Paired Speaking Test with Unequal Air Time
On a 12-minute paired speaking test, 24 candidates are tested by two examiners. Candidate A speaks about 70% of the time, interrupts twice, and initiates all three tasks. Candidate B produces eight short turns, answers accurately, uses several fillers, and rarely initiates. Examiner 1 rates B 2/5 for fluency because B contributed little; Examiner 2 rates B 3/5 because B's responses were accurate and coherent. The recording has a 25-second inaudible segment during B's longest turn, caused by a microphone fault. B requests a re-mark, arguing that the partner dominated and that the lost segment contained key content. The test center must decide whether to re-administer, adjust the score, or keep it, and must document the decision for quality review.
1. What is the valid interpretation of B's low participation?
Read the worked answer
B's low turn count is not direct evidence of low fluency. In a paired format, participation is co-constructed: A's 70% share and interruptions reduce B's opportunities. The valid question is whether B's eight turns provide enough evidence of the constructs in the rubric: accuracy, coherence, interaction, and fluency.
Eight short turns may be enough for grammar and coherence, but not for fluency if the rubric requires extended speech. The examiner should use turn-level evidence, not proportional talking time. The 25-second inaudible segment is a separate threat because it removes evidence from B's longest turn. Neither examiner should infer ability from dominance alone.
2. Recommend a decision among re-administration, score adjustment, and keeping the score.
Read the worked answer
Re-administration is the cleanest remedy if the test is high-stakes: give B a new paired task with a different partner and a working microphone, and use a fresh pair of examiners or at least the same rubric. A score adjustment upward is not defensible without evidence of what was lost.
Keeping the score is defensible only if the examiner notes and the audible portions clearly support the same trait profile; even then, document the recording fault. Because the inaudible segment covered B's longest turn, the evidence is incomplete. The decision should be recorded as a technical irregularity, not as a favor to the candidate.
3. What changes to test design and examiner training would reduce this problem in future sessions?
Read the worked answer
Train examiners to score individual performance from turn-level evidence, not from airtime. Set a participation protocol: the examiner can redirect a dominant candidate, and the rubric should state how interaction is credited.
For test design, use individual prompts within paired tasks, check microphones before each session, and keep a backup recorder. If paired format remains, collect evidence from at least two tasks per candidate. Quality review should track score differences between examiners and flag sessions with technical faults or extreme participation imbalance.