Expert Explains | What is normalisation of scores, and why it works in JEE but might not in NEET
As NEET moves toward CBT, rankings may depend on statistical normalisation instead of directly verifiable raw marks. This could raise questions about fairness and transparency.
Examinees of NEET exam finding their roll numbers before entering an exam centre in Kolkata on May 3, 2025. Photo: Express The proposed shift of the National Eligibility-cum-Entrance Test (NEET) from a single-session pen-and-paper (PnP) examination to multi-session computer-based testing (CBT) is primarily driven by repeated controversies over paper leaks, OMR tampering, transportation vulnerabilities, and operational failures. CBT is expected to reduce such risks through encrypted digital delivery and the elimination of printed papers.
The NEET-PnP is conducted in a single session, with the same paper for all candidates. Students receive carbon copies of their optical mark recognition (OMR) sheets, allowing independent verification of raw marks directly used for ranking. The system is thus transparent, verifiable, and based on a common benchmark.
NEET-CBT fundamentally changes this structure. With India’s IT infrastructure reportedly accommodating only about 1.2 lakh candidates per session, conducting NEET for nearly 24 lakh students may require around 20 sessions with different question papers.
Since no two papers can be identical in difficulty, comparing raw marks directly is statistically unfair. Rankings, therefore, depend on normalised percentile scores derived through statistical methods rather than directly verifiable raw marks, raising concerns about precision, transparency, fairness, and verifiability, especially since no normalisation method can ever be perfectly error-free or universally accepted.
How does normalisation work?
The fundamental challenge in multi-session CBT is to fairly compare candidates who attempt different question papers across sessions. The underlying assumption is that candidates appearing in relatively tougher sessions are likely to score lower raw marks than those in easier ones.
To compensate for such variations, raw marks are statistically normalised so that candidates are neither unfairly advantaged nor disadvantaged by their allotted session. The premise is that if all sessions have sufficiently similar candidate populations and comparable difficulty distributions, normalised percentile scores should theoretically produce fair rankings.
Most likely, the National Testing Agency (NTA) will use the same normalisation methodology for NEET-CBT as in JEE (Main), though NEET would involve nearly double the candidate volume.
A candidate’s performance is assessed relative to other candidates in the same session using percentile scoring. A percentile score indicates the percentage of candidates scoring equal to or below a particular candidate. Thus, the topper of every session receives a percentile of 100 irrespective of raw marks, while percentile values at lower scores vary depending on the number of candidates appearing in that session.
In the first stage, percentile scores are calculated for each subject and in the aggregate within every session. These are then placed on a common statistical scale to generate normalised scores, called NTA scores — often reported to several decimal places to reduce ties — and final rankings across all sessions.
Can normalisation distort merit and ranking?
It is the central controversy in multi-session CBT. Since rankings depend on statistically normalised percentile scores, candidates cannot independently verify precisely how normalisation across sessions was computed.
In NEET, where even a single mark can change ranks by thousands, tiny normalisation variations may significantly alter outcomes. For example, a candidate scoring 710 in one session may rank below another scoring 680 in another session, potentially affecting admission to institutions such as the All India Institute of Medical Sciences (AIIMS).
The concern is not that normalisation is mathematically invalid, but that even small statistical adjustments can significantly alter rankings and admissions, creating distrust when normalised scores fall below actual raw marks.
Why will normalisation severely impact NEET, not JEE?
NTA already uses normalisation in the Joint Entrance Examination or JEE (Main), where anomalies are evident. Its impact, however, remains limited because about 2.5 lakh top candidates proceed to IIT-JEE (Advanced) for admission to 23 Indian Institutes of Technology and a few other institutions for a few tens of thousands of seats.
For them, JEE (Main) mainly serves as an eligibility screening test, since admissions are ultimately based on IIT-JEE (Advanced), a single-session examination conducted without normalisation. For most remaining JEE (Main) candidates as well, precision in normalised scores matters less because lakhs of engineering seats are available across 31 National Institutes of Technology, 26 Indian Institutes of Information Technology, and numerous other public and private institutions, with several thousand seats remaining vacant each year.
NEET is fundamentally different. Over 20 lakh candidates compete for a very limited number of medical seats, including a few hundred at top institutions such as AIIMS and the Jawaharlal Institute of Postgraduate Medical Education and Research. With extreme rank compression, even fractional percentile differences or minor normalization adjustments may significantly alter ranks and admissions, especially for meritorious candidates near critical cutoffs.
However, normalisation has a limited impact on management-quota admissions in private colleges, where NEET largely functions as an eligibility examination.
What about global digital exams?
Global digital examinations such as Scholastic Assessment Test (SAT), Test of English as a Foreign Language (TOEFL), and Graduate Record Examination (GRE) differ fundamentally from NEET and JEE. They are adaptive tests, where subsequent questions depend on responses to earlier ones, unlike NEET and JEE, where all candidates in a session receive the same question set.
Moreover, they are not single-criterion rank-based admission tests. They assess aptitude, language proficiency, or academic readiness as only one component of broader admission processes.
NEET and JEE, in contrast, are single-criterion rank-based admission tests in which rankings alone determine admissions for limited seats.
China’s Gaokao — perhaps the closest examination to NEET in scale and stakes — still relies largely on raw-score-based merit systems.
What is the way forward?
Normalisation in NEET-CBT is the biggest challenge, given the scale and stakes of the examination. Question-difficulty balancing, session-equivalence testing, disclosure mechanisms, simulation studies, and independent technical oversight would become essential. Both raw and normalised scores should be disclosed to improve transparency.
The system would therefore need to be transparent, verifiable, explainable, and statistically robust. Yet possible ranking distortions are likely to generate distrust and controversy.
The larger question is whether an examination in which even a single mark determines life-altering outcomes should shift away from directly verifiable raw-score merit toward statistically processed rankings. Before any transition, alternative models should be openly debated with genuine academic experts rather than only through high-powered committees.
The author is a former computer science professor at IIT Kharagpur, IIT Kanpur, BITS Pilani, and JNU, and a former scientist at the Defence Research and Development Organisation and the Department of Science & Technology.
Share
