Convincing research designs make empirical economics credible. To avoid confounding, quasi-experimental studies focus on specific sources of variation. This could lead to a reduction in statistical power. Yet, published estimates can overestimate true effects sizes when power is low. Using fake data simulations, we show that for all causal inference methods, there could be a trade-off between confounding and exaggerating true effect sizes due to a loss in power. We then discuss how reporting power calculations could help address this issue.