Given a sample of size $n$ from a $N(\mu, \sigma^2)$ population, the Fisher Information $I(\mu)$ for the parameter $\mu$ is constant. If we want to construct a $100(1-\alpha)\%$ confidence interval for $\mu$, the width of the interval is proportional to: