I study a dispersed information economy in which agents choose how much attention to pay to macroeconomic events. I show that under certain conditions, attention and four widely studied measures of uncertainty are countercyclical: agents pay attention when they expect the economy to be in a bad state, and this increase in attention leads to higher (i) conditional volatility of aggregate output, (ii) dispersion of individual output, (iii) forecast dispersion about aggregate output, and (iv) forecast uncertainty about aggregate output. As agents pay attention, they react more to an event and their aggregate response generates high volatility. Because information is dispersed, agents’ beliefs and reactions diverge and each agent faces higher uncertainty about others’ aggregate response. All these implications are consistent with data. I evaluate the mechanism quantitatively in a dynamic dispersed information economy calibrated to U.S. forecast-survey data. In the calibrated economy, countercyclical attention generates countercyclical fluctuations in all four measures of uncertainty with cyclicality, magnitude, and persistence consistent with untargeted moments in the data. The analysis of the dynamic dispersed information economy requires a new solution method. Due to dispersed information, the economy features an “infinite regress problem” under which the equilibrium lacks a finite recursive state space. Existing methods addressing the problem are constrained to first-order approximations. These methods cannot capture attention and uncertainty fluctuations because these fluctuations are higher-order dynamics of the model. I develop a higher-order approximation method for dispersed information economies based on perturbation techniques to capture higher-order dynamics.