Sunday 24 October 2010

It's piqued my curiosity

I was having a wee look at the opinion poll done by Yougov for the Scotsman and noticed something that made me look again.  Remember the constituency polling figures were SNP 34%, Lab 40% Con 14%, LD 8%, but the unweighted figures were SNP 506, Lab 403, Con 138, LD 78 (no unweighted figures given for the Greens) - that would give the SNP 45% of the distribution among the four largest parties, Labour just under 36%, Conservatives just over 12% and the LibDems 7%.

I appreciate why the weighting is done and how it should increase the accuracy of the poll but I question whether Yougov might have gone a little far.  The change made by the weighting saw the SNP have 113 votes removed while the three other parties all saw increases (the LibDem vote wasn't increased by enough to make a difference to the percentage they got but it was increased).  If memory serves, Yougov has been a little off in its Scottish weighting before.  It's just a wee question I've got hanging about there, but maybe Scotland's Party is actually ahead just now.  Of course, even if the poll with its strange weighting is accurate, we're still ahead of where we were in November of 2006 when the SNP was at 32% in the constituency vote and 28% in the regional vote compared to the current 34% and 31%.

Och, what an interesting wee time we have ahead of us!

6 comments:

Robert said...

Hi Calum, great blog. Could you explain in easy terms why and what the weighting is about (for a simpleton like me!)

Cheers,
Robert

Calum Cashley said...

Not sure I could do as good a job of explaining it as the people who have already written on it.
http://www.straightstatistics.org/article/less-meets-eye-sceptic%E2%80%99s-guide-opinion-polls

http://www.ipsos-mori.com/newsevents/ca/ca.aspx?oItemId=299

http://news.bbc.co.uk/1/hi/magazine/8588747.stm

http://news.bbc.co.uk/1/hi/uk_politics/election_2010/8650282.stm

Wish I knew how to make these thigns actual links in the comments section. In essence, though, the weighting exercise is carried out to ensure that the poll sample is representative of the population. Yougov includes weighting for political preference (and who you voted for last time) so has additional factors to consider and therefore additional ways to refine their sample for accuracy - or additional ways to get it horribly wrong.

Vronsky said...

To create a link in a post use the template below. Replace the curved brackets with pointy ones ('less than' and 'greater than' signs), replace URL with the web address you want to link and LABEL with the text you want for the link (the text the reader will click on).

(a href="URL")LABEL(/a)

Here's a test link - to - uh - an Austrian brass ensemble singing Bohemian Rhapsody.

Calum Cashley said...

Marvellous, Vronsky, thanks very much. I'll have some BoRha later today and watch the reactions in the office!

masterymistery said...

this is why i avoided a career in politics...

recent research suggests politicians now rank below lawyers in thew most reviled stakes

masterymistery at cosmic rapture

Calum Cashley said...

No it doesn't.