This dissertation outlines three studies investigating linguistically diverse students’ (including bilingual and monolingual students) language use and self-efficacy in writing. Bilingual students are those who speak a language other than English, including students with various levels of English proficiency and language experience. The first two studies (Study 1 and Study 2) examine the relations of language use features to writing performance, while the third study examines which dimensions of writing self-efficacy are more crucial for educators to target. The Direct and Indirect Effects Model of Writing (DIEW; Kim & Graham, 2022) and the Writer(s)-Within-Community serve as the guiding theoretical frameworks for this dissertation. According to DIEW and WWC, many linguistic, socioemotional, and cognitive skills and knowledge are needed to write proficiently. The present dissertation focuses on language skills and self-efficacy in writing. Study 1 is a meta-analysis that aims to investigate the relations between syntactic features and writing performance systematically. Previous studies have inconsistent results regarding the relation between syntactic complexity and writing quality and the moderating roles of measurement (e.g., genre, syntactic complexity measures) and writers’ (i.e., language proficiency, age) in the relation. The results show a weak relation between syntactic features and writing performance, and the relation can be moderated by measurement and writers’ characteristics. Study 2 examines the predictive accuracy of machine learning in predicting multiple dimensions of writing quality (structure, evidence use, historical thinking, and language use) in secondary history source-based argument writing. In addition, based on linguistic features that contribute to the prediction of writing quality, the study further examines language use patterns in bilingual and monolingual writers with different EL designations. Results indicated that machine learning models can explain a higher variance in writing quality scores. English Learners (EL) performed lower on lexical sophistication and global cohesion, but bilingual students designated as proficient in English performed equivalently with their monolingual peers. Using the same data as in Study 2, Study 3 examines secondary students’ writing self-efficacy. The study examines (1) in which dimension(s) students feel less efficacious in writing, (2) which dimension(s) is more predictive of students’ writing quality, and (3) to what extent students of different EL designations perform differently. Results showed that students had the lowest writing self-efficacy in self-regulation, which was the dimension that was significantly associated with writing quality. In addition, EL students and students designated as Reclassified Fluent English Proficient (RFEP) had lower writing self-efficacy in ideation and revision, whereas their writing self-efficacy in self-regulation was equivalent to students who were initially fluent in English since they entered school. These findings suggest that the text-based linguistic features can provide information about students’ language skills and areas of improvement in an authentic writing context. Students, in general, had lower writing self-efficacy in self-regulation, which is a critical skill to navigate through the writing process. Students who had the experience of being an EL may have lower writing self-efficacy in certain aspects but had equivalent writing self-efficacy in self-regulation.