Using a large information Bayesian VAR, we approximate the flow of information received by economic agents to investigate the effects of government spending. We document robust evidence that insufficiency of information in conventional models could explain inconsistent results across samples and identifications (Recursive Structural VAR and Expectational VAR). Furthermore, we report heterogeneous effects of government spending components. While aggregate government spending does not appear to produce a strong stimulative effect with output multiplier around 0.7, government investment components have multipliers well above unity. Also, state and local consumption, which captures investment in education and health, elicits a strong response.