Further to Kate’s post on the developing scandal at the CBC, watch this:

Here’s what some of the main protagonists had to say when this, erm, educational tool was unveiled with great fanfare during last Sunday’s national news broadcast.
Reporter Paul Hunter:

“Politicians will have to take note…after all, CBC News will be reporting the (Vote Compass) results throughout the campaign.”

CBC News Editor In Chief Jennifer McGuire:

“As a public broadcaster we have an obligation around engaging Canadians in democracy…(Vote Compass) is a learning tool, you get to have insight into the actual party platforms…”

Cliff van der Linden, one of the four University of Toronto graduate students who, along with “some of Canada’s top political academics,” spent months preparing Vote Compass:

“People might be surprised at where they end up….they voted a certain way all their lives, and then, suddenly, they answer these questions truthfully and honestly, and at the end of the day, maybe they’re closer to another party.”

Update – From the comments;

I watched the video and went through the survey questions. The video’s methodology does not conclusively show bias, but it does demonstrate that something is amiss with the “compass” and that it is not to be trusted. I do not have access to the “compass” source code, so I can only guess how its analysis engine was implemented based on the methodology used in the video.
Is there a simple algorithmic explanation for why no matter which button the user always clicked, that the “answer” always came out as “Liberal”? I think there is. After reading through the questions, it appears that there are approximately an equal number of “positive” and “negative” questions that effectively cancel each other out. I expect that I am not being the most clear (and frankly I am having difficulty coming up with the phrases I need).
In an attempt to better describe the situation I am trying to explain (rather poorly I expect), imagine the following …
For type A questions, let the strongly disagree, somewhat disagree, neither, somewhat agree and strongly agree map to -2, -1, 0, +1, and +2 respectively. For type B questions, let them instead map to +2, +1, 0, -1, and -2. If there are an equal number of type A and type B questions (say conservative slant and liberal slant), then always clicking the same button will lead to a sum of 0.
If all the clicks were thus summed, then likely each party would be given a range of values that mapped to them. Assuming this is the algorithm that was used, then it would appear that anything that adds up to near zero would fall under the liberal “compass” point.
If that were the algorithm that was used, then I think that could explain the observed “bias”. I have not determined the distribution curve of sums, but my gut tells me that most will be near zero–and whatever party has been designated as the “sums near zero” will pop up as the answer most often.
Again, assuming that this was the algorithm, if this is deliberate, then there is bias. If it is not, then the developer who wrote this code ought to be slapped upside the head for failing to think.

Update – this is also very good.