Thursday, August 20, 2009

The upcoming election

Since the election is closing in, I've been looking around on some of the coverage in the media. Especially the "party selection tests" and the statistics offered.

First of, the so-called "party selection tests". Nearly all of the tests force the user to make an opninion about the statement presented, but some have the possibility of grading the importance of each statement, or chose one or several issues close to your heart. My experience is that I seem to get the same party as my top and bottom score on the different tests, but with a large variation of the parties in-between. But then, if you are average interested in news and politics, it's pretty easy to manipulate the test.

However, what I find interesting, is that some of these tests show statistics for all the users who participated. Pretty thin statistics, but still, here are a couple of examples:

First, here's NrKs test. It basically gives you a precentage how how much you agree with each party. Thus, the one with the highets percentage is your (obvious?) choice. They also publish the average score of everybody that has taken this test. The %-score is how much on average the user agrees with the party:

I liked the test, because you were able to rate your answers. But the average of all people who participated, shows some interessting results. I don't know about the varians here, but it seems to me that there are two likely scenarios coming from this graph:
1. Politics in Norway are so similar, that there is little difference between the parties. Hence, everybody agrees with approx. 1/3 or more of any party.
2. The test desgin is faulty, so it doesn't really differentiat between the parties.
VG has a different take on it. The result from this test just give the party best matched with your answers. No percentage or anything about how you match up with the rest. VG has also published results from all participants:

Interesting, it rates FrP at 35%. Though it is possible, I don't think FrP will get 35% of the votes this election. But what's more unlikely, is Ap to get a mere 16.5%. Which is interesting. So either the people who take this test is bias and/or the design of the test is faulty. If not, people don't know what the parties they vote for actually stand for. However, I'm leaning towards a faulty design theory. This test is one of the simplest I've seen for this election.

Aftenposten has the best test I've seen so far, but they will not publish their average results untill after the election.

I didn't find any tests on Dagbladet, but they do have some interessting statistics. They have plottet averages of several surveys since May:

This is interesting. Because unlike surveys with "going up 0.004% from last week", this shows the overall trends. And most notable, is a significant increase for FrP, mostly from May to July. And Ap is pretty stable.

Dagbladet also include data from several surveys performed in May. So, for fun, I decieded to calculate averages and uncertainty (at 2SD) and plot. I know this isn't correct, but it's the best I can do with the limited data available.

Now, here are some interesting observations. Ap as very little uncertainty, indicating two things: High precision on the measured data and a loyal group of voters. This is coherent with the time plot from Dagbladet. FrP has a lot of variation. This can be explained with the fact that they are increasing in numbers, as shown in the time plot. Also worth noting, is that we can't say for certain taht Høyre is large then SV.... Though I'd be really surprised if they were equal in size after the election. For Rødt and "Andre", the uncertainty is greater then the measured number. Anyway, the graph is just for fun and the numbers shouldn't be taken too serious.

Have a nice election, and don't forget to vote. Unless you vote for the wrong party, then feel free to stay home.

1 comment:

Wilhelm said...

..alternatively, if your vote disagrees with mine and effectively cancels each other out, we can enter an agreement in which we both abstain from voting. :-)

Good analysis, chief. By teh way, your Figure 1 from the NrK test is the most brutal example of complete lack of understanding of the concept of percent I have ever seen in my life.