A Bimodal Quiz in a Mixed ITI Batch: What the Mean Is Hiding
An instructor at a Rajasthan government ITI teaches the first unit of a computer fundamentals course - number systems, binary arithmetic and 1's/2's complement - to 48 trainees in a 50-minute daily period. The batch is mixed: 12 trainees have never used a computer before, 27 have school-level exposure, and 9 are graduates who already use spreadsheets. The instructor has taught the unit by lecture with board work for two weeks. A 20-mark formative quiz (10 one-mark recall items, 6 two-mark conversion problems, 4 one-mark true/false) gives a class mean of 9.8 out of 20; 31 of 48 score below 10. Item analysis shows 44 of 48 answered the recall items correctly, while only 9 solved the two's-complement subtraction item. Attendance is 90% and there are no discipline problems. The summative exam will be 60% MCQ and 40% practical on a machine. Two trainees complain the pace is too fast; seven say it is too slow.
1. Diagnose what the quiz data actually says about learning, using item difficulty and Bloom's levels, and explain why 'the class average is low' is the wrong conclusion.
Read the worked answer
The mean of 9.8 (49%) is a misleading summary because the distribution is bimodal and the paper samples two very different levels.
Item difficulty (p = proportion correct): the ten recall items have p ≈ 0.90 - 44 of 48 got them - which means they are easy and discriminate almost nothing. They test Remember: naming bases, writing a place-value table. The conversion items sit much lower, and the two's-complement subtraction item has p ≈ 0.19, near the floor. Only 9 of 48 could execute the borrow-and-invert procedure, which is Apply level.
So the class has mastered recall and has not mastered procedure. The average of a 0.90 block and a 0.19 block describes no actual trainee. The 12 first-time users are probably near the floor while the 9 graduates are carrying the recall block; averaging them hides both.
The four true/false items are also weak evidence: with 50% by guessing, a correct answer carries little information, and the expected score from guessing alone is 2 of 4.
Correct conclusion: instruction stayed at the Remember level for two weeks while the quiz - and presumably the summative paper - demanded Apply. The problem is the alignment, not the trainees' ability. The two 'too fast' and seven 'too slow' complaints are the same finding from the other side: one pace for a 12-to-48 spread.
2. Design a one-week remediation and differentiation plan inside the same 50-minute periods. Name the groupings, the activities, and the evidence you will accept that remediation worked.
Read the worked answer
Keep the syllabus identical; change the time and the mode per tier. Standard 50-minute period: 8 min whole-class mini-teach, 27 min tiered practice, 10 min whole-class check, 5 min exit ticket.
Groupings, based on the quiz item pattern rather than on the total score:
- Tier 1 (12 beginners, scored on recall only, missed conversion): instructor-led guided practice. Concrete-to-abstract sequence - an on-screen 8-bit counter/odometer, then a place-value grid, then worked examples with the scaffolding faded (I do, we do, you do). Maximum 6 per table so you can see every hand.
- Tier 2 (the 27 with partial conversion marks): paired problem sets with a checking routine - every answer verified by converting back, or by
bin()in Python. - Tier 3 (the 9 graduates): extension tasks (signed ranges, overflow detection, BCD, 1's vs 2's complement edge cases) plus a briefed role as peer tutors for 15 minutes a day. Brief them, or they will simply do the work for the beginners.
Evidence I will accept: day-by-day exit tickets (3 items, 2 conversion), and a parallel re-quiz on Friday with fresh items of the same blueprint, on which at least 80% of trainees solve 2 of 3 conversion items and no item has p below 0.3. If the two's-complement item stays near 0.2, the remediation failed regardless of how good the classes felt.
3. The summative exam is 60% MCQ and 40% practical. Show how you would align teaching and formative assessment to that, and name one specific change to the quiz itself.
Read the worked answer
Alignment is constructive: the intended outcome is 'converts and performs 2's-complement arithmetic correctly on a machine', so both the practice and the checking must look like the summative paper.
Teaching: every period includes one timed 10-item MCQ set in the summative format (so format familiarity is not a hidden variable), and one machine task per week - enter a negative decimal, convert with a calculator or bin(), subtract two 8-bit numbers, verify by adding back. Show the practical rubric in advance: correct method, correct answer, verification step, speed.
Formative: keep the weekly quiz but mirror the blueprint - 60% objective, 40% short practical/worked items with a mark for the check.
Specific change to the quiz: delete the four true/false items. They are guess-prone (expected 2 of 4 by chance) and inflate the mean without measuring anything. Replace them with two three-mark items that require showing the working and a verification step, so the paper distinguishes a trainee who guessed a conversion from one who can perform and check it. Then pre-test the new items on a small sample to confirm p falls in the 0.4–0.7 band.