Rater severity differences in English language as a second language speaking assessment based on rating experience, training experience, and teaching experience through many-faceted Rasch measurement analysis

Abstract

Evaluating candidates' answers in speaking skill is difficult and rarely explored. This task is challenging and can bring inconsistency in the rating quality among raters, especially in speaking assessments. Severe raters will bring more harm than good to the results that candidates receive. Many-faceted Rasch measurement (MFRM) was used to explore the differences in teachers' rating severity based on their rating experience, training experience, and teaching experience. The research uses a quantitative approach and a survey method to enlist 164 English teachers who teach lower secondary school pupils, who were chosen through a multistage clustered sampling procedure. All the facets involving teachers, candidates, items, and domains were calibrated using MFRM. Every teacher scored six candidates' responses in a speaking test consisting of three question items, and they were evaluated across three domains, namely vocabulary, grammar, and communicative competence. Results highlight that the rating quality was different in terms of teachers' rating experience and teaching experience. However, training experience did not bring any difference to teachers' rating quality on speaking test. The evidence from this study suggests that the two main factors of teaching and rating experience must be considered when appointing raters for the speaking test. The quality of training must be improved to produce a rater with good professional judgment. Raters need to be supplied with answer samples with varied levels of candidates' performance to practice before becoming a good rater. Further research might explore any other rater bias that may impact the psychological well-being of certain groups of students.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 25 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.