This dissertation selects two impactful factors that have been validated by previous research - discussion heterogeneity and incivility - as predictors to examine their effects on opinion expression, political polarization and response incivility. It fills a research gap by comparing 2 types of heterogeneity and homogeneity. It integrates group polarization theory in social psychology with empirical studies in communication by drawing analogy between oppositional discussion setting with outgroup interaction, and between homogeneous discussion setting with ingroup interaction. It also predicts that opinion expression leads to attitude polarization and justifies the prediction with cognitive dissonance theory.The study combines survey and experiment methods. The experiment selects Medicare-for-All as topic of interest and employs a 2 (incivility: civil vs uncivil) Ã 3 (heterogeneity of discussion setting: homogeneous, oppositional and mixed) posttest only control group design (N=528). It finds that individuals’ political position moderates the effect of discussion heterogeneity on polarization, and that opinion expression elicits polarization. However, effects of incivility were not found to be statistically significant.This study manifests decent validity of observed measure of actual opinion expression and self-report measure of expression willingness by triangulation. It also advances affective polarization measure by including thermometer ratings towards people of opposite opinions on a specific issue. Its findings also advance the cognitive dissonance theory and group polarization theory in communication discipline. Furthermore, the observed data seems to suggest a “spiral of noise” mechanism - people are most likely to join uncivil and mixed-opinion discussion, and with their opinion expression intensifying their own attitude and affective polarization and triggering others’ political expression and polarization, the dynamics that exposure to and expression of extreme opinions reinforce each other recur in turns and affect wider populace. Directions for future study are suggested.