Development and use of validated assessments in physics provide insight into student understanding of physics concepts and can be utilized to track student learning across time and instructional strategies. These tools can be used to gauge the efficacy of interventions designed to support students and persistence, and allow instructors to make data-based decisions about how they structure their classrooms to support student learning. Unfortunately, gaps in performance based on demographic groups can often appear in physics assessments. However, some statistical approaches can allow for identification of bias in assessment items, allowing for potential to reduce these "performance gaps." Additionally, studying how student identity influences their experiences can inform how these gaps are understood if they remain after bias within items has been addressed. This dissertation reports two distinct but complementary studies. The first study discusses development of a novel upper-division thermal physics assessment composed of coupled, multiple-response items. This study includes assessment item development and refinement; assessment implementation; and assessment validation, with an explicit focus on differential item functioning
and item response theory in addition to classical test theory approaches. The second, qualitative study focuses on how students' experiences in STEM vary based on race and gender. This includes investigations of perceptions of professor care and instruction styles; sense of belonging; and student perceptions of how race and gender impact those pursuing STEM. These studies in combination can inform practices intended to support students of all backgrounds in pursuing STEM and physics degrees.