I want to write:

$$P(y_{1}<Y\le y_{2}\mid Y\sim\mathcal{D}(\mu))$$

to say:

  • The probability of $Y$ being between $y_1$ and $y_2$
  • given that $Y$ is a random variable distributed according to distribution $\mathcal{D}$
  • when $\mathcal{D}$ is parameter by $\mu$

I want to leave the $\mathcal{D}$ in because $\mathcal{D}$ could be one of a number of different distributions, all of which are parameterised the same way.

but I've never seen a "distributed as" ($\sim$) on the right hand side of "given that" ($\mid$)

I was considering other notations:

  • $$P_{Y\sim\mathcal{D}(\mu)}(y_{1}<Y\le y_{2}\mid \mu)$$
  • $$P(y_{1}<\mathcal{D}(\mu)\le y_{2})$$
  • $$F_{\mathcal{D}(\mu)}(y_2) - F_{\mathcal{D}(\mu)}(y_1)$$

I want to use this as part of a numerical procedure to describe a way that a continuous space point is mapped to a probability mass vector: E.g. something like a function $\Omega$

$$ \Omega: \mathbb{R}\times\mathbb{D} \to \mathbb{R}^{100} $$ For $\mathbb{D}$ the space of 1-D continuous distributions parameterised by 1 scalar $$ \Omega(v,\mathcal{D})=\left(P\left(\dfrac{i-1}{100}<Y\le\dfrac{i}{100}\mid Y\sim\mathcal{D}(v)\right)\right)_{i=1}^{i=100} $$

So this is basically mapping a point $v$ and distribution $\mathcal{D}$ to histograms of that distribution when the first paramater is given by $v$

The random variable $Y$ is itself without real meaning -- just a step in describing the process

  • $\begingroup$ It's a slight abuse of notation, because you are not actually conditioning on anything, but people will understand easily what you mean. $\endgroup$ – Gordon Smyth Jun 30 '17 at 4:47
  • $\begingroup$ I deleted my post...your description of $\Omega(v,\mathcal{D})$ was pretty clear. However, where did $10$ come from in the second parameter? $\endgroup$ – user145807 Jun 30 '17 at 5:13
  • $\begingroup$ In this case it is an arbitrary value for the example, I'll just remove it entirely. $\endgroup$ – Lyndon White Jun 30 '17 at 5:15
  • $\begingroup$ I'd say if you use your "conditional" notation after explaining your mapping, then I'd think I could follow the rest of your argument...but by itself it's not self explanatory. $\endgroup$ – user145807 Jun 30 '17 at 5:28

This is an abuse of notation. How would you calculate $P(\,Y\sim\mathcal{D}(\mu)\,)$ ? $Y\sim\mathcal{D}$ is a definition that states: $Y$ is a random variable that is distributed as $\mathcal{D}$. If you say that $Y$ can follow different probability distributions, then what is the probability distribution of $Y$? For it to be a random variable it needs to have a probability distribution, so it cannot have multiple probability distributions.

  • $\begingroup$ What notation do you suggest instead? Your example of $P(Y \sim D)$ makes it clear that using this notation conflicts with considering it as a likihood function $\endgroup$ – Lyndon White Jun 30 '17 at 8:18
  • $\begingroup$ I have added some of the other notations I was considering to the question. $\endgroup$ – Lyndon White Jun 30 '17 at 8:27
  • 1
    $\begingroup$ @LyndonWhite but for the reasons outlined above: $Y$ cannot follow multiple probability distributions at the same time! $\endgroup$ – Tim Jun 30 '17 at 8:39
  • $\begingroup$ At the same time it can not. But the function $\Omega$ in my example is a reasonable function from the space of pairs of reals and distributions to the space of n-vectors of reals. $\endgroup$ – Lyndon White Jun 30 '17 at 8:51
  • $\begingroup$ Thing is $Y$ is not a random variable that exists outside of the definition of the function. It is not the distribution of something in a population. It is I guess a way of expressing evaluating CDFs in various sections $\endgroup$ – Lyndon White Jun 30 '17 at 9:02

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.