Liberal or National.

So, where have these numbers come from? It's from the Green column - down from 17% in Jan-Feb to 11% in Mar 2010 (down from 9% in 2007).

Is that actually possible?  Could it be the case that one out of three Greens voters have jumped ship in the last month?  I doubt it.

Some people on Twitter theorised that the fall was, at least in part, due to the announcement of the Carbon Tax.  Personally, I don't think so.

To have moved one's vote because of the Carbon Tax, this hypothetical Green voter needs to so fundamentally misunderstand that Federal/State divide that he/she thinks that their state vote matters on this issue, but have a nuanced enough understanding of the present state of Federal politics that he/she can see the Green fingers all over this move.

Such a voter may exist, but surely not more than one or two of them?

It is possible that this is some sort of Green backlash, but I would have thought that most people who already feel that strongly about the issue are already firmly lodged in the Coalition column.

Far more likely, in my view, is that this result is just statistical noise - a false positive, if you will.

Let's look at the small print.  Obviously only people who are home at the right time of day, interested in being part of a poll, and additionally able to speak English would be polled.  Newspoll no doubt balance for all these things, and besides, the same restrictions applied to the previous poll.

The important point, however, is that only 1000 voters were surveyed in this poll.  1000!  That's such a minute quantity of the NSW population that we're immediately taking a risk extrapolating.

Now, no poll can be truly random.  That goes without saying.   But it's still very easy to forget that polling is an inherently inexact science.

Even if the poll was 100% random, there will still be variation.  Indeed, this poll (like all good polls) tells us that the "maximum margin of sampling error is plus or minus 3.1".  Stretching back to the small amount of statistics I studied at uni, this tells us that there is a 95% chance that the "true" result is no more than 3.1% more and no less than 3.1% less than the stated figure.

Now, the Green vote plummeted 6%, so we're outside that standard deviation - but notwithstanding that, 5% of results will still fall outside that range.

At the end of the day. there may well have been a fall in the Green vote, and it may be a significant one.  But I severely doubt that it could possibly be 6%. I'd be very surprised if there was not an "inexplicable" rise in the Green vote in the next poll back towards the kind of figures we saw in Jan-Feb.

Polls are important.  They let us and politicians know what is going on, and when it is time to punt the old premier and rotate round again. If the Greens are smart, I'm sure they'll be calm about this result.  As the cliche goes, there is only one poll that counts.  

As a final note, according to these numbers, 11% of people think that Labor will win, including 4% of Coalition voters! I guess there are some questions no poll can satisfactorily answer.

No comments:

Post a Comment