Investigating the Nature of Corpus-based Variables and their Contributions to Item Difficulty

This study examines whether the two tests, the English Test for Certified Public Tax Accountants (ETCPTA) and the Test of English for International Communication (TOEIC) represent the target language use domain presented in a specific field of tax and accounting as ESP-based tests. This study compares the two tests by measuring corpus-based variables and their influence on item difficulty. The study intends to: I) examine input language through corpus analysis and study whether the corpus-based variables have correlations with item difficulty of the tests; and 2) investigate statistically significant variables and the extent to which they affect item difficulty compared to other variables. The two tests presented completely different characteristics in their input language. ETCPTA represented academically specific discourse, while TOEIC was characterized as general business-related discourse. Multiple Regression Analysis and ANOVA also confirmed that the ETCPTA item difficulty was influenced by content specific or academic vocabulary, while the TOEIC data were affected by frequently used vocabulary. In terms of authenticity of the ESP-based tests, neither ETCPTA nor TOEIC represent the target language use domain of tax and accounting. Thus, it is suggested that even more content-specific ESP-based tests be designed reflecting particular language use contexts.