I don’t imagine that Bob would disagree with anything I’ve written (in fact, I’m sure I remember you or Bob writing a very similar article pointing out that different pollsters are measuring slightly different things in a pre-election column- I looked for it when writing this but couldn’t find it on the new website!).

I think Bob’s original article did carry the implication that this the only reason for the difference – doubtlessly unintentional since I know he doesn’t think that. Take that difference into account and the consistency of the polls is even more notable.

I think you have missed the point of what Bob Worcester was trying to say. Of course it is true that that different polling methodologies can and do cause systematically different results, and Bob would certainly not disagree. There are consistent differences between the results of the different companies at the moment. But they are small ones.

Nobody is suggesting that these differences are no more than random variation – Bob does not “put it down to normal sample error”, he simply points out that the differences that exist are no bigger than would have to be expected if it were no more than normal sampling error. None of the polling companies use simple random sampling for their polls, so the statistical probability laws and the standard “margins of error” do not strictly apply in any case. In fact we use different sampling methods, we interpret the data in different ways, and for that matter we are not necessarily trying to measure the same thing.

You would think that we ought to be all over the shop – but, oddly enough, we are not. Although it is possible to detect systematic differences caused by the differences in method over a number of months, that difference is so small that it is within the compass of what we would consider the “margin of error” if were using simple random samples. Within any one cycle of polls, the differences between the various polls are not generally more than we might expect if they were all random samples and measuring something much less complex than voting intention; there are occasional outliers, of course, but the number of outliers from the whole group of polls over a period of months is no more than would be expected if they were all random samples (even though the different poll series do not have the same mean). Even though it is possible, knowing which company conducted which poll, to detect systematic differences between them, taken as whole they don’t look very different from what you would expect if they were all random samples.

What this means is that, despite their differences in methodology, all the polls are telling us the same thing. Two polls can only be considered to disagree if their results are so far apart that there is no possible “true” result within the “margin of error” of both of them. Bob’s point is that, except for the occasional outliers (and there are only about as many of them as one has to expect in any sampling exercise), none of the polls are disagreeing with each other in this sense. That is so even though with non-random sampling these margins of error are more a rule of thumb than something we can strictly expect to apply on statistical grounds (and even though the conventional plus-or-minus-three points margins of error which Bob is using assume a sample size of 1,000 when nearly all the polls have effective sample sizes considerably lower than that).

Furthermore (although Bob does not go into this in detail in his article for Political Betting), the polls are not static. If you take the record back to last year, before Gordon Brown became PM, you can find this same degree of consistency between the polls even when the state of public opinion was very different. Month after month the polls have been sufficiently in step as Labour support has risen and fallen that almost all have been within this narrow margin of the running average.

My impression is that their ratings are volatile because the majority of their support is entirely conditional, whereas Labour and Conservative are mostly tribal.

I’d have thought it was more sensible to take a contingent position as a voter and be able to switch the party you support – I certainly would not support a donkey in a rosette!

FWIW I describe myself as an independent liberal because I make my own mind up on questions as they are presented to me. I have no inherent objection to supporting any party except in the way the issues are framed.

If you do have some principled objection to voting for any particular party (which appears to be the case) then by definition you ARE generalising far too broadly to be accurate.

Each of the major parties do represent different philosophical traditions and behave differently for this reason, so of course it is entirely consequent that all ideologically-defined tribal supporters will fail to rationalise the poll fluctuations of their opponents accurately enough to help themselves.