Research Review: How Effective Is Utah Compose in Different Instructional Contexts?

Utah Compose provides a helpful framework to aid writing instruction. When implemented with an instructional focus on goal-setting and strategy use, results of experimental studies have shown improved writing quality over time. Additionally, there have been positive changes in students’ self-regulation, though their general attitudes and beliefs about writing remained unchanged. Stronger results have been observed in instructional contexts centered on building higher-level thinking skills and knowledge of writing in specific genres. An emphasis on goals and strategy use helps students develop good habits and control, while emphasis on surface features of writing results in less improvement. 

The Key Findings that follow include reference to MI Write. Utah Compose uses the same technology infrastructure as MI Write and includes the same features and functionality.

Key Findings

Added instructional support integrated with MI Write helps students improve overall writing quality.

•    Grade 5–8 students who used MI Write within an instructional context that included monthly writing practice and goal-setting activities showed significant improvements in argumentative writing (pretest to posttest) and in self-regulation (ability to set, plan, and achieve goals). [3]
•    Students who used MI Write combined with Self-Regulated Strategy Development (SRSD) instruction showed the most more improvement between pretest and posttest essays than students in two other instructional conditions (business-as-usual or teacher instruction combined with MI Write). Using strategies for argumentative writing, students in the SRSD condition produced essays that were longer and included more genre-specific elements. [1]
•    Traditional writing instruction paired with different conditions for teacher feedback, given either through MI Write or Google Docs frameworks, showed no differences in student writing quality from pretest to posttest in a study of Grade 6 students, but teachers gave different types of feedback. With automated feedback, teachers gave more higher-level feedback on content, while teachers who gave feedback through Google Docs tended to focus more on corrections and grammar. [4]

Implementation of MI Write is less effective in educational contexts lacking a standard curriculum or rigorous core instruction.

•    In a large school without a standardized ELA curriculum, writing performance of Grade 3–5 students showed uneven improvement from fall to spring: Grade 5 students showed no changes, while Grade 3 students improved across all genres and Grade 4 students improved in narrative writing. [3]
•    Students must be given enough opportunities to practice and receive timely feedback. More “distributed practice” vs than “massed practice” is recommended for most effective implementation in writing curriculum and to “transform writing outcomes.” [2]
•    Available tech resources, professional training, and instructional time are important factors to consider for effective implementation of MI Write. [2]



1.    Palermo, C., & Thomson, M. M. (2018). Teacher implementation of self-regulated strategy development with an automated writing evaluation system: Effects on the argumentative writing performance of middle school students. Contemporary Educational Psychology, 54, 255–270.

2.    Wilson, J., Huang, Y., Palermo, C., Beard, G., & MacArthur, C. A. (2021). Automated feedback and automated scoring in the elementary grades: Usage, attitudes, and associations with writing outcomes in a districtwide implementation of MI Write. International Journal of Artificial Intelligence in Education, 31, 234–276.

3.    Wilson, J., Potter, A., Cordero, T. C., & Myers, M. C. (2022). Integrating goal-setting and automated feedback to improve writing outcomes: A pilot study. Innovation in Language Learning and Teaching, 17, 518–534.

4.    Wilson, J., & Roscoe, R. D. (2020). Automated writing evaluation and feedback: Multiple metrics of efficacy. Journal of Educational Computing Research, 58(1), 87–125.