Slashdot Mirror
OOXML Vote and the CPI Corruption Index
Tapani Tarvainen writes "It turns out there's an interesting correlation between Transparency International's 'corruption perceptions index' and voting behavior in ISO's OOXML decision. Countries with a lower score (more corruption) on the 2006 CPI were more likely to vote in favor of OOXML, and those with a higher score were less likely. According to the analysis, 'This statistics supports with a P value of 0.07328 the hypothesis that the corrupted countries were more likely to vote for approval (one-tailed Fisher's Exact test). In other words, simplified a bit: the likelihood that there was no positive correlation between the corruption level and probability of an approval vote, that is, this is just a random effect, is about 7%.' Of course, correlation doesn't prove causality."
Not only is 0.07 not significant, they used a 1-tailed test, rather than a 2-tailed test. If they had used the 2-tailed test, the p-value would have been 0.14, which is REALLY not significant. You're only ever justified in choosing the 1-tailed test over the 2-tailed one if you know for certain which way the influence is pushing. If, for example, one could make the case that the OOXML vote would have gone the other direction, with the more corrupt countries voting against it (a case we have no a priori reason to discard), then the use of a 1-tailed test is inappropriate here.
Actually, having read TFA, I'm pretty sure that correlation isn't appropriate at all here. The corruption scores are discrete, categorical values, rather than continuous values. This calls for nonparametric methods. Start with chi-square and move on from there. You can't do correlation with a straight face if your variables are discrete, since there's no guarantee that the "distance" in corruption between 2 and 3 is the same as the distance between 4 and 5.