 Solution

For fixed $\theta >0$, the central limit theorem givesThe weak law of large numbers gives $\overline{X}\to \theta$, so Slutsky theorem permits replacing $\theta$ in the standard error by $\overline{X}$. With $z=z_{0.025}\approx 1.96$, an approximate 95% confidence interval isThe lower endpoint may be truncated at zero because the parameter is nonnegative. This is a large-sample normal approximation for fixed positive $\theta$; an all-zero or very sparse sample does not justify the plug-in normal approximation. At the degenerate parameter value zero, every observation is zero almost surely.