The online actions and words of a person can reveal their political sentiments and how they may vote at the polls. For decades, the dominant strategy of determining voter sentiment on policies relied on slow and often inaccurate polling. The creation and subsequent popularity of numerous social media sites, namely Twitter, has presented an opportunity for researchers to apply machine learning models to identify voter stances towards relevant political issues. Stance detection is a sub-task of natural language processing that involves algorithmically determining the stance that a text contains towards a given topic. With recent developments in NLP models and architectures, prior researchers have successfully trained stance detection models to predict the winning candidates in national-level elections. However, the viability of stance detection towards specific policies in city-level and state-level elections is relatively unexplored. In this paper, we train a novel transformer neural network architecture that accurately classifies Twitter users' stances towards Proposition 16 of California's 2020 election. To that end, we created a novel annotated data set of all Tweets regarding Proposition 16. Because not all tweets were opinionated, we also trained a model to filter out irrelevant and neutral tweets. Ultimately, we achieved 82% overall accuracy. Moreover, our model accurately predicted the result of the Prop 16 election. Our results show that a stance detection model can provide a unique perspective to help politicians make decisions that represent their constituents.