With three polls in the last week from  IPSOS-Mori, YouGov and Scottish Progressive Opinion, we anoraks have had plenty to chew over.  They all employ quite different methodologies, so comparing them is a bit like trying to find points of similarity between broccoli, carrots and onions.

But they all indicate a significant shift in support towards the SNP.  Without any doubt, the electoral sands are shifting in Salmond’s favour, though we should be wary of hailing the victors based on what is being reported, for there are some fine details that are worthy of consideration.

First, the figures are being massaged a little at the edges.  This pre-occupation with dismissing don’t knows and uncertains to vote has a bearing on the findings.  The IPSOS-Mori poll gives the most detail with which to examine this phenomenon.  The headlines gave the SNP an 11 point lead on the constituency vote (45% to Labour’s 34%) and put them head by 10 points on the regional vote (42 to 32).  But that finding is based on participants’ voting intention and which party they are inclined to support and how certain they are to vote.  Which means rather than simply being a pure who will you vote for result, it has other things factored in as well.

One immediate consequence of doing this is to lower the sample size from the 1002 people polled to 681 respondents (or 667 if the weighted figure is used and we want to confuse matters further).  Thus, the margin for error increases on a lower number and moreover, it is universally accepted that you need a sample of 1000 to make a survey fully representative.

If we include all participants, the sample size goes back up and the figures shift slightly.  On the constituency vote it now looks like SNP on 43 and Labour on 34, a margin of 9 per cent, and on the regional vote, it is 40 to 32, a difference of 8 points.  See below for how this might impact on seats in Holyrood.

The sample then drops dramatically for the question on which party people are inclined to support.  Looking only at the findings for the constituency vote – I’m trying to simplify this, honest – we see that only 294 responded to the question and of that number, 88 were undecided or didn’t know and 73 refused to answer.  This means that only 133 people gave a substantive response, yet this finding was applied to the whole sample to come up with the headline finding ie SNP on an 11 point lead.

So what happens if we take out all the extraneous stuff and simply ask all people whom they intend to vote for?  We get a very different result.  On the constituency vote, the SNP is on 29%, Labour on 20%, Tories on 7%, Lib Dems om 6% and the don’t knows are on 23%.  For the list, it’s 28% SNP, 21% Labour, the Tories and Lib Dems on 7%, Greens on 4% and the don’t knows on 18%.

It is still a commanding lead for the SNP but suddenly, the large numbers of don’t knows – as also found by Scottish Progressive Opinion - become more relevant.

The adjustments for certainty to vote are much more subtle in YouGov’s poll findings but it is not clear how many people fall out of the sample because that detail is not given.  However, the vagaries of the electoral system mean that even a percentage point of difference – Labour is on 32% when the findings are adjusted compared to 33% when they are not – results in Labour gaining another couple of seats when those findings are entered into ScotlandVotes predictor.

It is also worth noting that YouGov’s sample size has increased considerably.  In March’s poll, it was 1025; last week’s Scotland on Sunday poll was based on a sample of 1135 adults;  and this week’s sample was 1332.  It should make for a more accurate picture but there is still considerable weighting being applied, one presumes on the basis of Westminster voting record rather than previous Scottish Parliament vote share.

The general point holds true: every time the raw data is poked around with, no matter how stringently rules are adhered to, there is a risk of contamination and affecting the findings.

And every time don’t knows are airbrushed out of the equation, the findings are being skewed somewhat.  Some of these will be genuinely unsure voters who will make up their minds at the last minute, some as late as when they reach the polling booth;  others will actually be won’t says ie they know how they are going to vote but they won’t share it;  others can now be considered as won’t votes.  At this late stage of the campaign, if people have not made their minds up, often they simply will not bother to vote at all.

So while the large numbers of undecideds might still give the Labour party a glimmer of hope, their potential for causing a swing back is diminishing day by day.  Indeed, some will simply make for the winners’ bandwagon which is more good news for the SNP.

Finally, a percentage poll lead does not translate into a gain by that margin.  A ten point poll lead translates into a much smaller swing which means most Labour MSPs will escape, albeit with their majorities scythed.  The SNP’s potential gains are just as likely to come from the Tories and the Liberal Democrats but in all cases, local factors and incumbency come into play.  On the face of it, Ayr and Galloway and West Dumfries should fall but both have longstanding, respected Conservative MSPs.  Moreover, both Tavish Scott and Iain Gray are unlikely to lose their seats and Kevin Stewart in Aberdeen Central, as depute leader of a council forced to make horrendous cuts to balance the books, might find it harder to shift an incumbent MSP than all the polls suggest.

It all adds up to a great big headache and a couple of truisms:  all to play for and the only poll that counts is the real one on 5 May.