Predicting CAT Percentiles: Why Coaching Centers Are Screwing It Up – ARKS Srinivas
Every year, right after CAT (and other exams), various Training centers post their analysis of the exam and make predictions of the scores & percentiles. As most CAT aspirants can tell you, predictions are usually way off.
In many cases training centers are racing to post their analysis and prediction the earliest without doing the necessary in depth analysis or perhaps without the intellectual capacity to do so.
|Percentile||Actual Score||VistaMind||TIME||Bulls Eye|
|99.5||194||190 (-4)||NA||180 (-14)|
|99||180||181 (+1)||166 (-14)||150 (-30)|
|97||156||162 (+6)||NA||130 (-26)|
|95||143||142 (-1)||133 (-10)||118 (-25)|
|90||124||123 (-1)||NA||100 (-24)|
In other cases, Training centers are giving up on prediction and making vague and useless statements like attempt 20Q with 90% accuracy for a good score. Lets see some data
Here is what some other National Level trainers are saying?
|70% accuracy for a 98%ile|
Really, that is all the data that they provide for a student who wishes to know where they stand. Ok. If I take 65 attempts and 70% accuracy (45-46 correct) and apply negative marking (say 12 of 19 errors have negative marks) then I get a score of 126 for 98%ile. As the table shows 126 marks get you close to 90%ile not 98.
CL does even better. Here is what they said on CAT day.
VA: 26 attempts at 80-85% accuracy
DI: 20 attempts at 85-90% accuracy
QA: 28-30 attempts at 85-90% accuracy
And if you do that, you get a GOOD Score. No Idea what CL means by good score. Your guess is as good as mine.
Why is prediction a very difficult science?
No access to Question paper
In case of a paper and pencil test, a student can get the question paper home and then test it out against a answer key made by experts. So we can get a score that is reasonably accurate. Of course actual accuracy of the score depends on accuracy of the key and keys posted by different experts can vary as can be seen in XAT 2016 exam. But in case of online exams, one cannot even access a question paper post exam, which makes score calculation guesswork.
Scaling of scores
Thankfully CAT this year was conducted only in two slots unlike multiple slots in the past. However, the Question paper for both the slots was different and hence scaling is used to bring all students onto one scale. From an analysis and prediction point of view this requires a very good analysis of relative difficulty level of the paper in each slot.
Score to Percentile conversion: Sectional and Overall
This is perhaps the toughest part. Scores in each section vary depending on number of Qs, time available, student allocation of time and relative difficulty. Even something like order of sections can affect the scoring patterns. To get a reasonable idea, one has to consider past data, factor in changes in exam pattern, relative difficulty level etc.
Ok, I get it. Prediction is difficult. So what? Why does prediction matter anyway?
Some students look at predictions just to get an idea of how they have done, for satisfaction. But for most student’s predictions matter, because they have to apply to Bschools. And a lot of good Bschools put in a application last date, which is before the results come out. And Bschool applications cost a lot of money. A student may end up spending a cool 10 to 15k on Bschools where he/she doesn’t have a chance (over prediction) of miss out on a good Bschool because a training center under predicted his/her score.
And surprisingly, many other institutes that claim to be so special and so successful have simply abdicated their responsibility by giving vague predictions like attempt 28-80 Qs with 85-90% accuracy to get a good score? Come on, you have 100s of trainers, tens of thousands of students (for feedback). Surely you can do better that.
How does VistaMind get it right?
No substitute for Experience. VM trainers have trained students for CAT for over 15 years with great success, having sent 1000s of students to the IIMs and top Bschools.
Write the bloody exam. Too many so-called experts are doing what we call as armchair analysis. It is easy to advise students to attempt X Qs at 90% accuracy. Unless you take the test, understand the difficulty level and time pressure, one can’t get it right. VistaMind faculty write the exam every year, to get a real feel of the exam and then do the analysis. Gut feel + Data.
Take the time, do it right. Too many Institutes and experts are in tearing hurry to post the analysis. Be the first to post online, try to get website hits, acquire student data and cross sell products. And in doing so, they make colossal errors: they get the key wrong, their predictions are way off. Please note, they end up demotivating students (Or giving them false hopes). Students who base their B school applications on that basis lose out badly.