While writing this article, I am prepared for the backlash that I might receive from the readers. The article is about my understanding of ‘Normalization’ and what can one do to counter the unnecessary stress that results because of the same.
In the time of paper-pencil test, when the same or a similar test paper (with reordering of the questions) was administered to all test takers it was relatively easier to compare students’ performance. With the introduction of an online exam, the test and the questions vary across different test days. Hence normalization is needed – scaling up/down of scores using psychometrics, statistics and test takers’ responses to a standard set of questions. This helps evaluate students on a common standard and is necessary because the difficulty level of the test varies across test days.
Why is normalization needed?
Let us consider a simple hypothetical scenario: I am attempting CAT this year along with a friend. Both of us are equally good in Quantitative Ability. But while my strength lies in Algebra and Geometry, his strength is Problem Solving (Time, speed, distance, work, profit and loss etc.) and Arithmetic. Since we gave the exams on different days our test papers were different. We both got considerable questions from our respective strong areas and performed equally well. But does that mean that we are equally good? No. Say on a scale of 1-5, I got questions of difficulty level 4 from Geometry. Being my strong area I did well. On the other hand he got questions of difficulty level 5 from Problem Solving. Now the only way to benchmark his and my performance is by looking at the relative scores of people who gave the test with us. In my slot out of 80/100 students were able to work out the questions from Geometry. On the other hand, in my friend’s slot only 50/100 students were able to work out questions from Problem Solving. That means when we take a large enough sample size, my friend’s performance is much better than average as compared to mine (even when we are scoring equally well).
Personal choices that bring in subjectivity – a student might not like geometry at all but might be strong in every other area in quant. This is countered by taking a large sample size.
This is a hypothetical scenario explaining normalization and its need in simplest possible language. The idea I am trying to put across is that comparing students across two different tests, requires psychometrics as well as statistics to get an idea of how people attempt a particular set of questions.
Deepak Nanwani is the co-founder of One52.com, an online adaptive solution for GMAT, MBA and UG exams. An alumnus of IIT Guwahati and IIM Bangalore, he is a master strategist for all competitive exams.