Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Suppose that a random variable $X \in [0,1]$ is drawn according to the density $f(x|\theta)$ conditional on the realization of $\theta \in [0,1]$. From $x$, we can generate $c$ according to some rule $c=g(x)$. We have know that $g$ is one-to-one. This procedure will give a distribution of $c$ given a realization of $\theta$. Now consider a median of $c$ and how the median of $c$ changes when $\theta$ changes. Under what condition on $f$ can it be that the median of $c$ is distinct for each $\theta$ (or at least there is only a measure zero set of $\theta$ that will give the same median of $c$.

share|improve this question

1 Answer 1

For every $\theta$, the median is $g(y_\theta)$ where $y=y_\theta$ solves the equation $$ \displaystyle\int_0^yf(x|\theta)\mathrm{d}x=\frac12. $$ Introducing a random variable $X_\theta$ with density $f(\ |\theta)$, this reads as $P(X_\theta\le y_\theta)=\frac12$. A way to ensure that $\theta\mapsto g(y_\theta)$ is one-to-one is to ask that $g$ is strictly monotonic, say increasing, and that $\theta\mapsto y_\theta$ is increasing as well. One can ensure this last property by requiring that the family $(X_\theta)$ is stochastically increasing.

Recall that one says that two random variables $U$ and $V$ are such that $V$ stochastically dominates $U$ if $P(V\le w)\le P(U\le w)$ for every $w$. In words, $V$ is small less often than $U$ is, so this relation is indeed a way of specifying that $V$ is larger than $U$, in some sense.

Coming back to your setting, one asks that for every parameters $\theta<\lambda$ and every $x$ in $(0,1)$, $P(X_\lambda\le x)<P(X_\theta\le x)$ (note the strict inequality signs and that one excludes the cases $x=0$ and $x=1$). A way to know this holds is to build some random variables $\tilde X_\theta$ and $\tilde X_\lambda$ on the same probability space such that $\tilde X_\theta$ is distributed like $X_\theta$, $\tilde X_\lambda$ is distributed like $X_\lambda$ and $\tilde X_\theta <\tilde X_\lambda$ with probability $1$.

Reference: the first chapters of the book Lectures on the Coupling Method by Torgny Lindvall.

share|improve this answer
    
Thank you. But I need this property to hold not only for a monotone $g$ but also for every $g$. On the other hand, it does not have to be that median is increasing in $\theta$. –  Thales Apr 3 '11 at 5:51

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.