Friday, November 15, 2013

What to look out for in the independence referendum polls

Because there has only been one poll so far this year showing an outright lead for the pro-independence campaign, it's been easier for the overwhelmingly anti-independence media to gloss over the huge differences in the results produced by different pollsters.  Panelbase's most recent poll wasn't all that far away from 'statistical tie' territory, while on the other extreme the last two polls from Ipsos-Mori have shown very substantial leads for No.  The four other major pollsters have come up with various shades of grey in between those two extremes.  To put it in a nutshell, the polls are all over the place, and we simply don't have a clue what the true state of play is.  However, one thing that has remained fairly constant is the extent of difference between each pollster.  There's the odd exception to that rule (for example TNS-BMRB reported a dramatic change in public opinion last year that no other pollster detected), but in general we all know that an Angus Reid poll will tend to be more favourable for the Yes side than a YouGov poll, but less favourable for Yes than a Panelbase poll.

The point being that it's very important not to look at any given Ipsos-Mori or YouGov poll, and think "oh, this means Yes needs to gain another X amount of support".  It might mean that, but there are very strong reasons for doubt.  What would probably be more useful is to use the poll for an extrapolation of what we'd expect other pollsters to show, based on the assumption that the now-familiar differentials will remain stable.  Here, for example, are various benchmarks for interpreting Ipsos-Mori polls -

If Ipsos-Mori show a 20 point lead for No, then Panelbase would probably be showing a very small Yes lead.  Angus Reid would be showing a small No lead, and the other three pollsters would be showing No leads of between about 8% and 11%.

If Ipsos-Mori show a 15 point lead for No, then both Panelbase and Angus Reid would be showing leads for Yes (albeit in Angus Reid's case only just).  ICM, TNS-BMRB and YouGov would all be showing very modest leads for No, of between 3% and 6%.

If Ipsos-Mori show a 10 point lead for No, then Panelbase, Angus Reid, ICM and TNS-BMRB would all be showing leads of various sizes for Yes.  YouGov would be showing a tiny No lead.

So although it would be wonderful to go into polling day with a lead for Yes across all pollsters, it isn't essential.  In Ipsos-Mori's case, we just have to get the No lead down a bit - the company could still be showing a double-digit lead for No, while four of their five competitors show leads for Yes.  Without a crystal ball handy to tell us which pollster is going to be proved right, that would be more than enough for us to reasonably conclude that we have at least a very good chance of prevailing.

3 comments:

  1. My money is on Yes never being allowed to be ahead.
    I predict a 65 Yes / 35 No result, but the polls in the week before the referendum will show a 95 No / 5 Yes.

    ReplyDelete
  2. I'm no expert, but what I have been noticing is that while the press have been screaming that Yes has not been showing any positive movement in the Polls, they haven't reported that the 'No' side have steadily lost support in all of the polls.

    Or to put it another way, if our MSM was balanced in their reporting, they would be screaming 'Polls continue to show trend towards a Yes victory'

    but hey 'never interrupt your enemy while he's making a mistake'

    ReplyDelete
  3. For us old-timers the 1970 General Election showed Labour coasting to victory according to all the opinion polls except one. The exception proved to be right.

    ReplyDelete