In #UPSC 2018, Muslims scored 13 marks LOWER than non-Muslims on average in written exams BUT scored 12 marks HIGHER in interview!!! This interview difference used to be 5-8 marks since 2013.

I ran t-test to see if this marks diff was by chance or by design. Statistics say that probability of this being by chance is 1 in 10000. In other words, it is far more likely that you get 10 continuous heads in coin tosses than Muslims scoring 12 marks higher than rest in #UPSC!

Consider this happening at least since 2013 (I couldnt get data before that). Sometimes its 4 marks more, sometimes 8, this time 12. I did a rough calculation and found that it is more likely that you get 50 consecutive heads than Muslims getting higher interview marks in #UPSC!

Just for cross-check, if anyone wants to verify, here are my results:
Ave Written score of M & NM = 793 and 806 respectively
Ave Interview score of M & NM = 181 and 169 respectively
Std deviation for above = 210 and 284
Clearly, data science says this is by design.

Whether one somehow find candidate details during written exams can be best answered by an insider. But in interviews, panel clearly knows who the candidate is. And interview is essentially a subjective assessment. Here Muslims excel every year. But lag in writtens of #UPSC!

I also ran an analysis of the claim that Hindus benefit by reservations and Muslims do not. What was interesting is that 19 out of 28 Muslims selected this year (68%) in #UPSC took benefits of reservations under OBC or ST. The cap for others is 50%.

Additionally, as per Minority Ministry schemes, coaching for #UPSC is completely free for Muslims. It is a different matter that typically minority is defined as 3-5% of population world over.

Conclusions are clear. Statistically, there are reasons to believe that selection in topmost service of nation is influenced by religion. This influence may be unbiased if you are ready to believe that a coin gets 50 heads but is still not biased!! #UPSC doesnt seem secular.

#UPSC 751 ranker (of 759 total) is a madrasa trained specialist in Dawaah or conversion. He studied maths science beyond class 3 for first time for Upsc preparation, and took Urdu as optional as well as medium of language. Has 11 siblings all of whom are Moulvis.

140 Ranker Ambul Samaiya (aided by Minority Ministry) got 804 in written and 206 in #UPSC Interview.
141 ranker Ajay Jain got 872 (68 more) but only 138 in Interview.
If selections were anonymous, Ambul would have ranked 410 and Ajay Jain 35.

As a Data Scientist, I can tell you that if anything happens that has lesser probability than 50 consecutive heads, and keeps happening at least 6 times without break, then either it is by design. or we have witnessed rarest miracle of history. Must be in Ripley Believe It Or Not

What also needs to be analysed is how many of these 759 were close relatives of bureaucrats and which ones. What were their scores. I leave that to government because they can rightfully probe that based on my analysis of significant statistical bias in UPSC scoring.

The highest score in Interviews for #UPSC 2018 was 206. 3 Muslims got this fantastic score. Ironically all the three had poor written scores.
Ayman Jamal (499) had 751 in writtens. R500 had 810
Rehana(187) had 797. R188 had 835
Ambul (140) had 804 .R141 had 872

I re ran analysis with new list of 30 Muslims who cracked #UPSC this year. 21 were OBC or ST. This means 75% took caste based reservation benefits compared to 50% cap overall. This means non Muslim OBC were at disadvantage - both in seats and interview marks.

Comparison of 2017 #UPSC with 2018 is interesting to Data Scientist.
In '18, avg = 181.2, stdev = 14.5 (28)
In '17, Avg M Ivw marks = 172, Std deviation =16 (51 guy)

Normally, deviation increases with more numbers for unbiased samples

Here it decreases.
Plus score rises.

If I do the same analysis for non-Muslim candidates,
In 18, avg = 169, sdev = 16.8
In 17, avg = 167, sdev = 16.7
Almost same deviation with just 2 more marks in 2018.

For Muslims, deviation reduced even though sample size reduced! And scores jumped. Statistically, it shows bias.

Also interesting is that this year I couldnt find #UPSC marks on their website. I had to get the data from internal sources. However one can RTI. I am also willing to share my detailed analysis with govt if they decide to probe an event with probability 1 in million

FAQ Times.
Q - But Muslim proportion is 18% of Indian population. How is there bias if only 28 out of 759 are selected?

A - This is a common sampling error. Why less or more people from a community are seen in any specific domain can depend on many things...cont...

..For ex, Kerala dominates nursing. Others dominate other good domains, even bad domains. Thats a diff research.
However what matters is that those carefully selected must be similar irrespective of religion, and selection must have no scope for bias. #UPSC data raises q here.

Q - Why you are worried that 70% Muslims selected are OBC? That means social justice.

A - 1. Some studies say OBC population is same in Hindu and Muslim. Hence this wide divergence comes to notice.
2. They narrative is that Hinduism is casteist, Islam is equality....

...this narrative gets busted. What emerges is - Muslim society has more caste issues than Hindu.
3. When u see this as cascading - getting more interview marks, less variation in these marks, OBC benefits, yr after yr - this makes is a Ripley Believe It or Not rare probability.

I also analysed scores of Muslim grads of Zakat Foundation to test the effectiveness of their special preparations. The finding is opposite. The average interview score of 17 Zakat fellows is ~175 compared to 188 for 13 non-Zakat Muslims.

I could not explain Zakat data. They carefully select candidates and give special training. Yet Zakat fellows dominate the lower 50 percentile of successful Muslims. Interview scores are significantly lower than rest though still higher than non-Muslims. Written score also lower.

I admit this is small dataset. Despite that the variation in Interview marks among Zakat grads is even lower. Not sure what to make out of it. It seems most Zakat fellows just barely make it to UPSC due to consistently better marks than non-Muslims.

I have been told recently that one candidate who I assumed to be Muslim based on data available on Zakat Foundation site was actually a non Muslim Jain. If we adjust for this one data point, yet story doesn't alter significantly..