The COVID-19 pandemic is a global crisis that has forced governments around the world to implement large-scale interventions such as school closures and national lockdowns. Previous research has shown that partisanship plays a major role in explaining public attitudes towards these policies and beliefs about the severity of the crisis. However, the cognitive roots of this phenomenon remain poorly understood. In principle, partisan gaps in policy support could emerge from cost-benefit analyses from individuals with dissimilar perceptions about the severity of the pandemic, as proposed by rational models of partisan behavior. Alternatively, polarized responses may be driven by social identity motives that are unrelated to individual beliefs, as predicted by theories of tribal partisanship. Here, we tested the predictions of these two models across four experiments (N=1980) performed in four different countries (Argentina, Uruguay, Brazil, and the United States). Participants forecasted the number of COVID-19 deaths in their country after considering either a high or low number. Then, they rated their agreement with a series of interventions. This anchoring procedure, which experimentally induced a large variability in the forecasted number of deaths, did not modify policy preferences. Instead, we observed that partisanship independently modulated the optimism of forecasts and participants’ support for COVID-19 policies. These results, which are against the predictions of the rational partisanship model, have strong policy implications. In particular, our findings suggest that communication strategies aimed at informing the public about the severity of the pandemic will not substantially change levels of support for COVID-19 interventions.