In the immediate aftermath of the Scottish referendum there are many details of the campaign and the data to examine (expect more here next week). One particularly interesting feature is how the polls carried out during the final weeks of the campaign compare with the actual result, writes Oliver Hawkins in the House of Commons Library blog, Second Reading.

The following chart shows the distribution of polls whose last day of fieldwork fell during the final two months of the campaign. The columns show the number of polls reporting a given percentage of people intending to vote Yes, once undecided voters are excluded. The dotted green line shows the actual percentage of people voting Yes in the referendum. Both the polls and the actual result are rounded to the nearest percentage point.

As the chart shows, the most frequent result for Yes was 47%. Superficially this doesn’t seem too far from the actual result of 45% — it’s within the margin of error for a single poll. But looking at the polls in aggregate, it’s clear there is a systematic difference between the estimated level of support for Yes and the percentage of people who actually voted that way: 23 of the 29 polls conducted in the last two months of the campaign estimated support for Yes at 46% or more.

Of course, what is missing from this analysis is any consideration of the trend in the level of support for Yes during the final two-month period. It’s possible there was a late swing to No. But looking at the 14 polls conducted in the week before polling day, all of them estimated the level of support for Yes at 46% or more, with an average result of 48%. Furthermore, as Lord Ashcroft’s post-referendum poll indicates, most voters made up their minds much earlier than this.

Further analysis of the Scottish referendum results will be published next week here on the blog and in our forthcoming research paper.