The principle of indifference suggests without prior knowledge the [[prior]] distribution should be a [[uniform distribution]]. For discrete cases, the prior probability for each of $n$ hypotheses should be $1/n$. For continuous cases where $\theta \in (a, b)$, the prior probability for any subinterval $I = (c,d)$ should be $P(I) = \frac{d-c}{b-a}$ While uninformative priors may sound appealing as they appear to answer the frequentist objection to the prior, in fact the outcome of a Bayesian analysis may depend on how the parameters of the problem are considered, or what type of ignorance we claim. For example, if we wanted to estimate the area of a square, we could focus on either the area $A$ or the length of a side $S$. Because $A=S^2$, it is not possible to apply a uniform prior distribution to both $A$ and $S$ at the same time. For this reason, the uninformative prior is not invariant to different parameterizations of the problem. The [[objective prior]] is a response to this critique.