Mains 2019 Preparation and Discussion Thread

@John_TitorExactly! Thanks for corroborating my stand. (Finally. This view is usually not appreciated by the fraternity I've interacted with and is met with- You mean UPSC is wrong and you're right?)

Anyway. It's high time UPSC changed this method. I mean how difficult would it be to prepare a new empirical model based on statistics from post-revision period (one optional).

I have no knowledge of statistics, but still I think it could be done. Weighted coefficients could be assigned in accordance with the strength for different optionals. The sample size does not vary abruptly for any optional. It is more or less predictable. Like, we won't see 1000 people taking up Kannada literature ever!

I don't understand why there has been no action in this direction. Where are the statisticians of our country?

@sat_009
May be u r correct. But official stand of upsc is that scaling hoti hai. In RTI also they will admit about scaling. So we can only guess.

@AzadHindFauzsaid inMains 2019 Preparation and Discussion Thread:

@RagsIf only you can decipher the hebrew written in this, you'd be able to know-

https://indiankanoon.org/docfragment/1530075/?big=0&formInput= scaling of marks

Assuming 1000 candidates per paper is audacious. And so the marks obtained by candidates are the only way to decide on the difficulty of the optional paper after all. This because they don't account for the degree of subjectivity that differs across optionals. For some literature optionals there are only 2-3 candidates appearing. Highest doesn't even reach 300 but the lowest is around 180 for them.

@John_TitorI agree that the assumption that 1000 people take each optional is wrong. Any distribution can be approximated as normal distribution if we have atleast 30 data points. That's a well established principle in statistics. The more data points we have, the better normal approximation we get. What I don't understand is, how they'll apply the principle to subjects that see less than 30 people? This would do terrible injustice to optionals with very low turnout.
Another objection I have is to compare performances across optionals. Sometimes some subjects are given extremely hard like 2015 geography or 2017 maths. People struggled hard to even understand some questions. In such a case how can the performance of top 5 percentile assumption be made?

@vamgaducan you explain me what with this 5 percentile thing

@vamgaduExactly! Treating marks obtained as the only viable method to judge the difficulty of a subject in upsc is plain injustice.

@sat_009"It is fair to assume that the mental ability (and consequent performance) of candidates in all optional subjects are about the same at very score range. We can assume that top 5% of say History candidates are comparable in ability to the top 5% of say Geography candidates." This is what's written in the link pasted above. So, what they are saying is we'll try to do justice to all the optional subjects by comparing the performance of those how scored more than 95% of the candidates taking that paper.

Let us take two different scenarios. You want to judge how well 100 monkeys climb trees against how well 100 fish swim in water. Now, the method is to find out the time taken by the top 5 monkeys to climb tree and find out time taken by 5 fish to swim. Then you assume that these two should be equal. In order to make them equal, you adjust the scores in one set of data to match the other set keeping in mind that the top 5 should be roughly the same.
what i object in this method are,

1. which data are you adjusting? is it the monkey group data or the fish group data? how do you decide which data to be taken as baseline from among 40 odd optionals?
2. How can you say that the ability of fish to swim in water is comparable to the ability of monkey to climb a tree?
3. Even after assuming you can compare their abilities, what about the difficulty of the task? how will you decide the height the monkey has to climb as against the fish swimming say 100m in water? similarly, how can you compare the performances of candidates in a very tough (say) math paper with the performance of an easy (say) pol.sci paper or vice versa.
4. When the scores are being adjusted statistically, doesn't it offer discretion to UPSC to fix the scaling factors? Does it not take away the objectivity of the paper? A small difference in the scaling factor may change your optional score by as low as 1-2 marks if you're in the middle of the bell curve to about 10-15 marks or more if you fall on either extreme. this feels scary to me.
   I generally don't post on forum these days, just follow it. But i thought i should share this. What i said is true to the best of my knowledge, if I made any mistake that may be because of my incomplete understanding. Feel free to enlighten.

I think there is no normalisation between subjects, but only between different examiner of the same subject taking top 5 marks given by each of them.

@foxmulderThis case primarily was related to prelims. For mains, they might be using certain corrections to normal distribution for optionals in which very few people appear.

@abhinav23raj436Okay. But the explanation they gave is for normalizing marks across different optionals.

And, that's why the talk of scrapping the optional paper comes. UPSC would do more good than harm by introducing a common paper like public policy or something which is actually meaningful for administration and doing away with 40 different kind of optionals.

The act of normalization fails on moral grounds when done with super-simplistic assumptions.

Grabs popcorn

@AzadHindFauzYes. There should be some common papers instead of optionals

@Dreadful

Thanks to you all for liking this comment

@vamgaduwhat is ur optional

@Dreadful

Please ignore it.

My friend has mischievously posted this without my knowledge. Just he revealed it.

My apologies.

Moral of the story is one needs to be in top 5% in their optional to expect a rank.

Bhai yaha koi Chemsitry optional se hain kya ? Kaisa hua paper ?

@vamgaduattempt is made to equate papers, and common scores are arrived at. For instance,if in an easy maths paper top five percentile score between 380-400, and in difficult history paper top five percentile score between 280-300, the equal percentile groups between optionals are equated. In the above example, the scores of math candidate is lowered, where as the scores of history candidates is upscaled. So that the top five percentile in both history and math are in the same range. That is why, generally, toppers in each subjects have their scores in the range 300-330(barring a few aberrations), irrespective of difficulty level of the paper. The final scores in optional are not your actual scores but scaled ones.

@Deltavery arbitary.

@RaufNaScaling is done for both optional papers separately or together, i have seen people with very low in one and very high in amother how is that possible if equalisation is on total of two