Wednesday, December 07, 2005
[680] Of Pengkalan Pasir and Malaysian general election in 2008: hypothesis testing
[680] Of Pengkalan Pasir and Malaysian general election in 2008: hypothesis testing
So, Barisan Nasional won Pengkalan Pasir with a slim margin. A recount confirmed BN's victory. But does this mean PAS in deep shit? More precisely, if this by-election acts as a signal for the next Malaysian general election, does this mean support for Barisan Nasional's growing in Kelantan?
Quite hard to say. Let's see some statistics to see how hard it is.
First of all, we know from the New Straits Times that the first count showed that BN received 7419 votes, PAS 7290, Ibrahim Ali 414 and there were 160 spoiled votes. In total, 15283 votes.
Now, let's assume two things to make life simpler.
Kick Ibrahim Ali out of our equation, like how he was kicked from UMNO. And then, we shoved all the spoiled votes into Ibrahim Ali's throat before we kick him. That leaves us a total of 14709.
Thanks to the two assumptions, now we have only BN and PAS to worry about. The assumptions help making things easier since it allows us to use bootstrapping method - the easiest way to get a standard error. Bootstrapping gives us a standard deviation of roughly 0.5 with 60.64 for SE.
Now, my hypothesis is that the difference between BN and PAS - a difference of 129 votes - is zero. That makes t = 129/60.64, which makes t about 2.13. t is as in the t-distribution.
Using t-test, the hypothesis at 25% significance level is rejected.
However, at 10% and lower significance levels, the hypothesis is significant. Doesn't necessarily mean it must be accepted but it can't be rejected.
So, did anybody win outright statistically? I don't know. It's likely the answer is no. More importantly, if this is a signal of things to come in 2008, Pengkalan Pasir is a crystal ball with industrial defect.
Anyway, given this entry is written at 2 AM, you might be reading crap with bad statistics. Hah!
Heh. I myself am not convinced with the statistics; the hypothesis testing in particular looks odd. I've done this before extensively but I haven't touched real statistics in a long time. So, you are more than welcomed to check them up or even refined them to include Ibrahim Ali and the spoiled votes.
Quite hard to say. Let's see some statistics to see how hard it is.
First of all, we know from the New Straits Times that the first count showed that BN received 7419 votes, PAS 7290, Ibrahim Ali 414 and there were 160 spoiled votes. In total, 15283 votes.
Now, let's assume two things to make life simpler.
Kick Ibrahim Ali out of our equation, like how he was kicked from UMNO. And then, we shoved all the spoiled votes into Ibrahim Ali's throat before we kick him. That leaves us a total of 14709.
Thanks to the two assumptions, now we have only BN and PAS to worry about. The assumptions help making things easier since it allows us to use bootstrapping method - the easiest way to get a standard error. Bootstrapping gives us a standard deviation of roughly 0.5 with 60.64 for SE.
Now, my hypothesis is that the difference between BN and PAS - a difference of 129 votes - is zero. That makes t = 129/60.64, which makes t about 2.13. t is as in the t-distribution.
Using t-test, the hypothesis at 25% significance level is rejected.
However, at 10% and lower significance levels, the hypothesis is significant. Doesn't necessarily mean it must be accepted but it can't be rejected.
So, did anybody win outright statistically? I don't know. It's likely the answer is no. More importantly, if this is a signal of things to come in 2008, Pengkalan Pasir is a crystal ball with industrial defect.
Anyway, given this entry is written at 2 AM, you might be reading crap with bad statistics. Hah!
Heh. I myself am not convinced with the statistics; the hypothesis testing in particular looks odd. I've done this before extensively but I haven't touched real statistics in a long time. So, you are more than welcomed to check them up or even refined them to include Ibrahim Ali and the spoiled votes.
