Grade Inflation in CBSE 10+2 exams is an old story now. We know owing to excellent work of Prashant Bhattacharji and Debabrghya Das who “hacked”" into CBSE and ISC results archive for mining the data grade inflation or rather distortion in CBSE and ISC boards respectively. The fact that they hacked for this data which should have been open is another travesty ,though it ISC board seems to be taking lead in this regard.

In this blog my key objectives will be:

  1. Visualise distributions from years 2004 to 2015
  2. Visualise skewness from 2004 to 2015
  3. Understand What caused grade inflation in first place
  4. Examine if grade inflation was across all grades or across some grades alone(differential)

Data gathering

I had no access to raw data so i used these yearly histograms and digitized them to raw data . Once i had access to marks and respective percent histogram, i simulated the data back to raw number distribution in R.

so Let’s roll-

Lets visualise distribution of all years marks from 2004-2015.

So We have a got a good beautiful density plot instead of separately looking at histograms in original visualisation but let’s loook at 70 to 100 where the action is.

We can clearly see here that grade inflation/distortion in 90+ territory had started early with progressive higher percent of kids entering 90+ territory in progressive years but it was in 2012-13 that whole 90+ grade exploded and incidentally that was when Prashant Bhattacharji and Debabrghya Das figured this out.

One of the reasons given was the decision of IIT and NIT to include board marks as weightage, the policy turned out to be disastrous in retrospect and was scrapped in 2016.

But while visualisation is alright lets look at skew which is a statistically marker for asymmetry of marks curve which should be normal. A negative skew or left skew means , the peak is right (that is higher marks are awarded and there is grade inflation ) and tail is to left ahile positive skew is just opposite. All years distribution look to be negatively skewed as there is lesser amount of marks in less than 30 grade than above 70 marks

similarly kurtosis measures if the distribution has fat tails or tall peaks. A value greater than three suggest tall peak and fat tails though it is not absolute.

year skew kurtosis
2004 -0.33 2.90
2005 -0.48 3.46
2006 -0.34 2.77
2007 -0.31 2.90
2008 -0.20 2.88
2009 -0.25 2.90
2010 -0.35 3.15
2011 -0.17 2.60
2012 -0.22 2.71
2013 -0.31 2.80
2014 -0.41 2.91
2015 -0.40 2.71

Lets see if inflation was for one grade or across all grades

We see while numbers between 60 and 70 have fallen while number between 70-80,80-90,90-100 have all risen but 90-100 slope is highest and the curve accelerates from 2012

Lwt’s visualise fill diagram

Let’s plot animation of cumulative percent

But while it is good estimate lets visualise cumulation percent till 33+ grade instead of B2 only

We see while Pass percentage roughly stage same around 80-82% in each year 90+ grade goes high from 0.5% in 2004 to around 7.5% in 2015

Lets visualise it as odds with 2004 baseline

One of the reasons relative probability of 90+ goes high is because ot starts from very small base, but so does 33+. The above diagram shows grade inflation is substantially driven by 90+ grade to larger extent and 80+ 70+ to smaller extent

Lets’ plot animation of the same

As a bonus i plot here the correlation of surnames ( extracted from names by regex) in CBSE 2014 board results and UPSC exam from 2013-16. Data file is here .CBSE data was provided by Prashant Bhattacharji while UPSC data was compliled by me and other analysis on UPSC can be found here

Key Takeaways

  1. IITs inadvertently played a role in grade inflation
  2. Earlier there was random variation of skew since 2010,persistent negative skew due to grade inflation.
  3. Pass percentage stayed same despite grade inflation
  4. Grade inflation was driven by 90+(predominantly),80+ and 70+
  5. Grade Inflation has made board exams redundant
  6. Scientific techniques should be used for designing question and fitting the curve