Microfeatures influencing writing quality: the case of Chinese students’ SAT essays

Calculating the Relative Importance of Multiple Regression Predictor Variables Using Dominance Analysis and Random Forests

“…Variable Importance Plot for the Data from Goh et al (2020) The dominance weight of Commas was close to zero in magnitude, and its importance as a predictor was trivial. The problem of Commas being overrated as an important predictor was simply because Goh et al (2020) interpreted its importance using its standardized beta coefficient. As witnessed in this empirical example, one cannot trust the face value of the standardized beta coefficients as a measure of predictor importance.…”

“…Table 4 provides the correlations and standardized beta coefficients reported by Goh et al (2020). Based on these results, Goh et al argued that three length-related features (i.e., word count, the Coleman-Liau readability index, and the number of words per sentence) were the top three most important variables for predicting the SAT writing score.…”

“…Therefore, Goh et al suggested that students should limit their use of stop words to the minimum level required. Goh et al (2020) employed stepwise multiple regression analysis because their primary purpose was to build the best regression model with the highest predictive power possible. This method is also known as "statistical" regression analysis.…”

“…Considering the several significant problems inherent in the use of stepwise regression analysis, Nathans et al (2012) advised against its use for assessing variable importance. Accordingly, standard multiple regression was used to reproduce the analysis of Goh et al (2020); then, DA was applied to these data. (Goh et al, 2020).…”

“…Accordingly, standard multiple regression was used to reproduce the analysis of Goh et al (2020); then, DA was applied to these data. (Goh et al, 2020). R 2 and β were recalculated using standard multiple regression.…”