Blog entry by Orji Anyianuka

Picture of Orji Anyianuka
by Orji Anyianuka - Sunday, 21 March 2021, 11:00 PM
Anyone in the world

Many people don't trust the government and the general impression is that the government cooks up data.  This mistrust gets worse because the government often publishes data that do not reflect reality. 

Interestingly, it's very common for the government to present factual data, but the data itself is lying. Let's say the government comes up with data showing that a higher percentage of students in the federal government colleges passed JAMB than students in private schools, and then conclude that the government schools are therefore better than the private schools.

A possible conundrum is that everyone knows that the private schools have better teachers and that the teachers work harder, and therefore expect students in the private schools to score higher in Jamb than students in the government schools. However, if the total population of students enrolling into government schools are divided based on their IQ, the data may reveal something totally different and lead to a different conclusion.

What if data shows that a higher percentage of students who enrol into the government schools have higher IQ and that a higher percentage of students who get admitted into the private schools are students who are unable to pass the entrance exams to the government schools due to lower IQ? What if the data shows that seventy per cent of the low IQ students improved so much that they got higher JAMB scores than the students from the government schools? The government still wouldn’t be lying to say that the government schools did better than the private schools, rather it's the data that lied because of the way the data of the high IQ students and the low IQ students were lumped together.

An obvious takeaway from this phenomena called the Simpson's Paradox is that statistics can mislead when data of different groups are lumped together. And one way to take good care is to separate data from different groups and take a closer look.

[ Modified: Thursday, 10 June 2021, 10:55 AM ]