The prior distribution is meant to reflect our knowledge or degree of belief in the value of $\theta$ before any new data are observed (e.g., expert opinion, prior knowledge). The prior should be independent of the likelihood, which means that one should not use the data to inform the prior. Two general classes of prior exist - **Uninformative priors** assign equal probability to all hypotheses. Uninformative priors are based on the [[principle of indifference]]. - **Informative priors** incorporate background information. In some cases, Bayesian inference can be simplified when the prior and likelihood is part of a [[conjugate family]]. [[prior predictive distribution]]