engleski

Analyzing different regression methods: the case of predicting text comprehension

College-level text comprehension is a complex cognitive skill especially for junior undergraduates as novices to thefield with little experience in academic reading. Comprehension becomes even more demanding in the international context when foreign language (FL) texts are in question and if readers still have difficulties with grammar or vocabulary. It is therefore interesting to analyse students' comprehension at an early stage of their studies to help find a remedial course of action and remove possible weaknesses which may obstruct academic progress. Methods of statistical regression may be used in such analysis to predict text comprehension taking into account factors such as language competence (e.g. general language proficiency, text-specific vocabulary), prior knowledge of the topic, metacognition or gender. We analyse the use of several regression methods for selecting text comprehension predictor variables: univariate and multivariate (hierarchical, stepwise, and regularization methods - lasso and ridge). The analysis is based on the data on factors of comprehension collected in a business school. Text comprehension was measured by the quality of notes (outlines) business students made while reading a business text in FL (English). Our results show similarities and differences between the methods used. All the regression methods point to the same variables of importance (gender, text-specific vocabulary, prior knowledge of the topic), but different effects are observed depending on the method applied

hierarchical regression, regularization regression methods, stepwise regression, text comprehension

nije evidentirano