Problem from Bayesian Data Analysis (Bayesian interpretation of non-Bayesian estimates): Consider the following estimation procedure, which is based on classical hypothesis testing. A matched pairs experiment is done, and the differences $y_1, ... , y_n$ are recorded and modeled as independent draws from $\text{N}(\theta, \sigma^2)$ where $\sigma$ is known. The parameter $\theta$ is estimated as the average observed difference if it is ‘statistically significant’ and zero otherwise:

$$\hat{\theta} \equiv \begin{cases} \bar{y} & & |\bar{y}| \geqslant z_{0.025} \cdot \sigma / \sqrt{n}, \\[6pt] 0 & & \text{otherwise}, \end{cases}$$

where $z_{0.025} \approx 1.96$ is the normal quantile for a 5% significant test. I am hoping to find a prior distribution for $\theta$ that gives this estimator as a Bayesian estimator. I was hoping that an improper prior parameterised by $\alpha>0$ with the following form would work:

$$\pi(\theta) \propto \begin{cases} 1+\alpha & & \theta = 0, \\[6pt] 1 & & \theta \neq 0. \end{cases}$$

My hope was that the posterior would be a normal distribution centered at $\bar{y}$ with variance $\sigma^2 / n$ except at $\theta = 0$ where there would be a probability spike. Depending on the parameter $\alpha$, the posterior mode would be either at 0 or $\bar{y}$. Would appreciate feedback.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.