Unique and shared roles of the LLAMA subtests for prediction of initial L2 achievement: An application of regression commonality analysis

Abstract

Little research has examined the relations of the LLAMA subtests beyond predictive correlations and simple regressions. In this secondary analysis of data from Bokander (2020), we use regression commonality analyses (RCA) to address multicollinearity by decomposing the LLAMA predictive variance into unique components for each subtest alone and for each possible subtest combination. Fifty-five students with Germanic L1 backgrounds completed the LLAMA, followed by an introductory Swedish course, and then a written C-test. LLAMA-D, sound-sequence recognition, was the most important unique predictor of L2 achievement. LLAMA-E (sound-symbol association) unique variance and shared variance with LLAMA-D and LLAMA-B (vocabulary learning) was the next most important contributor to prediction. Similar to results for MLAT, these results demonstrate the major role of phonetic script/sound-symbol relationship skills both uniquely and shared with other subtests. The most important difference is the equally important, distinct role of speech sound-sequence recognition, a skill not previously included in aptitude tests prior to the LLAMA. The paper concludes with a discussion of the strengths and limitations of regression commonality analysis, which appears to have considerable usefulness for studies involving prediction.