Slashdot Mirror
Hacker Exposes Evidence of Widespread Grade Tampering In India
Okian Warrior writes "Hackaday has a fascinating story about Indian college student Debarghya Das: 'The ISC national examination, taken by 65,000 12th graders in India, is vitally important for each student's future: a few points determines which university will accept you and which will reject you. One of [Debraghya]'s friends asked if it was possible to see ISC grades before they were posted. [Debraghya] was able to download the exam records of nearly every student that took the test. Looking at the data, he also found evidence these grades were changed on a massive scale."
Good thing he's living in the US then.
He tried to kill me with a forklift!
not even hacking just URL typing with fixed ID numbers
The test results were manipulated. There are missing scores (from 1-100) on a test taken by 150,000 students. That is not possible. They have been bumped up to passing. The graphs show jagged peaks separated by gaps rather than a curve. Unless his data is incomplete or has been manipulate, there is no reasonable explanation for the jagged charts.
It definitely does not represent standardization to a score of 100. It's not an even distribution of peaks. It is pushed up above the failing mark, and there is no gap from 94-100. Furthermore, all the different tests in different subjects show the same gaps. This is not reasonable at all.
What does "ls -l" do? Please describe below.
That kind of thing. So, I'm not surprised if institutions are manipulating test scores. India is more about the perception of computer savvy developers than the reality of it.
No sigs in BETA. Beta SUCKS.
The Indian system of education doesn't work like that. Here's a post I made on another forum: You can theoretically attain all marks in the 0-100 range because there is no scaling up. Each paper has components that together total upto a 100. For example, there could be 10 1-mark questions, 15 2-mark questions, 4 3-mark questions, 3 4-mark questions and 6 6-mark questions. Each question can be graded to a fraction of it's worth. So you can get 1.5 on a 2-mark question, 0.5 on a 3-mark question, etc. Thus theoretically, all possible combinations of scores are possible. The absence of certain scores is evidence of tampering. SOURCE: I appeared for the CBSE exams last year. The system is similar, though not the same.
Was I the only one reading the article thinking, "Finally, a developer from India that can think deeply about a problem without being told what to do, and then write software that works..."?
Peter predicted that you would "deliberately forget" creation 2000 years ago...
1. Teachers have to ensure that their class marks have a certain average and median before they submit them. There can't be too many failures either.
2. Teachers know not to give a grade of 49 if the pass is 50 since the student will argue to get that missing point. If you want to be safer, just don't give out anything in the forties.
3. If a test gives letter grades, that equates to a particular number. A = 85, A- = 83, and so on. In that case, no one gets an 84, ever.
"We are here on Earth to fart around. Don't let anybody tell you any different!" -- Kurt Vonnegut
i believe that was a joke, aimed at the 'indian english'. just sayin' :)
Rich
You should read the original http://deedy.quora.com/Hacking-into-the-Indian-Education-System. The missing scores are in extremely suspicious positions. For example, there are no scores of 32,33 and 34, and the minimum pass grade if 35. That looks pretty close to a bump to get people to pass. This doesn't look like someone not understanding the grading system. It looks like manipulation. Frankly, speaking as someone who does a fair bit of grading, yes one can get weird distributions from legitimate adjustments, but they don't look like this.
Correct; typical example could come from counting, then plotting, discrete data. Number of children in a family, doors on a car...
Note that whilst you might expect a normal distribution, with events (exam results) distributed evenly but randomly about the mean, the fact the the guy found something that certainly looks non-normal, (he did not do normality tests, but having looked at his results, I don't think he needed to), does not itself prove that the results were altered.
Imagine a 'perfect' exam, where the expected (average) result for the student population was 50 out of 100, or 50%
Now imagine an (equally unlikely) 'perfect' candidate population.
If you plotted the exam results, you could expect the population to be centered on a mean result of 50, with half the scores higher, half lower.
If you had a (really getting unlikely now) 'perfect' education system, there would be a low standard deviation in your data, let's say 2%
If the results could be modelled with the Gauss curve, then 99.73% of your distribution would be at +/- 3 sigma (standard deviations) from the mean.
So lowest expected score of 50-2*3=46, with highest of 56.
Of course, candidate abilities could be much more varied than this, so sigma could be anything...5%, 10%
Anyway, getting to the point, if the mean of a what you *might* be expecting to be a Gauss / Normal curve is shifted sufficiently towards a 'hard' limit, (in our example, you cannot score less than 0%, or more than 100%, so both are 'hard' limits, or 'boundaries'), then the data (example results) do tend naturally to 'pile up' against the limit. (Think of a snow plough pushing snow aganist a wall - it's go nowhere to go, except up).
Thus you get a non-normal distribution, (typically better modelled with a lognormal or Weibull curve, not Poisson).
But WHAT can cause the mean to shift? For this example:
- Either the exam is "too easy", or
- The students are all very good (yeah, same thing,really), or
- The marking system is biased.
I'll leave you to draw your own conclusions on that one, but I've personally found that in India, (as in other places, including the USA), a little cash can go a long way...
But that was not the most compelling evidence of bias; that would be the very strange 'missing' data points, (especially close to critical scores such as the 35 pass. /endoldstatsbore