This study aimed to identify second language (L2) developmental indices while controlling for the effects of first language (L1). More specifically, this study investigated the differences in the use of metadiscourse markers among learners from different L1 backgrounds. The following research questions were explored: (1) Which metadiscourse markers can be used as developmental indices to predict learners' proficiency levels? (2) How strongly do learners' L1s impact their L2 development? To answer these questions, multilevel ordinal logistic regression analysis was performed comparing the metadiscourse in English essays written by three Asian learner groups (i.e., Japanese, Thai, and Taiwanese). The results suggest that the frequencies of transitions and self-mentions are the best predictors of language development of metadiscourse categories. Additionally, this study showed that the multilevel regression model can measure the significance of each metadiscourse category more accurately than the single-level model. Therefore, we must consider the influence of learners' L1s for precise detection of the linguistic features that predict their proficiency levels. Through this application of multilevel ordinal regression analysis to learner corpus research, this study illustrated the effectiveness of multilevel analysis for tracking the language acquisition process.