CONFIDENTIAL

Normalization
1. Purpose: If test is conducted in batches with different set of questions, there is
possibility that the difficulty level may differ and in such an event to have
scores of both batches comparable, process Normalization is adopted.

2. Process: Mean and Standard Deviation is ascertained for the Base as well as
Targeted Batch. Formula is applied using these figures to the Scores of
Targeted Batch and Normalized score is obtained.

The base factors for the calculation are:

A) Proportion of Deviation
B) Difference between Target Value and Average Value
C) Average Value

and the formula used to get Normalized Score is A x B + C.

The elements comprising the above factors are modified to achieve precise
results based on of data (Scores of candidates in different Batches) resulting in
different methods. These methods are explained in Annexure “A”.

3. Assumptions / Pre-requisite: Basic assumption in this process is that scores in

both Base Batch and Targeted Batch have Normal Distribution.

The main disadvantage of Normalized scores is that they always assume a

normal distribution. But if this assumption is not met, the scores cannot be
interpreted as a standard proportion of the distribution from which they were
calculated.

It is said that you need to take at least 30 samples, to be "sure" that you have
an exact enough mean and deviation estimates.

It is interesting to note here that, even if a sample is taken from one batch, the
size needs to be adequate enough, and then only it can represent approximate
distribution of that batch. (Ref : Central Limit Theorem)

The choice of n = 30 for a boundary between small and large samples is a rule
of thumb, only. There is a large number of books that quote (around) this value,

October 2020 Page 1

 CONFIDENTIAL

for example, Hogg and Tanis' Probability and Statistical Inference (7e) says
"greater than 25 or 30".

A. Considering above said Assumptions / Pre-requisite Normalization can not

be done under following circumstance and the score needs to be left as it is
for ranking purpose.
• In case the size of Base Batch or Target Batch is less than 30
• In case the test Question Papers are not comparable (i.e. with different
subject matter content, different pattern / level)

B. The distribution of scores is normal. It may be skewed towards right or left

to some extent depending upon the overall performance of candidates.

C. Even if Average Score of 2 Batches is same the distribution of Scores may

be different. This will reflect the comparative difficulty level of questions
administered for different Batches.

October 2020 Page 2

 CONFIDENTIAL

D. In the following diagram the B (data plotted in dotted line) represents more
difficult Questions used in the Batch as compared to A and C

E. Annexure “A” enumerates different methods for normalizing scores using

different formulas. There is no hard and fast rule for the purpose of selecting
the method to be applied, however considering the formulas used, inferences
could be drawn as to the circumstances in which they are applied.

4. Conclusion: Considering what has been stated above and also various methods
mentioned in annexure “A”, In our opinion :
a) Normalization cannot be done under following circumstance and the score
needs to be left as it is for ranking purpose.
• In case the size of Base Batch or Target Batch is less than 30.
• In case the test Question Papers are not comparable (i.e. with different
subject matter content, different pattern / level)
b) As regards Method B and C :
• There is a factor “Top 0.1% candidates. If the absolute value of the
same is expected to be at least 30, then the batch size needs to be

October 2020 Page 3

 CONFIDENTIAL

minimum 30,000. In view of this these methods are not recommended

where batch size is less than 30,000.
• Where the size of batches does not differ significantly, concept of
Base Batch can be adopted and therefore Method B will be suitable.
• In other case Method C will be suitable

c) Method A or D can be used in all other cases however Method A may be

preferred where the examination is related with specific domain knowledge
whereas Method D would be suitable for General Knowledge. It is desirable
that the size of Base Batch and Target Batch is not less than 300.

October 2020 Page 4

 CONFIDENTIAL

Annexure A
Method A
- Score Normalization using Mean and Standard Deviation of Base / Standard
and Target Batch

Xn = (S2/S1)*(X-Xav) + Yav

Where:
Suffix 1 and suffix 2 represent two sets of marks.
S represent standard deviation.
X and Xav represent raw score and average score for set 1.
Y and Yav represent raw score and average score for set 2.
And Xn = Normalized score.
Supposing set 1 is to be scaled against set 2 (which is declared as standard)

In this method there are following 3 factors:

A) Ratio of SD of Base / Standard Batch to SD of Targeted Batch
B) Difference of Score of Candidates and Average Score of the Target Batch
C) Average Score of Base / Standard Batch
Normalized Score = A x B + C

Batch with maximum average with minimum 70% of the overall average
attendance is considered as the Base / Standard Batch.

October 2020 Page 5

 CONFIDENTIAL

Method B
̅𝑡𝑔 − 𝑀𝑞𝑔
𝑀
̂𝑖𝑗 = 𝑔𝑚
𝑀 𝑥(𝑀𝑖𝑗 − 𝑀𝑖𝑞 ) + 𝑀𝑞
̅𝑡𝑖 − 𝑀𝑖𝑞
𝑀
Where,
̂𝑖𝑗 = 𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 𝑚𝑎𝑟𝑘𝑠 𝑜𝑓 𝑗𝑡ℎ 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒 𝑖𝑛 𝑖 𝑡ℎ 𝑠ℎ𝑖𝑓𝑡 (𝑢𝑝 𝑡𝑜 5 𝑑𝑒𝑐𝑖𝑚𝑎𝑙 𝑝𝑙𝑎𝑐𝑒𝑠)
𝑀
𝑀̅𝑡𝑔 = 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑚𝑎𝑟𝑘𝑠 𝑜𝑓 𝑡𝑜𝑝 0.1% 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠 𝑐𝑜𝑛𝑠𝑖𝑑𝑒𝑟𝑖𝑛𝑔 𝑎𝑙𝑙 𝑠ℎ𝑖𝑓𝑡𝑠
𝑔
𝑀𝑞 = 𝑆𝑢𝑚 𝑜𝑓 𝑚𝑒𝑎𝑛 𝑎𝑛𝑑 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 𝑚𝑎𝑟𝑘𝑠 𝑜𝑓 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠
𝑖𝑛 𝑡ℎ𝑒 𝑒𝑥𝑎𝑚 𝑐𝑜𝑛𝑠𝑖𝑑𝑒𝑟𝑖𝑛𝑔 𝑎𝑙𝑙 𝑏𝑎𝑡𝑐ℎ𝑒𝑠
̅𝑡𝑖 = 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑚𝑎𝑟𝑘𝑠 𝑜𝑓 𝑡𝑜𝑝 0.1% 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠 𝑖𝑛 𝑖𝑡ℎ 𝑠ℎ𝑖𝑓𝑡
𝑀
𝑀𝑖𝑞 = 𝑆𝑢𝑚 𝑜𝑓 𝑚𝑒𝑎𝑛 𝑎𝑛𝑑 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑖𝑡ℎ 𝑠ℎ𝑖𝑓𝑡
𝑀𝑖𝑗 = 𝑅𝑎𝑤 / 𝑆𝑐𝑎𝑙𝑒𝑑 𝑀𝑎𝑟𝑘𝑠 𝑜𝑏𝑡𝑎𝑖𝑛𝑒𝑑 𝑏𝑦 𝑗𝑡ℎ 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒 𝑖𝑛 𝑖 𝑡ℎ 𝑠ℎ𝑖𝑓𝑡
𝑔𝑚
𝑀𝑞 = 𝑆𝑢𝑚 𝑜𝑓 𝑚𝑒𝑎𝑛 𝑚𝑎𝑟𝑘𝑠 𝑜𝑓 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠 (𝑖𝑛 𝑡ℎ𝑒 𝑠ℎ𝑖𝑓𝑡 ℎ𝑎𝑣𝑖𝑛𝑔 𝑚𝑎𝑥. 𝑚𝑒𝑎𝑛)
𝑎𝑛𝑑 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠 𝑐𝑜𝑛𝑠𝑖𝑑𝑒𝑟𝑖𝑛𝑔 𝑎𝑙𝑙 𝑏𝑎𝑡𝑐ℎ𝑒𝑠
______________________________________________________________________________

1. In this case additional elements like Average and Standard Deviation of top
0.1% of overall candidates as well as that of targeted batch are brought in the
picture.
2. By using Proportion of difference ( Average score of Top 0.1% candidates
minus Average+ SD of all candidates’ score) for all shifts to targeted shift the
purpose to normalize the data more precisely is achieved if there is significant
variation in marks scored by top 0.1% candidates in different batches.
3. Concept of Base / Standard and Target Batch is maintained.
4. Since ratio as indicated in 2 is used as one of the factor for the purpose of
normalization of candidate’s score, instead of difference between score of the
candidate and average score of the Base Batch, difference between score of the
candidate and average + SD of score of the Base Batch is taken.
5. Having taken proportionate difference as stated above Average + SD of score
for all batches is added to it. Here SD is also added because while calculating
proportionate difference SD is also deducted from candidate’s score.

October 2020 Page 6

 CONFIDENTIAL

Method C

̅𝑡𝑔 − 𝑀𝑞𝑔
𝑀
̂𝑖𝑗 = 𝑔
𝑀 𝑥(𝑀𝑖𝑗 − 𝑀𝑖𝑞 ) + 𝑀𝑞
̅𝑡𝑖 − 𝑀𝑖𝑞
𝑀
Where,
̂𝑖𝑗 = 𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 𝑚𝑎𝑟𝑘𝑠 𝑜𝑓 𝑗𝑡ℎ 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒 𝑖𝑛 𝑖𝑡ℎ 𝑠ℎ𝑖𝑓𝑡 (𝑢𝑝 𝑡𝑜 5 𝑑𝑒𝑐𝑖𝑚𝑎𝑙 𝑝𝑙𝑎𝑐𝑒𝑠)
𝑀
𝑀̅𝑡𝑔 = 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑚𝑎𝑟𝑘𝑠 𝑜𝑓 𝑡𝑜𝑝 0.1% 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠 𝑐𝑜𝑛𝑠𝑖𝑑𝑒𝑟𝑖𝑛𝑔 𝑎𝑙𝑙 𝑠ℎ𝑖𝑓𝑡𝑠
𝑔
𝑀𝑞 = 𝑆𝑢𝑚 𝑜𝑓 𝑚𝑒𝑎𝑛 𝑎𝑛𝑑 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 𝑚𝑎𝑟𝑘𝑠 𝑜𝑓 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠
𝑖𝑛 𝑡ℎ𝑒 𝑒𝑥𝑎𝑚 𝑐𝑜𝑛𝑠𝑖𝑑𝑒𝑟𝑖𝑛𝑔 𝑎𝑙𝑙 𝑏𝑎𝑡𝑐ℎ𝑒𝑠
̅𝑡𝑖 = 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑚𝑎𝑟𝑘𝑠 𝑜𝑓 𝑡𝑜𝑝 0.1% 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒𝑠 𝑖𝑛 𝑖𝑡ℎ 𝑠ℎ𝑖𝑓𝑡
𝑀
𝑀𝑖𝑞 = 𝑆𝑢𝑚 𝑜𝑓 𝑚𝑒𝑎𝑛 𝑎𝑛𝑑 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑖𝑡ℎ 𝑠ℎ𝑖𝑓𝑡
𝑀𝑖𝑗 = 𝑅𝑎𝑤 / 𝑆𝑐𝑎𝑙𝑒𝑑 𝑀𝑎𝑟𝑘𝑠 𝑜𝑏𝑡𝑎𝑖𝑛𝑒𝑑 𝑏𝑦 𝑗𝑡ℎ 𝑐𝑎𝑛𝑑𝑖𝑑𝑎𝑡𝑒 𝑖𝑛 𝑖 𝑡ℎ 𝑠ℎ𝑖𝑓𝑡

____________________________________________________________________________________

This method is similar to Method B except change in “Factor C” i.e. instead of

“Sum of mean marks of candidates (in the shift having max. mean) and Standard
Deviation of candidates considering all batches” , “Sum of mean and Standard
Deviation marks of candidates in the exam considering all batches” is considered.

Therefore, this method is appropriate where all Batches are equally important or
unique and no Batch can be taken as a Base / Standard Batch.

October 2020 Page 7

 CONFIDENTIAL

Method D - Using Equi-Percentile

Total Candidates in Shift - Rank of the candidate in that shift

Percentile Rank of Candidate = ------------------------------------------------------------------------------- X 100
(Total candidates in the shift -1)

In this method Batch-wise Percentile Rank of Candidate is calculated as per the

above formula and thereafter All data is merged for further process of short-listing.
Once the percentiles are calculated, the batch having the most number of
candidates appeared is considered as Base Batch.
If Equated Score is required to be displayed, instead of Percentile Rank, before
merging data of all batches the same is calculated using the following formula.
However this does change Merit Order as prepared based on Percentile Rank of
Candidate.

𝑌2 − 𝑌1
𝑌 = 𝑌1 + 𝑥(𝑋 − 𝑋1)
(𝑋2 − 𝑋1)

Where:
Y = Equated Score rounded up to 2 decimal places
Y1 = Marks corresponding to immediate lower percentile form Batch II
Y2 = Marks corresponding to immediate upper percentile form Batch II
X1 = Immediate lower percentile form Batch II
X2 = Immediate upper percentile form Batch II
X = Percentile of the Candidate of the respective Batch

=======

October 2020 Page 8