An advanced vocabulary comparison casebook: comparative adjectives and adverbs, more, less, fewer, better, worse, higher, lower, faster, slower, improvement, decline, baseline, benchmark, reference point, percentage change and the question “compared with what?”

Maya reads a headline: Students improved by 20%.

She smiles. “That sounds large.”

Hana asks the question that changes the sentence:

“Twenty per cent compared with what?”

Jia Jun tries another line:

The second class performed better.

Ethan looks up from the table.

“Better than which class? On which measure? Over what period? Starting from which score?”

The vocabulary of comparison is deceptively simple.

Cambridge Grammar teaches that comparative adjectives and adverbs compare one person, thing, state or action with another, commonly through forms such as bigger, more useful, less complicated, better, worse, faster and constructions with than. It also shows that nouns can be compared through more, less and fewer.

That grammar is necessary.

It is not sufficient for advanced reading.

Because every comparison hides a structure:

target × measure × reference point × time window × population × scale.

Change the reference point and the same observation can look:

• better;
• worse;
• larger;
• smaller;
• faster;
• slower;
• improved;
• unchanged;
• dramatic;
• minor.

NIST measurement guidance makes the same idea concrete in technical work: comparison requires a reference base or accepted reference value. A difference cannot be interpreted properly without knowing the value against which it is being judged. NIST’s percentage-difference documentation likewise defines percentage difference relative to an explicit reference value.

This article owns one narrow applied job in the eduKatePunggol advanced vocabulary chain: how learners detect, reconstruct and test the hidden baseline behind comparative language so that “more”, “less”, “better”, “worse”, “higher”, “lower” and “improved” do not float without a reference point.

It is distinct from Advanced Vocabulary Causality | The Cause–Mechanism Audit, which asks whether one thing produced another; Advanced Vocabulary Sequence | The Timeline Audit, which owns order and state change; and Advanced Vocabulary Argument | The Qualification Audit, which owns claim-strength calibration.

The Baseline Audit asks one question first:

Compared with what?

Then it asks five more:

1. What is being compared?
2. On which measure?
3. Against which baseline or benchmark?
4. Across which time period?
5. In which population or group?
6. On what scale—absolute difference, percentage change, percentage points, rank, rate or ratio?

The teaching examples below are constructed. The external sources support the general grammar and measurement principles; the Baseline Audit is an original classroom application.

The forty-five-second route: never accept a comparative without rebuilding its missing half

When you read a comparative claim, do this:

1. Circle the comparative word. More, less, fewer, better, worse, higher, lower, faster, slower, stronger, weaker, improved, reduced, increased, decreased.
2. Write the target. What changed or differs?
3. Write the baseline. Compared with what?
4. Write the measure. Score, time, error count, percentage, rate, rank, cost, frequency?
5. Write the window. Compared with when?
6. Check the scale. Absolute points or relative percentage?
7. Check the group. Same learner? Same class? Different cohort? Different difficulty?
8. Test whether a different reasonable baseline would reverse or shrink the claim.

A sentence such as The new method is better is not wrong because it is short.

It is incomplete because better asks the reader to supply an absent comparison.

Reader routes: Comparative grammar · Hidden baseline · Improved and reduced · Percentage change · Benchmarks · Reference classes · AI baseline drift · Case files · Practice · Sources.

Laboratory A: comparative grammar tells us that a comparison exists; reasoning tells us what the comparison means

Cambridge Grammar describes comparative adjectives as forms that compare one person or thing with another and allow us to say that one has more or less of a quality. British Council examples use forms such as better, bigger, more expensive and happier, often with than when the second term is stated.

The reasoning problem begins when the second term disappears.

Case 1. Better than what?

This revision method is better.

Possible baselines:

• better than doing nothing;
• better than rereading;
• better than the learner’s previous method;
• better than another method on delayed recall;
• better for speed but not accuracy;
• better for one learner but not another.

The adjective better carries no fixed meaning without a dimension and comparator.

A high-level reader mentally expands it:

Better [than X] [on measure Y] [for population Z] [over time window T].

Case 2. Higher can be good, bad or neutral

The second class had a higher average.

Higher average what?

• score?
• error rate?
• absenteeism?
• reading speed?
• time spent on homework?

Higher is directional, not evaluative.

A higher score may be desirable.

A higher error rate may be undesirable.

A higher homework time may indicate diligence, inefficiency or harder work.

The reader must not smuggle “better” into “higher”.

Case 3. Lower can hide whether the denominator changed

The class had a lower error rate this week.

Suppose:

• Week 1: 8 errors in 80 attempted items;
• Week 2: 6 errors in 40 attempted items.

The raw number of errors is lower.

The error rate is higher.

This is a central lesson:

counts and rates can move in opposite directions.

Case 4. More and fewer require us to identify what is being counted

Cambridge Grammar distinguishes more, less and fewer in noun comparisons. The traditional pattern uses fewer with plural countable nouns and less with uncountable nouns.

Advanced reasoning adds another question:

More what?

Fewer errors can mean:

• fewer total errors;
• fewer errors per question;
• fewer serious errors;
• fewer repeated errors;
• fewer errors on easier questions only.

The noun after the comparative often needs operational definition.

Case 5. Faster is incomplete without accuracy

Ethan completed the paper faster.

If he also made twice as many mistakes, faster may not be an improvement.

Many educational comparisons involve trade-offs:

• speed versus accuracy;
• breadth versus depth;
• fluency versus precision;
• support versus independence;
• quantity versus quality.

A comparative on one dimension can hide deterioration on another.

The comparative completion test

Whenever you see a comparative, complete this sentence:

“X is more/less/better/worse/higher/lower than ______ on ______ over ______ for ______.”

If the blanks cannot be filled, the comparison is not yet interpretable.

Laboratory B: the hidden baseline — the missing half of the comparison

NIST describes a reference base as a source of authority or agreed reference used for comparison in measurement. Classroom reasoning uses the same architecture even when the stakes are smaller.

The baseline can be:

• a previous score;
• a starting state;
• a control group;
• an accepted standard;
• a cohort average;
• a target;
• a personal best;
• an age-level expectation;
• a different method;
• a hypothetical counterfactual.

Different baselines answer different questions.

Case 6. Same score, three stories

Hana scores 78.

Compared with her last score of 70:

Hana improved by 8 points.

Compared with the class average of 82:

Hana scored 4 points below the class average.

Compared with her personal target of 75:

Hana exceeded her target by 3 points.

None of the sentences is necessarily false.

Each selects a different baseline.

This is why sophisticated comparison requires:

baseline disclosure.

Case 7. Baseline choice can manufacture drama

A programme’s average rises from 50 to 60.

Possible statements:

• Scores increased by 10 points.
• Scores increased by 20% relative to the starting score.
• The final average remained 15 points below a target of 75.

All can be true.

They answer different comparison questions.

The problem begins when the writer presents one as if it were the only natural description.

Case 8. Personal baseline versus population baseline

Maya moves from 55 to 68 while the cohort average is 75.

Personal-growth framing:

Maya improved substantially.

Population-position framing:

Maya remains below the cohort average.

Educationally, both can matter.

A good report can hold both simultaneously:

Maya improved by 13 points from her previous result but remains 7 points below the current cohort average.

This preserves progress without hiding current standing.

Case 9. Benchmark drift

Suppose the benchmark itself changes.

Year 1 class average: 70.

Year 2 class average: 76.

Ethan scores 74 in both years.

Absolute performance is unchanged on this score scale.

Relative standing declines from above-average to below-average.

“Better” and “worse” depend on which baseline is being tracked.

The baseline ladder

1. Self-over-time: compared with the learner’s earlier state.
2. Peer: compared with a cohort or class.
3. Standard: compared with an external requirement.
4. Target: compared with a desired future state.
5. Alternative method: compared with another strategy.
6. Counterfactual: compared with what would likely have happened otherwise.

A mature comparison names which rung it is using.

Laboratory C: improved, reduced, increased and decreased — change words always inherit a starting point

Improved sounds like a complete idea.

It is not.

Improved compared with:

• the previous attempt?
• a baseline test?
• a control group?
• an expected trajectory?
• the learner’s average?
• a target?

The same is true of reduced, increased, decreased, rose, fell, grew, shrank, accelerated and slowed.

All are relational verbs.

They need a before-state, comparator or expected state.

Case 10. Improved from one unusually bad result

Maya usually scores around 72.

One difficult paper: 58.

Next paper: 70.

Headline:

Maya improved by 12 points.

True relative to the immediately previous paper.

But relative to her longer-run average, she mostly returned to normal.

The baseline choice changes the story:

• previous paper baseline → large improvement;
• usual-performance baseline → recovery toward normal;
• target 80 → still below target.

The comparative vocabulary should reveal which story is being told.

Case 11. Reduced from which denominator?

Week 1:

• 10 errors;
• 100 attempted questions.

Week 2:

• 8 errors;
• 50 attempted questions.

Raw error count fell by 2.

Error rate rose from 10% to 16%.

Therefore:

Errors were reduced

can be true if it means the raw count.

It is misleading if the intended claim is that performance became more accurate.

CDC guidance on counts and rates makes this principle explicit in another domain: counts are only directly comparable when the underlying populations are the same or approximately the same size; rates correct for differences in population size or exposure period.

The classroom translation is simple:

When opportunity changes, compare rates as well as counts.

Case 12. More correct answers can coexist with lower accuracy

Paper A:

• 40 attempted;
• 36 correct;
• 90% accuracy.

Paper B:

• 80 attempted;
• 64 correct;
• 80% accuracy.

Paper B contains more correct answers.

Paper A has higher accuracy.

If the purpose is productivity, Paper B may look better.

If the purpose is precision, Paper A may look better.

The measure chooses the winner.

Case 13. Faster can mean less complete

Jia Jun finishes in 70 minutes instead of 85.

He leaves two questions blank.

Sentence A:

Jia Jun completed the paper 15 minutes faster.

True as elapsed time.

Sentence B:

Jia Jun became more efficient.

Not automatically true.

Efficiency requires a relationship between resources used and useful output produced.

Finishing earlier by abandoning work may improve speed while damaging completeness.

Case 14. A lower score can still be better performance

Paper A:

• score 82;
• easy paper;
• class average 84.

Paper B:

• score 78;
• much harder paper;
• class average 68.

Raw score fell.

Relative performance improved.

This is why exam analysis cannot treat every score difference as learner change.

The task changed too.

The change audit

1. Starting value: what was the actual baseline?
2. Final value: what changed?
3. Measure: count, rate, score, time, rank or proportion?
4. Opportunity: did the number of attempts or exposures change?
5. Difficulty: was the task comparable?
6. Reference group: did the cohort or benchmark change?
7. Longer-run baseline: was the starting point unusually high or low?

Laboratory D: percentage change, percentage points and the denominator trap

Percentage language is where comparative vocabulary begins to look mathematical.

It is also where hidden baselines become easiest to miss.

NIST expresses percentage difference relative to a specified reference value. The Office for National Statistics distinguishes percentage change from percentage-point change: when a percentage moves from 10% to 9%, that is a fall of 1 percentage point, not a fall of 1%.

This distinction belongs in advanced vocabulary because the words percent, percentage, points, relative, increase and decrease alter the comparison.

Case 15. Ten percentage points versus twenty per cent

A score rate rises from 50% to 60%.

Absolute percentage-point change:

+10 percentage points.

Relative percentage change:

(60 − 50) / 50 = 20%.

Both describe the same move.

They are not interchangeable.

Sentence A:

Accuracy rose by 10 percentage points.

Sentence B:

Accuracy rose by 20% relative to the starting accuracy.

The second sounds larger because the baseline sits in the denominator.

Case 16. Small baseline, huge percentage

A learner moves from 1 successful retrieval to 3.

Absolute increase:

2 successful retrievals.

Relative increase:

200%.

“A 200% increase” sounds enormous.

It is mathematically correct relative to a tiny starting value.

This is why large relative percentages should trigger the baseline question immediately.

Case 17. The zero-baseline problem

A learner had 0 correct uses of a target structure in Week 1 and 4 in Week 2.

How much percentage increase?

The ordinary relative-change formula divides by the starting value.

Division by zero makes that percentage-change formulation undefined.

Write:

Successful uses increased from 0 to 4.

Do not invent an infinite or arbitrary percentage.

Case 18. A 50% reduction is not the reverse of a 50% increase

Start at 100.

Reduce by 50% → 50.

Increase 50 by 50% → 75.

To return from 50 to 100 requires a 100% increase.

Relative changes are asymmetric because the denominator changes.

This matters whenever headlines compare rises and falls as if equal percentages cancel.

Case 19. “Twice as high” versus “200% higher”

Baseline: 10.

New value: 20.

20 is:

• twice the baseline;
• 200% of the baseline;
• 100% higher than the baseline.

200% higher would mean baseline plus two additional baselines:

30.

This is a common language-number trap.

Case 20. Percentage of what?

Twenty per cent of students improved.

Denominator:

• all enrolled students?
• students who took both tests?
• students who completed the intervention?
• students with complete data?

A percentage is a fraction with the denominator disguised by the percent symbol.

Advanced readers reconstruct the denominator.

The percentage audit

1. What is the numerator?
2. What is the denominator?
3. What is the starting value?
4. Percentage change or percentage points?
5. Is the baseline unusually small?
6. Is the baseline zero?
7. Would the absolute change tell a less dramatic but more useful story?

Counts, rates and ratios: “more” can disappear when we divide

CDC guidance gives a powerful general rule: counts can be compared directly only when the populations or opportunities behind them are comparable; rates use a denominator to adjust for different population sizes or exposure periods.

In learning:

error count ÷ opportunities = error rate.

correct answers ÷ attempted answers = accuracy.

words retrieved ÷ words tested = retrieval success rate.

questions completed ÷ minutes = throughput.

One numerator can support several comparisons depending on the denominator.

Case 21. More errors in the stronger student

Student A:

• 10 errors;
• 200 questions attempted;
• 5% error rate.

Student B:

• 6 errors;
• 60 questions attempted;
• 10% error rate.

A has more errors.

B has the higher error rate.

“Who makes more mistakes?” has two legitimate readings.

The analyst must specify whether they mean count or rate.

Case 22. More reading does not automatically mean greater reading exposure per day

Month A:

• 300 pages over 30 days;
• 10 pages/day.

Month B:

• 280 pages over 20 active reading days;
• 14 pages per active reading day.

Month A has more total pages.

Month B has greater intensity on reading days.

Different denominators answer different behavioural questions.

Case 23. The denominator changes midstream

A school reports:

The proportion of students reaching the target rose from 60% to 70%.

But Year 1 includes all enrolled students while Year 2 includes only students who sat the final assessment.

The percentages no longer share the same denominator definition.

Before comparing rates, compare denominator rules.

The denominator audit

1. What was counted?
2. Out of how many opportunities?
3. Was the denominator defined the same way in both groups?
4. Did exposure time differ?
5. Did task quantity differ?
6. Would rate reverse the raw-count comparison?

Laboratory E: benchmarks, targets, norms and standards are different baselines

The word benchmark is often used as if every reference point were interchangeable.

They are not.

A benchmark can be descriptive:

the class average is 72.

A standard can be normative:

students are expected to reach 75.

A target can be aspirational:

Hana aims for 85.

A baseline can be historical:

Hana started at 68.

A reference value can be agreed for measurement or calibration.

Each supports a different comparison.

Case 24. Above average is not the same as meeting the standard

Class average: 62.

Required standard: 70.

Student score: 66.

Two statements:

• The student performed above average.
• The student did not yet meet the required standard.

Both can be true.

“Above average” is relative to peers.

“Meets standard” is relative to a criterion.

Peer comparison and criterion comparison answer different questions.

Case 25. Meeting the target can coexist with declining relative position

Maya target: 75.

Maya score: 76.

Class average: 80.

She met the personal target.

She is below the class average.

If last term the class average was 72, her relative position may have worsened even while she crossed her target.

This is why good progress reporting can contain several baselines without contradiction.

Case 26. “Best” depends on the objective function

Method A:

• highest immediate quiz score;
• lowest delayed retention.

Method B:

• slightly lower immediate quiz score;
• stronger delayed retention.

Which method is best?

For tomorrow’s quiz, perhaps A.

For long-term learning, perhaps B.

Superlatives hide the criterion as well as the comparison group.

Cambridge Grammar defines superlatives relative to the whole group under comparison. Advanced reasoning adds:

best on which outcome, over what time horizon, among which options?

The benchmark audit

1. Historical baseline: where did this person or system start?
2. Peer norm: how does it compare with others?
3. Criterion: does it meet a fixed standard?
4. Target: does it reach the desired state?
5. Best available alternative: what other option is being compared?
6. Objective: what outcome defines “better” or “best”?

Laboratory F: reference classes — the comparison group can decide the conclusion

A reference class is the group against which a person, score, rate or outcome is compared.

The vocabulary may say:

• above average;
• below typical;
• higher than peers;
• among the strongest;
• top-performing;
• unusually low.

Each phrase is meaningless until the group is named.

Case 27. Above average in which group?

Ethan scores 80.

Compared with all students in the cohort, average 72:

Ethan is above average.

Compared with the advanced class, average 84:

Ethan is below average.

Same score.

Opposite comparative.

The reference class decides the relation.

Case 28. “Top 10%” does not mean “scored 90%”

NIST defines percentiles through ordered data and the proportion of observations below or above a value.

A learner at the 90th percentile is positioned relative to a reference distribution.

That does not mean the learner answered 90% of questions correctly.

ETS likewise distinguishes percent-correct scores from percentile rank and notes that percent-correct scores are not necessarily comparable across different tests because difficulty can differ.

Three quantities:

• raw score;
• percentage correct;
• percentile rank.

They answer different questions.

Case 29. Same percentile, different mastery

Suppose Hana is at the 80th percentile on two different tests.

Test A is easy and she scores 92% correct.

Test B is difficult and she scores 68% correct.

Her relative standing can be similar while raw mastery measures differ.

Percentile describes position in a distribution.

It does not directly describe what the learner knows.

Case 30. Average changes when the population changes

Year 1 average is calculated across all students.

Year 2 average excludes students who withdrew.

The average may rise partly because the composition of the group changed.

“The average improved” may therefore describe:

• the same people learning more;
• different people being included;
• both.

When the reference class changes, the baseline moves beneath the comparison.

Case 31. The filtered comparison

A learning programme reports:

Students who completed all ten sessions improved more than students who completed fewer sessions.

Possible interpretation:

more sessions helped.

Alternative:

students able or motivated to complete all ten sessions may already differ from those who did not.

The comparison group contains a selection effect.

The Baseline Audit and Cause–Mechanism Audit meet here:

first ask who is being compared;

then ask whether the observed difference was caused by the treatment.

The reference-class audit

1. Who is in the comparison group?
2. Who is excluded?
3. Did group membership change across time?
4. Is the comparison with all students, similar students, top performers or completers only?
5. Does the comparative reverse under another reasonable reference class?

Rank, percentile, score and mastery: four scales that answer different questions

Educational reporting often compresses these into “better” and “worse”.

Keep them separate.

Raw score

How many marks or points were earned?

Percentage correct

What proportion of the available marks or items were correct?

Rank

Where does the learner sit in ordered comparison with others?

Percentile

What proportion of the reference distribution falls below the learner’s score, subject to the specific percentile convention used?

Mastery

What can the learner reliably do?

A learner can rise in rank because others fall.

A learner can improve mastery while falling in rank if peers improve faster.

A learner can keep the same percentile on tests with different difficulty.

A learner can score the same raw mark while the item mix changes.

“Better” cannot tell us which of these moved.

The rank audit

1. Absolute or relative performance?
2. Same test or different tests?
3. Same reference group?
4. Same difficulty?
5. Rank, percentile or mastery?

Equality language: “as good as” hides the same comparison machinery

Cambridge Grammar describes as … as constructions as comparisons of equality.

But equality on one dimension does not mean identity overall.

Case 32. As accurate as, but not as fast as

Maya and Hana both score 90% accuracy.

Maya takes 75 minutes.

Hana takes 55.

Maya is as accurate as Hana.

True.

Maya performs as well as Hana.

Depends on whether “well” includes speed.

Equality language needs the same measure audit as comparative language.

Case 33. Not as bad as is not the same as good

The second draft was not as weak as the first.

This means the second draft improved relative to the first.

It does not mean the second draft is strong.

Comparative improvement and absolute adequacy are separate.

Case 34. “Similar” requires tolerance

The two classes performed similarly.

How similar?

• within 1 point?
• within 5 points?
• statistically indistinguishable?
• same rank band?
• same qualitative profile despite different averages?

“Similar” is a tolerance claim.

The acceptable distance must be explicit or contextually justified.

Superlatives: best, worst, highest, lowest, fastest and most effective require a search space

A superlative compares one item with a whole set.

The hidden question is:

whole set of what?

Case 35. “The best method” among three weak methods

Three methods are compared.

Method B performs best.

That does not mean B is optimal.

It means B was best among the tested set on the chosen measure.

A missing fourth method might outperform all three.

Superlatives need a declared search space.

Case 36. Highest score in a small sample

Ethan had the highest score.

Among:

• three students?
• thirty students?
• the whole level?
• students who completed the paper?

The prestige of highest changes with the comparison set.

The superlative audit

1. What is the full comparison set?
2. Which measure defines best/worst?
3. What time horizon?
4. Were all plausible alternatives included?
5. Does “best among tested options” get wrongly inflated into “best overall”?

Laboratory G: AI baseline drift — when a rewrite quietly changes what the sentence is compared with

AI systems often improve fluency by completing missing structure.

That is useful when the missing structure is grammatical.

It is dangerous when the missing structure is evidential.

Comparative language is particularly vulnerable because the comparator is often omitted in natural prose.

Case 37. “Improved” becomes “improved significantly”

Source:

The average score increased from 70 to 73.

AI rewrite:

The class improved significantly.

Two problems appear.

• Improved adds a positive evaluation rather than merely reporting increase.
• Significantly may imply practical importance or statistical significance without evidence for either.

Safer:

The class average increased by 3 points, from 70 to 73.

Case 38. “Higher” becomes “better”

Source:

The intervention group had a higher average homework completion time.

AI summary:

The intervention group showed better homework performance.

Higher time does not automatically mean better performance.

It may mean:

• greater persistence;
• slower working;
• more difficult homework;
• more distraction;
• more complete work.

The AI converted a directional comparison into an evaluative conclusion.

Case 39. “Compared with last week” becomes “better than usual”

Source:

Maya completed more questions than last week.

AI rewrite:

Maya completed more questions than usual.

Last week is one observed point.

Usual means a broader historical baseline.

The model enlarged the comparison set.

If last week was unusually low, the rewrite can be misleading.

Case 40. Relative gain loses its denominator

Source:

Accuracy rose from 50% to 60%, a 10-percentage-point increase.

AI concise rewrite:

Accuracy improved by 20%.

The 20% relative change is mathematically defensible.

But the rewrite changes the representation.

A reader may hear “20 percentage points”.

Good editing preserves the scale when the scale is material.

Case 41. “Top among the four tested methods” becomes “the best method”

Source:

Method C had the highest delayed-recall score among the four methods tested.

AI headline:

Method C is the best revision method.

The model changed:

• one outcome into overall quality;
• four tested options into the universe of revision methods;
• a study result into a general recommendation.

This is superlative expansion.

Case 42. Rank improvement becomes learning improvement

Source:

Ethan moved from 12th to 8th in class rank.

AI report:

Ethan’s academic performance improved.

Maybe.

But rank can rise because:

• Ethan’s score rose;
• other students’ scores fell;
• both;
• the class composition changed.

Rank is relative position.

Learning is change in capability.

They can correlate.

They are not synonyms.

A practical AI prompt for baseline preservation

Use:

Rewrite for clarity without changing the comparator, baseline, denominator, reference class, time window, scale or evaluation criterion. Preserve the distinction among raw counts, rates, percentages, percentage points, rank, percentile and mastery. Do not replace higher/lower with better/worse unless the source explicitly evaluates the direction. Do not replace “than last week” with “than usual”, “highest among tested options” with “best overall”, or a rank change with a learning change. Flag every alteration to more, less, fewer, better, worse, higher, lower, improved, reduced, increased, decreased, fastest, slowest, best, worst, average, typical, benchmark, target, percentile and percentage.

The baseline invariance test

Before accepting a rewrite, compare source and rewrite on six fields:

1. Comparator: same thing being compared against?
2. Measure: same outcome?
3. Scale: same units?
4. Population: same group?
5. Time: same period?
6. Evaluation: did a neutral difference become better/worse?

Extended case files: comparison claims under pressure

Case file 1. The student who “improved” while the paper got easier

Constructed record:

• Paper A: score 65; class average 60.
• Paper B: score 72; class average 75.

Raw-score story:

The student’s score improved by 7 points.

Relative-to-class story:

The student’s position relative to the class declined from 5 points above average to 3 points below average.

Which is correct?

Both.

What cannot be concluded from these two numbers alone?

That underlying mastery definitely improved.

The task difficulty may have changed.

The Baseline Audit preserves both observations without forcing them into one verdict.

Case file 2. The tutor advertisement with a tiny starting value

Constructed claim:

Students made a 300% improvement in advanced vocabulary use.

Underlying count:

• baseline: 1 correct target use;
• final: 4 correct target uses.

Relative increase:

300%.

Absolute increase:

3 uses.

The percentage is not false.

It becomes misleading if it is presented without the baseline because the dramatic percentage depends on a tiny denominator.

Better disclosure:

Correct target uses increased from 1 to 4 in the sampled task.

Case file 3. More homework completed, worse homework accuracy

Week A:

• 20 questions completed;
• 18 correct;
• 90% accuracy.

Week B:

• 35 questions completed;
• 28 correct;
• 80% accuracy.

Quantity improved.

Accuracy declined.

A parent saying you did better this week needs to decide which objective matters.

If the child was previously under-practising, greater volume may be useful.

If accuracy matters, the decline needs attention.

Good comparison can hold a trade-off:

You completed 15 more questions, but accuracy fell by 10 percentage points. The next goal is to keep the higher volume while restoring accuracy.

Case file 4. The “more efficient” learner

Before:

• 80 minutes;
• 45 marks attempted;
• 40 correct.

After:

• 65 minutes;
• 40 marks attempted;
• 38 correct.

Elapsed time improved.

Marks attempted fell.

Accuracy among attempted marks rose slightly.

Overall marks fell.

“More efficient” is under-defined.

A better diagnosis specifies the ratio:

Accuracy per attempted item improved, but coverage fell and total marks declined.

Case file 5. The school report that changes the denominator

Year 1:

• 100 students enrolled;
• 60 reached target;
• 60% of enrolled students reached target.

Year 2:

• 100 students enrolled;
• 80 sat final test;
• 56 reached target;
• 70% of tested students reached target.

Headline:

Target attainment rose from 60% to 70%.

Problem:

The denominators differ.

Measured against all enrolled students, Year 2 has 56% attainment.

The language of improvement depends entirely on the denominator definition.

Case file 6. The percentile that falls while learning improves

Term 1:

• score 60;
• 70th percentile.

Term 2:

• score 72;
• 65th percentile.

The learner’s score rose.

Relative position fell.

This can happen if peers improve more quickly or if the reference group changes.

Do not interpret percentile decline automatically as learning decline.

Case file 7. The “best” revision strategy that wins only the immediate test

Strategy A:

• immediate score 90;
• one-week score 55.

Strategy B:

• immediate score 82;
• one-week score 76.

Which is better?

If the goal is immediate performance, A.

If the goal is retention, B.

If both matter, the evaluation needs weights.

“Best” cannot be separated from objective and time horizon.

Case file 8. The AI report that chooses a flattering baseline

Source data:

• previous score: 60;
• current score: 68;
• cohort average: 74;
• target: 75.

AI report:

The student showed strong performance, improving substantially to 68.

The improvement is real.

“Strong performance” is a new evaluative judgement.

Compared with the target and cohort average, current performance remains below both.

Balanced version:

The score increased by 8 points from 60 to 68. This is meaningful progress from the previous result, although the current score remains 6 points below the cohort average and 7 points below the target.

The full Baseline Audit

1. Comparative word: what word signals a relation?
2. Target: what item or group is being described?
3. Comparator: what is it being compared with?
4. Measure: score, rate, count, rank, time, cost, frequency, accuracy?
5. Direction: higher/lower, more/less, better/worse?
6. Evaluation: is the direction being treated as desirable?
7. Baseline type: historical, peer, standard, target, alternative or counterfactual?
8. Denominator: out of how many opportunities or people?
9. Scale: raw points, percentage, percentage points, ratio, percentile?
10. Time window: compared across which period?
11. Reference class: who is included?
12. Difficulty: are the tasks comparable?
13. Trade-off: did another important measure move the opposite way?
14. Search space: if a superlative is used, among which options?
15. Alternative baseline: would another reasonable baseline change the story?

Independent practice: thirty baseline decisions

For each item, identify the hidden comparison structure. Do not answer only “true” or “false”. State the baseline, measure, scale and any missing denominator.

Use this five-step route:

1. Underline the comparative.
2. Name the target.
3. Name the comparator.
4. Name the measure and scale.
5. Rewrite with the baseline exposed.

Practice set A: more, less, better, worse, higher and lower

1. The new method is better. What four missing pieces are needed before this becomes analytically useful?

2. The second class had a higher average. Why does higher not automatically mean better?

3. Week 1 has 8 errors in 80 attempts; Week 2 has 6 errors in 40 attempts. Which week has fewer errors? Which has the lower error rate?

4. Ethan finished faster. Name one important outcome that could have worsened.

5. The second draft was not as weak as the first. Does that establish that the second draft is strong?

6. The two classes performed similarly. What tolerance information is missing?

Practice set B: baselines and benchmarks

7. Hana scores 78. Her previous score was 70, class average 82 and target 75. Write three correct comparison sentences using three different baselines.

8. A learner’s score is unchanged at 74 while the class average rises from 70 to 76. What changed and what did not?

9. A student scores above the class average but below a fixed proficiency standard. Explain why both statements can be true.

10. A method is “best” on immediate quiz score but worse on delayed recall. What hidden criterion must be named?

11. Maya improved substantially. What baseline would you ask for first?

12. A student rises from 55 to 68 while the cohort average is 75. Write a balanced sentence that preserves both progress and current standing.

Practice set C: percentage language

13. Accuracy rises from 50% to 60%. State the percentage-point change and the relative percentage change.

14. Correct uses rise from 1 to 3. Why does “200% increase” need the starting value beside it?

15. Correct uses rise from 0 to 4. Why is ordinary percentage change from baseline not defined?

16. A value falls from 100 to 50, then rises by 50%. Does it return to 100? Explain.

17. Baseline 10, new value 20. Complete: 20 is ___ times the baseline, ___% of baseline and ___% higher than baseline.

18. Twenty per cent of students improved. Name two plausible denominators that would produce different interpretations.

Practice set D: counts, rates, rank and percentile

19. Student A makes 10 errors in 200 questions; Student B makes 6 errors in 60. Who has more errors? Who has the higher error rate?

20. Why can percent-correct scores from two different tests be hard to compare directly when difficulty differs?

21. What does being at the 90th percentile mean, and what does it not mean?

22. A learner’s percentile falls while their raw score rises. Give one explanation.

23. A class average rises after several low-scoring students withdraw. Why can the class improved be misleading?

24. A programme compares students who completed all sessions with those who did not. What selection issue should be considered before saying more sessions caused better outcomes?

Practice set E: AI baseline drift

25. Source: The average rose from 70 to 73. AI: The class improved significantly. What did the model add?

26. Source: The group had a higher homework completion time. AI: The group performed better. What changed?

27. Source: Maya completed more questions than last week. AI: Maya completed more questions than usual. Why is the second baseline broader?

28. Source: Method C had the highest delayed-recall score among four tested methods. AI: Method C is the best revision method. List three expansions in the AI claim.

29. Source: Ethan moved from 12th to 8th in rank. AI: Ethan learned more. Why does rank change not prove capability change?

30. Source data: previous 60, current 68, cohort 74, target 75. Write a sentence that reports progress without hiding the other two baselines.

Answer commentary

1. At minimum: comparator, measure, population/context and time horizon. Better is evaluative and relational; without those fields, readers cannot know what superiority is being claimed.

2. Higher only indicates direction on a scale. A higher error rate or homework completion time can be undesirable or ambiguous. Evaluation must come from the meaning of the measure, not the comparative adjective itself.

3. Week 2 has fewer raw errors: 6 versus 8. Week 1 has the lower error rate: 10% versus 15%. The numerator fell while the denominator fell more sharply.

4. Accuracy, completeness, number of attempted questions or quality of reasoning could worsen. Speed is one dimension.

5. No. Not as weak establishes a relative improvement only. Both drafts could still be weak against an external standard.

6. We need a threshold or criterion for similarity: for example within 2 points, within measurement uncertainty, within the same performance band or statistically indistinguishable under a specified analysis.

7. Examples: Hana improved by 8 points from her previous score. Hana scored 4 points below the class average. Hana exceeded her target by 3 points. One observation can support several honest comparisons.

8. Ethan’s raw score is unchanged. Relative standing moves from above the class average to below it because the benchmark changed.

9. The class average is a peer norm; the proficiency standard is a criterion. A score can exceed peers while still falling short of the required capability level.

10. The objective and time horizon. “Best” could mean best for immediate performance, delayed retention, efficiency, transfer or some weighted combination.

11. Ask “compared with which prior value or expected trajectory?” Then check whether the baseline was representative or unusually low/high.

12. Example: Maya improved by 13 points from 55 to 68, while her current score remains 7 points below the cohort average of 75.

13. +10 percentage points. Relative increase = 10/50 = 20%.

14. The dramatic 200% depends on a tiny starting value. The absolute change is only 2. Both should be visible when practical significance matters.

15. The standard relative-change formula divides change by the baseline. A baseline of zero creates division by zero. Report the move from 0 to 4 directly.

16. No. After falling to 50, a 50% increase adds 25, reaching 75. The percentage increase uses the new 50 baseline. Returning to 100 requires a 100% increase from 50.

17. 2 times; 200%; 100% higher.

18. For example, 20% of all enrolled students versus 20% of students who completed both assessments. Attrition can change the apparent rate.

19. A has more raw errors: 10 versus 6. B has the higher error rate: 10% versus 5%.

20. Percent correct describes performance on the particular item set. If Test B is harder, 70% there may represent more knowledge than 80% on Test A. ETS explicitly warns that percent-correct scores are not directly comparable across different tests without additional scaling or equating information.

21. It describes position relative to a defined reference distribution: roughly, the score exceeds a large proportion of that group under the percentile convention used. It does not mean 90% correct.

22. Peers may have improved faster, or the reference group may have become stronger. Relative position can fall while absolute capability rises.

23. The population changed. A higher average can be produced by removing lower scores even if the remaining students did not learn anything new.

24. Completion may reflect motivation, resources, schedule stability or starting ability. Completers and non-completers may not be exchangeable before the intervention dose differs.

25. It added evaluation and perhaps unsupported statistical/practical importance. A 3-point increase is the direct observation; significantly needs its own justification.

26. A neutral directional difference became a positive evaluation. More time could indicate persistence or inefficiency; the source does not decide.

27. Last week is a single specified comparison point. Usual implies a broader distribution of past behaviour. The rewrite invents a historical baseline.

28. It expands one outcome into overall quality, four tested options into all possible revision methods, and a bounded study result into a general recommendation.

29. Rank depends on everyone else. Ethan can rise because others fall or because group membership changes. Capability change must be measured directly.

30. Example: The score rose by 8 points from 60 to 68. This represents progress from the previous result, while the current score remains 6 points below the cohort average and 7 points below the target.

The comparison independence test

A learner has strong baseline awareness when they can increasingly:

• spot comparative adjectives and adverbs quickly;
• reconstruct an omitted than-phrase;
• separate direction from evaluation;
• ask “better on which measure?”;
• distinguish raw counts from rates;
• reconstruct hidden denominators;
• identify historical, peer, criterion, target and counterfactual baselines;
• report progress and current standing at the same time;
• distinguish percentage change from percentage-point change;
• notice tiny baselines behind large percentages;
• handle zero baselines correctly;
• understand the asymmetry of percentage rises and falls;
• distinguish “twice as much” from “200% higher”;
• separate raw score, percent correct, rank and percentile;
• recognise that percentile rank is relative to a reference group;
• recognise that rank can change without capability changing;
• notice when the comparison group changes;
• notice when task difficulty changes;
• notice when completion or selection changes the sample;
• identify trade-offs such as speed versus accuracy;
• treat similar as requiring a tolerance criterion;
• treat best as requiring an objective and a search space;
• avoid promoting “higher” into “better” without justification;
• detect AI baseline expansion;
• detect AI denominator loss;
• detect AI superlative expansion;
• write comparison sentences with explicit comparator, measure, scale and time window;
• test whether another reasonable baseline would materially change the story.

The comparison vocabulary field guide: words that change the baseline, scale or judgement

The most dangerous comparison words are not difficult words.

They are familiar words whose reference point has become invisible.

A strong learner does not merely know what more, less, better, worse, higher, lower, improved, average, typical, similar, best and double mean.

They know what information each word demands from the rest of the sentence.

Directional comparatives: more, less, fewer, higher, lower, faster, slower

More. A quantity is larger relative to a comparator. Ask what is counted and whether exposure or opportunity changed.

Less. A quantity is smaller. In careful grammar it commonly modifies uncountable nouns; in analysis the more important question is which quantity and denominator are being reduced.

Fewer. Traditionally used with plural countable nouns. Fewer errors still does not tell us whether the error rate fell.

Higher. A value moves upward on a scale. It is directional rather than automatically desirable.

Lower. A value moves downward. A lower score may be worse; a lower error rate may be better; a lower completion time may be better only if completeness is preserved.

Faster / slower. Comparison on time or rate. Speed alone cannot establish efficiency or quality.

Evaluative comparatives: better, worse, superior, inferior

Better. Requires an objective. Better for immediate score, retention, transfer, independence, cost, speed or accuracy?

Worse. The mirror problem. A result can be worse on one dimension and better on another.

Superior / inferior. Stronger formal versions of better/worse. They can sound scientific while concealing an undefined criterion. Ask superior on what measure and within what uncertainty.

Evaluative comparatives are more demanding than directional ones.

Higher needs a scale.

Better needs a scale plus a value judgement about the direction.

Change words: increase, decrease, rise, fall, improve, decline, grow, shrink

Increase / rise / grow. The later value exceeds the baseline on a specified measure.

Decrease / fall / shrink. The later value is below the baseline.

Improve. Adds evaluation. A score can increase without performance necessarily improving if the task becomes easier. A completion time can decrease without efficiency improving if work is skipped.

Decline. Often evaluative as well as directional. A rank can decline while mastery improves if peers improve faster.

Reference-to-standard words: meet, exceed, surpass, fall short, match

Meet. Reach a defined criterion or threshold.

Exceed / surpass. Go beyond a stated benchmark, target or previous value. These words are uninterpretable unless the benchmark is visible.

Fall short. Remain below a criterion. This says nothing about whether progress occurred from baseline.

Match. Reach approximately the same value under a specified measure. The acceptable tolerance matters.

A learner can improve, exceed a previous personal best and still fall short of an external standard.

Those statements are compatible because they use different reference points.

Equality and similarity: equal, equivalent, comparable, similar, as … as

Equal. Same value on a specified measure. Equal scores do not imply equal knowledge profiles.

Equivalent. Often stronger than equal because it can imply interchangeability for a purpose. Two tests can have equal average scores without being equivalent measures.

Comparable. Suitable for meaningful comparison. This does not mean identical. It means differences in measurement, population or conditions are controlled or small enough for the intended comparison.

Similar. Close enough under an unstated or stated tolerance. Advanced writing should make the tolerance defensible.

As … as. Marks equality on the named dimension. As accurate as does not mean as fast as or as independent as.

Relative-position words: average, above average, below average, typical, unusual

Average. Usually refers to a mean in everyday reporting, but the word can also be used loosely to mean ordinary. Clarify whether the statistic is actually a mean and which population it describes.

Above / below average. Relative to a specified group and measure. The same score can be above average in one class and below average in another.

Typical. Often means representative or usual, not necessarily numerically average. A distribution can have an average that few individuals actually resemble.

Unusual. Requires a reference distribution or expectation. Unusual among whom, when and under which conditions?

Rank words: leading, top, bottom, highest, lowest, ahead, behind

Leading / ahead. Relative position in a defined field. The field can be tiny or large.

Top / bottom. Rank language. Top 10% is a relative-position statement, not a statement about percentage correct.

Highest / lowest. Superlatives that require the comparison set to be declared.

Ahead / behind. Can refer to score, schedule, age expectation, syllabus progression or rank. The dimension must be supplied.

Proportional words: double, triple, half, twice, three times, ratio

Double / twice as much. New value equals two times the reference value.

100% higher. Also equals two times the baseline.

200% higher. Equals three times the baseline, not twice.

Half. Fifty per cent of the baseline value. A fall by half and a later rise by half do not cancel because the second percentage uses a smaller baseline.

Ratio. Compares one quantity directly with another. Ratios preserve relationships that raw differences can hide.

Magnitude words: slightly, marginally, moderately, substantially, considerably, dramatically

These words evaluate the size of a difference.

They need a scale or context.

A 3-point change can be:

• tiny on a 1,000-point scale;
• large on a 10-point scale;
• statistically detectable but practically unimportant;
• practically important despite not being statistically precise in a small sample.

Significant deserves special caution because it can mean important in everyday English or statistically significant in technical writing.

Do not let one meaning impersonate the other.

The field-guide rule: every comparative word creates a blank

Train the learner to hear the blank.

Better… than what, on what?

Higher… on which scale?

Improved… from which baseline?

Average… of which group?

Similar… within what tolerance?

Best… among which options, for which goal?

The word becomes precise when its blank is filled.

Cross-subject comparison: the baseline changes its costume but not its logic

English: comparative claims in source analysis

A source may claim:

Young people are reading less.

The Baseline Audit asks:

• less than which age group or earlier period?
• less in minutes, books, pages or frequency?
• print only, or all long-form reading?
• same survey method?
• same population?

A comprehension answer becomes stronger when it can distinguish the explicit comparison from the reader’s imagined one.

Mathematics: additive and multiplicative comparison are different languages

From 10 to 20:

• increase of 10;
• twice as large;
• 100% increase;
• ratio 2:1.

These are four ways to describe one relation.

Students need to know when a question asks for:

• absolute difference;
• relative difference;
• ratio;
• percentage change;
• percentage-point change.

A vocabulary error can become a mathematics error before any arithmetic begins.

Science: fair comparison requires controlled conditions

Plant A grew more than Plant B.

Useful only if the comparison conditions are interpretable.

• same species?
• same starting size?
• same time window?
• same light and water except the variable being tested?
• growth measured by height, mass or leaf count?

The scientific control condition is a formalised baseline.

The phrase grew more becomes a scientific result only after the baseline architecture is stable.

Humanities: totals and per-capita rates tell different stories

Country A has more total cases than Country B.

Country A also has ten times the population.

Total count answers one question.

Per-capita rate answers another.

Historical and social comparisons also need stable definitions:

• same borders?
• same age categories?
• same measurement systems?
• same purchasing-power basis?
• same inflation treatment?

“More” across unequal populations or changing definitions can be mathematically true and analytically weak.

Examinations: progress and standard should be tracked together

A learner can improve sharply from 45 to 62 and still remain below the required examination standard.

A learner can remain above the passing standard while declining from 85 to 76.

Both dimensions matter:

• change from self-baseline;
• distance from performance criterion.

Progress language should not erase readiness language.

A parent route: compare the child without turning comparison into identity

Families use comparison constantly.

Your sister finishes faster.

You did worse this term.

You are below average.

This is much better.

Comparison can diagnose.

It can also humiliate, confuse or hide the actionable cause.

A practical family sequence is:

1. Start with self-over-time. What changed in the child’s own process?
2. Add the external standard. What remains to be learned for the real task?
3. Name the measure. Speed, accuracy, completeness, independence or score?
4. Avoid sibling identity comparisons. “Your sister is better at English” hides too many dimensions and invites a trait conclusion.
5. Compare processes when possible. “You checked two questions this time instead of none” creates an actionable next step.

A parent can say:

Your score rose by 8 points from the previous paper, which is real progress. The current score is still 6 points below the target, so our next job is to preserve the gain and repair the remaining gap.

This sentence does two things at once:

• recognises growth;
• preserves the performance standard.

It does not require fake celebration or unnecessary criticism.

A tutor route: keep two baselines and two denominators

A useful tuition progress system tracks at least two baselines:

• self-baseline: is this learner improving from their own earlier state?
• criterion baseline: are they approaching the level required for independent performance?

It also tracks at least two kinds of quantity:

• counts: how many correct answers, errors, retrievals, prompts or unfinished questions?
• rates: correct answers per attempt, errors per opportunity, prompts per task, questions per minute?

And it should track paired trade-offs:

• speed × accuracy;
• quantity × quality;
• support × independence;
• immediate performance × delayed retention;
• breadth × depth.

One metric alone creates false victories.

If prompts fall from six to two while accuracy is stable, independence improved.

If speed rises while accuracy collapses, automaticity did not improve.

If raw correct answers rise only because twice as many questions were attempted, rate must be checked.

If rank rises but independent task quality is unchanged, relative standing improved without clear evidence of capability growth.

The tutor’s comparison language should reveal the mechanism of progress rather than merely announce that something is “better”.

A compact comparison dashboard for one learner

• Score: 64 → 72.
• Target: 75.
• Accuracy: 82% → 88%.
• Completion: 90% → 100%.
• Time: 95 min → 85 min.
• Prompts needed: 5 → 2.
• One-week retention: 60% → 74%.

Now improved can be unpacked:

The learner’s score increased by 8 points and is now 3 points below target. Accuracy rose by 6 percentage points, the paper was completed fully, completion time fell by 10 minutes, external prompts fell from five to two, and delayed retention rose by 14 percentage points.

The long sentence is more useful because the baseline has been made visible.

One late comparison in Punggol

The four students are looking at a progress card.

Previous score: 50.

Current score: 60.

Class average: 66.

Target: 70.

Jia Jun says, “A twenty per cent improvement. Huge.”

Maya says, “But still below average.”

Hana says, “And still below target.”

Ethan writes four lines:

• +10 raw points;
• +20% relative to the starting score;
• 6 points below class average;
• 10 points below target.

“Which one is the truth?” Maya asks.

“All four,” Hana says.

Jia Jun looks at his headline again.

“So ‘huge’ is the part I invented.”

Ethan nods.

They rewrite:

The score increased from 50 to 60, a gain of 10 points or 20% relative to the starting score. This is clear progress from baseline, while the current score remains 6 points below the class average and 10 points below the target.

Nothing has been hidden.

Nothing has been exaggerated.

The comparison now tells the reader where the learner came from, where the learner is, and what remains.

The real goal: make the reference point visible

Comparison words are compact because they omit information that context often supplies.

That efficiency is useful in ordinary conversation.

It becomes dangerous in analysis, reporting and persuasion when the omitted baseline changes the conclusion.

Better.

Better than what?

Higher.

Higher on which measure?

Improved.

Improved from which starting point?

Twenty per cent.

Twenty per cent of what, relative to what?

Above average.

Average of which group?

Best.

Best among which options, for which objective?

The Baseline Audit does not make every sentence longer.

It makes the writer know what the short sentence is leaving out.

A comparison is only as honest as its reference point.

Research notes and further reading

The sources below support the grammar, measurement, percentage, rate and score-interpretation principles. The Baseline Audit, resident-character cases, parent/tutor routes, dashboard and practice sequence are original instructional material.

• Cambridge Grammar: Comparison — overview of comparative structures.
• Cambridge Grammar: Comparative and superlative adjectives — forms such as bigger, more, better, best and their comparison functions.
• Cambridge Grammar: Comparing nouns — more, less, fewer and related noun comparisons.
• Cambridge Grammar: Comparisons — equality and comparative constructions including as … as.
• British Council LearnEnglish: Comparative and superlative adjectives — practical comparison grammar.
• NIST: Reference base — reference values and bases used for meaningful measurement comparison.
• NIST: Percentage difference — percentage difference relative to a specified reference value.
• NIST: Percentiles — interpretation of percentile position within ordered data.
• UK Office for National Statistics: Percentages — distinction between percentage change and percentage-point change.
• CDC Field Epidemiology Manual: Describing epidemiologic data — why counts and rates answer different comparison questions when populations or exposure differ.
• ETS Standards for Quality and Fairness — distinctions among percent-correct scores, percentile ranks and comparability across tests.

Continue through the advanced eduKatePunggol vocabulary chain

• Advanced Vocabulary Causality | The Cause–Mechanism Audit
• Advanced Vocabulary Sequence | The Timeline Audit
• Advanced Vocabulary Cohesion | The Reference Audit
• Advanced Vocabulary Perspective | The Framing Audit
• Advanced Vocabulary Agency | The Hidden-Agent Audit
• Advanced Vocabulary Synthesis | The Contrast Audit
• Advanced Vocabulary Argument | The Qualification Audit
• Advanced Vocabulary Reading | The Assumption Audit
• Advanced Vocabulary Editing | The Counterexample Test
• Education: Regarding Vocabulary