Describing and measuring instructional quality of mathematics lessons is a common goal amongst mathematics education researchers. Such work takes several forms such as classifying and coding instructional moves and student activity or providing high-level rubric-based scores in relation to categories. In this work, we share an innovative mixed methods approach to analyzing lesson data that includes both a time-based classification of instruction and an overall scoring component. Using the Math Habits framework, our project team analyzed a set of 97 fourth-eighth grade mathematics lessons including overall scores. From this qualitative analysis, we developed quantitative models to predict overall scores and better understand the ways that individual codes do or do not contribute to overall lesson score characterizations. [For the complete proceedings, see ED630210.]