More Polling Bogosity

I promise that I’ll shut up about polls soon, but I couldn’t ignore the results that were posted, with graphs, on the front page of today’s Vancouver Sun. The accompanying article is here (b0rked URL corrected), but it may disappear behind their magical subscription-only Berlin wall. The important part of the article is noted in small print under the graphs, and deep within the article:

The poll was conducted on June 16, the day after the English debate, and is accurate within four percentage points, 19 times out of 20.

A mere 600 people? And four percentage points? That means that national Liberal support could be at 31% or 39% percent–a government-making difference if there ever was one. Things get more absurd when you consider that they include numbers for the provinces. Of those 600 surveyed, they spoke to maybe 70 British Columbians–and they’re actually citing this as statisically meaningful information? They’re claiming they know how BC will vote based on the 70 people (another estimated 350 people refused the request) who agreed to speak with a pollster? That’s simply shameful reporting, and, statistically speaking, has almost no bearing on reality.

Incidentally, can someone help me with some math? If there’s a statistical error of 4% for 600 people, how does that change when we examine a subset of 70 people. It increases, obviously, but by how much? Surely you don’t increase the error percentage by the same factor that you divide the set by (in this case, about 9)?