Political expression through social media has already taken root as a form of political participation. Meanwhile, democracy seems to be facing an epidemic of incivility on social media platforms. With this background, online political incivility has recently become a growing concern in the field of political communication studies. However, it is less clear how a government's performance is linked with people's uncivil political expression on social media; investigating the existence of performance evaluation behavior through social media expression seems to be important, as it is a new form of non-institutionalized political participation. To fill this gap in the literature, the present study hypothesizes that when government performance worsens, people become frustrated and send uncivil messages to the government via social media. To test this hypothesis, the present study collected over 8 million posts on X/Twitter directed at US state governors and classified them as uncivil or not, using a neural network-based machine learning method, and examined the impact of worsening state-level COVID-19 cases on the number of uncivil posts directed at state governors. The results of the statistical analyses showed that increases in state-level COVID-19 cases led to a significantly higher number of uncivil posts against state governors. Finally, the present study discusses the implications of the findings from two perspectives: non-institutionalized political participation and the importance of elections in democracies.