A B1 English learner can read a rubric that awards four points for ‘coherence’ and still have no idea why one paragraph is easier to follow than another. Put two anonymous paragraphs side by side and ask which one works better, though, and the learner has somewhere concrete to begin. For writing assessment, comparison should often come before the rubric. The choice creates a need for language such as focus, sequence and support. The rubric can name those features after the learner has seen them. This article is for a university student in Taipei who can finish a 250-word English response but receives comments such as ‘develop your ideas’ and ‘improve organisation’. Those phrases describe a destination. They do little to show the next move on Tuesday evening.
Why a rubric can arrive too early
Learners see a label, not a move.
Analytic rubrics divide writing into categories and make grading more transparent. They also compress difficult judgements into short descriptors. ‘Clear progression’ may refer to the order of claims, the relation between evidence and conclusion, the placement of a qualification, or all three. A learner who has not yet built a mental picture of strong work must interpret the descriptor and diagnose the draft at the same time. Because the words apply to many genres and levels, even a careful learner may circle the right box while remaining unable to locate the sentence that weakened the response or decide what to change. Pairwise comparison lowers that first demand. The learner reads Text A and Text B in response to the same prompt, then chooses which is more successful for one stated purpose. The first prompt might be narrow: ‘Which paragraph makes its reason easier to follow?’ A second comparison can ask about evidence. Only after several choices does the class name the recurring grounds for judgement.
Researchers call the method comparative judgement. In assessment research, judges compare two performances instead of assigning each one an isolated score. A 2024 study in Assessing Writing found that community judges could produce reliable and valid rankings of second-language texts, although texts within a narrow proficiency band were harder to distinguish. The study concerned assessment rather than classroom learning, yet it establishes a useful boundary: comparison can support consistent judgement, and close cases demand more care. Read the study and its limitations.
Comparison has done its first job.
What learners attend to
Classroom value lies in the explanation after the choice. ‘B is better’ gives the teacher almost nothing. ‘B states the claim in sentence one, then each example refers back to it’ exposes a usable principle. The learner has converted a felt difference into evidence. A study of an International Trade English writing class compared two ways of working with examples. One group applied a list of criteria. Another made comparative judgements. The comparison group evaluated more examples and tended to attend to content and structure at a higher textual level, while the criteria group focused more on local suggestions for improvement. The effect on the students’ own writing was moderate, not dramatic. The authors also warned that learners can construct a faulty standard without guidance. The open study supports comparison as a learning activity, not as a self-running cure.
That distinction matters. A teacher should sample the reasons learners give, challenge a weak reason and show why two defensible judgements can differ. Writing quality has several dimensions. A vivid example cannot rescue a paragraph whose conclusion contradicts its claim. A clean structure can carry thin evidence.
A six-text lesson
Learners have to name the evidence.
Start with six short responses to one prompt, ideally drawn from previous work with names and identifying details removed. Three pairs are enough for a first lesson. Make the differences visible without turning one sample into a cartoonishly bad answer. If every pair has an obvious winner because one text contains twelve spelling errors, learners will judge proofreading rather than argument.
For each pair, use one decision and one reason:
- Choose the response that answers the prompt more directly. Underline the line that decided it.
- Compare a fresh pair for use of evidence. Mark the sentence where the evidence begins doing explanatory work.
- Compare the last pair for sequence. Draw arrows between claims that genuinely depend on each other.
Then collect the reasons on the board. The teacher can group ‘gets to the point’, ‘answers in the first line’ and ‘doesn’t wander’ under a provisional label such as focus. The class has earned the term. Now show the relevant rubric row and ask which phrases match the evidence they marked. I'd rather use six paragraphs of 130 words than two polished model essays of 650. More comparisons reveal that quality can be achieved in several ways. They also reduce the chance that learners copy the surface vocabulary of one admired answer.
Revision has to follow
Judging examples can become an elegant detour from writing. The learner may become good at spotting another writer’s weak ending while leaving the same problem in their own draft. A transfer step belongs in the lesson. Ask each learner to reopen a recent paragraph and make one revision based on a criterion the class generated. The instruction should identify an observable change: move the claim above the background, add the missing link between evidence and conclusion, or delete a sentence that serves a different question. Ten minutes is enough for one serious change.
One revision is enough.
This sequence fits well with giving feedback before showing the grade. Both approaches protect the period when the learner is still deciding what to do with the work. A score can wait until after the comparison and revision. Once the number appears, attention often shifts to the result.
Model-text research gives another caution. In a study of low-intermediate secondary learners, students compared their picture-based stories with two native-speaker models. They noticed many lexical problems but found relatively few solutions in the models. The researchers’ report shows why exposure alone is thin feedback. Learners need a question that directs attention and a chance to use what they notice.
Where comparison fails
Pairwise work is a poor fit when the samples differ on every dimension. If Text A has stronger ideas, easier grammar and a familiar topic, no one can tell which feature drove the decision. Match the task, length and approximate level. Change one main dimension at a time when teaching beginners.
It also cannot replace a criterion-referenced decision. A learner preparing for IELTS, GEPT or a university marking scheme eventually needs to know how performance maps onto published standards. Comparison builds the learner’s eye. The rubric connects that eye to the required scale. Teachers still have to explain genre expectations and check factual accuracy. Use the narrowest possible exit ticket. Put two final paragraphs on screen. Ask learners to choose the one with the clearer evidence-to-claim link and circle the exact words that create it. Then have them add one such link to yesterday’s draft before they close the file.
Circle the new link, then close the file.