0
$\begingroup$

Let's assume a probability mass function $P$ on the discrete domain $\{0,...,N\}$ and a density function $f$ and the existence of two real factors $a$ and $b$ so that we have for all numbers $k$ in $\{0,...,N\}$:

$$P(k) = a * f(k * b)$$

What can be said about the theoretical / mathematical relationship between $P$ and $f$? Can $f$ be used to understand $P$ and vice versa?

(I hope my question is not too fuzzy - in case it is, I would adjust it according to suggestions)


To add some context:

library(VGAM) #drayleigh

N <- 100
p <- function(k,N) k/N * ifelse(k>1, prod(1 - (1:(k-1)/N)), 1)
plot(1:N, sapply(1:N, p, N))

a <- 0.104
b <- 1 / 9.7
points(1:N, drayleigh(1:N * b) * a, col="red", type="l")

p defines a random process (I think it is a Markov chain) that will sample from {0,1} until 1 is chosen and then stops. The probability of choosing 1 increases linearly from 0 to 1 within N+1 steps.

enter image description here

The Rayleigh distribution is a special case of the Weibull distribution which is used for modelling mortality. The process I define above could obviously serve as a very simple model for mortality - the older you get the higher the probability of dying until an age of 100 where dying can be considered as certain.

But I am wondering whether I could f.x. infer from my simple process an intuitive idea that helps understanding what the Rayleigh distribution captures. And because I expect this situation to come up again in the future I am asking for a general answer. But of course insights into this specific context are welcome.

$\endgroup$
1
$\begingroup$

Here is one possibility (for $b\neq 0$): let your $P$ be the probability mass function of a random variable $X$ and let $Y$ have the same distribution as $bX$. Then we see that $$P(X=k)=P(bX=bk)=P(Y=bk)=af(bk)$$ where $a=1$ and $f=P$. Now if you let $a$ be any real number, you will in general not get a probability mass function on $\{0,\ldots,bN\}$ but a multiple of it, a signed measure.

Edit:

Because $P$ is a probability mass function, $af(bk)$ must be nonnegative for all $k\in\{0,\ldots,N\}$ and fulfil $\sum_{k=0}^N af(bk)=1$. So clearly $a\neq 0$. Without loss of generality, assume $a>0$ (if not, replace $f$ by $-f$). Because $a>0$ and $P$ defines a probability measure, also $f$ must be nonnegative. We now find that $a = 1/(\sum_{k=0}^N f(bk))$. Now define the function $g$ on $\{0,b,\ldots,bN\}$ via $g(bk)=af(bk)$. Then $g$ is the probability mass function of $bY$. So you see that $b$ scales the state space of the random variable $X$ while $a$ may be any positive number that just scales the total mass of $f$, with the total mass of $af$ being one.

$\endgroup$
3
  • $\begingroup$ (+1) the idea is a tad basic, but of course, why not $\endgroup$ – Raffael Apr 11 '14 at 12:32
  • $\begingroup$ I am sure more clever persons will give better answers :) $\endgroup$ – binkyhorse Apr 11 '14 at 12:34
  • $\begingroup$ rescaling the domain offers a different way to look at it - but I currently can't see how this yields to a new insight. maybe a more clever person can tell me how $\endgroup$ – Raffael Apr 11 '14 at 12:36

Your Answer

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

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