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.

Consider two random variables $X$ and $Y$ with joint distribution $F_{X,Y}$ and strictly positive density function $f_{X,Y}$. Additionally, let $x^*$ be the value of $x$ that solves: $$ \Pr[Y\leq y\mid X=x^*]=1-k $$ for some constant $k \in [0,1]$.

I would like to know the minimum conditions I should impose on $F$ in order to ensure that $$\frac{\partial x^*}{\partial k}\geq 0$$

It seems quite intuitive that $COV[X,Y]>0$ would do the job, but this is not the case. I found in the literature that Affiliation* or Decreasing Inverse Hazard Rate** are sufficient, but I would like to know if there are weaker conditions.

*The pdf $f$ is affiliated if $x \leq x'$ and $y \leq y'$ imply $f(x, y')f(x',y)\leq f(x', y')f(x,y)$

** $Y$ is Inverse Hazard Rate Decreasing in $X$ if $\frac{F(y|x)}{f(y|x)}$ is non-increasing in $x$ for all $y$ (where $f(y|x)$ denotes the pdf of $Y$ conditional on $X=x$).

NOTE: This question has also been posted on the Stats Stackexchange site (link here)

share|improve this question
    
Nothing wrong with posting this here, but there is a stats site in the stackexchange group which might be a better fit. –  Gerry Myerson Jul 17 '13 at 12:46
    
@GerryMyerson: Thanks. I'll try that. Is it generally OK to post the same question in two forums? –  EOO Jul 17 '13 at 13:14
2  
It's generally NOT OK to do that. You could 1) wait to see whether you get a good answer here, or 2) delete the question here, and post to the other site, or 3) have it on both sites BUT with a note on each site linking to the post on the other site (transparency!). 4) you could (flag it and) ask a moderator to migrate it to the other site. –  Gerry Myerson Jul 17 '13 at 13:19
    
Thanks -thought so. Will wait and try one of the options later on. –  EOO Jul 17 '13 at 13:22
    
Still waiting? $ $ –  Did Nov 11 '13 at 14:54

1 Answer 1

up vote 0 down vote accepted

In the mathematical statistics literature (for example here) there is a condition termed Regression Dependence. More specifically, for two real random variables $X$ and $Y$, it is said that $Y$ is positively regression dependent on $X$ if: $$ \Pr[Y\leq y |X=x] \quad \text{is non-increasing in }x $$

(Page 1143, Some Concepts of Dependence, E. L. Lehmann. The Annals of Mathematical Statistics, Vol. 37, No. 5 (Oct., 1966), pp. 1137-1153)

This condition is precisely the desired property, namely that $$ \frac{\partial \Pr[Y\leq y |X=x]}{\partial x}\leq 0 $$

As noted on the reference provided for the question, this condition is weaker than Affiliation (and also weaker than Decreasing Inverse Hazard Rate)

share|improve this answer

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.