Scotland independence vote could be as close as Quebec referendum: Éric Grenier

WATCH LIVE: CBC News will provide live coverage of Scotland referendum results online and on CBC News Network, starting at 5 p.m. ET on Thursday.

On Thursday, Scots will decide whether to leave the United Kingdom and dissolve their union of over three centuries. The polls suggest the outcome could be as close as the one that almost split Canada apart 19 years ago.

• VIDEO: Nahlah Ayed goes inside the 'Yes' camp
• Scottish referendum: The issues that may decide outcome

The campaign for Scottish independence certainly appears to have the momentum.

Among decided voters, support for independence averaged around 40 per cent in polls conducted throughout 2013, improving only slightly to 41 per cent at the beginning of 2014, leaving the No campaign with a sizable lead. Even without dropping the roughly one-in-five Scots who were undecided, support for a continued union frequently surpassed the 50 per cent mark.

By the spring, however, the Yes campaign began to pick up steam. From March to July, support for independence increased to an average of between 43 and 45 per cent among decided voters. But that gap was still a relatively comfortable one for the No side.

• Both sides scramble to convert swing voters in Scottish independence vote
• David Cameron warns vote is forever
• Scotland referendum: Warnings mount about economic harm

Polls have become consistently better for the "Yes Scotland" campaign since the second debate in mid-August between its leader, Alex Salmond, and the leader of the "Better Together" side, Alistair Darling.

Since then, polls have not dipped below 46 per cent support for independence among decided voters. Most have pegged support to be between 47 and 49 per cent.

A handful have even given the edge to the Yes side. But averaging out the latest polls puts support for independence just shy of 49 per cent, with support for the status quo at just over 51 per cent. Though the polls have been broadly consistent (the three most recent surveys, published Tuesday, put the gap now at four points), the margin of error blurs the gap considerably.

It comes down to undecideds

That assumes, however, that undecided voters will break proportionately between the two camps.

The consensus view in general is that undecideds tend to break disproportionately toward the status quo, opting at the last moment not to take a step into the unknown. This is what seems to have occurred in the 1995 Quebec referendum, when polls suggested the Yes side was on track for a slim victory.

Yet, this has so far not been the case in the Scottish referendum campaign. From March to July, support for independence stood at an average of 36 per cent, with 46 per cent supporting continued union and 17 per cent remaining undecided.

In August, the number of undecideds fell to about 13 per cent, with neither the Yes nor the No campaigns benefiting more than the other. Support for independence increased by two points to 38 per cent. Support for union was up three points to 49 per cent — a proportional increase.

But lately support has swung strongly to the Yes side. Undecideds have fallen again, by four points to an average of nine per cent, but support for independence has surged by six points to 44 per cent, with support for the "Better Together" campaign dropping to 47 per cent. In other words, not only has "Yes Scotland" drawn some undecideds toward its option, it has also converted some former unionists.

If Salmond can get the remaining undecideds to break as disproportionately in his favour by Thursday's vote, he may prevail. He needs roughly two-thirds of undecideds to come over to his side to surpass "Better Together" and get over the 50 per cent mark.

That is a tall order, and the history of referendum polling is on the No side. But the campaign has brought the United Kingdom to the brink of dissolution and Scotland achingly close to independence. History could be rewritten tomorrow.

Note: This article reviews trends in public opinion surveys. Methodology, sample size and margin of error if one can be stated vary from survey to survey and have not been individually verified. Read more about threehundredeight.com's methodology here.