2
$\begingroup$

enter image description here

Hello. I inserted an image because it's easier to understand my questions.

I don't understand how pdf works. If I say I want te probability of X=1, it returns me 0,667. But what's that? Is it the whole area under the triangle of base 1 and height 0,667? So why doesn't A=b*h/2 equal to 0,667? A=2/6=1/3 ;

Also, if I do the integral of the function (2/3)*x shouldn't it give me the same result as f(1) = 2/3 ? Since the fdp says that the area below the graph is the probability of X=k..

Lets say my random variable is the kilograms of rice that people buy in a store. The probability o X=1 kilogram is zero, because the area under the graph is zero. So it makes more sense to say we want P(0

$\endgroup$
  • $\begingroup$ By definition, $\mathbb{P}[a\leq X\leq b]=\int_a^bf_X(u)\,du$. In your case, $\mathbb{P}[X=1]=0$ because $X$ is a continuous random variable, i.e. $\mathbb{P}[X=1]=\mathbb{P}[1\leq X\leq 1]=\int_1^1f_X(u)\,du=0$. The key point is that there is a difference between $\mathbb{P}[X=c]$ and $f_X(c)$. The former is always zero when $X$ is a continuous random variable; the latter need not be zero (but could be). The function $f_X(u)$ is a probability density, not a probability. $\endgroup$ – symplectomorphic Apr 3 '17 at 0:57
  • $\begingroup$ The intuitive idea is that $f_X(x) \Delta x$ tells you (approximately) the probability that $X$ is in the range $[x,x+\Delta x]$. (Here $\Delta x$ is a tiny positive number. The approximation is good when $\Delta x$ is very small.) $\endgroup$ – littleO Apr 3 '17 at 1:15
1
$\begingroup$

In this example, you can think of the area under the PDF between two limits (say, between $1$ and $3$) as the probability that the random variable falls between those two limits. Here, a triangle of base $2$ and height $\frac23$ yields an area, and thus a probability, of $\frac23$.

More generally, the height of the PDF represents a kind of probability rate: the variable falls within an infinitesimal range around $1$ of width $dx$ with an infinitesimal probability of $\frac23 dx$. Around $2$, where the height of the PDF is $\frac13$, the rate is $\frac13 dx$ per $dx$ width of interval. And so on. That is why the integral of the PDF produces a probability. Note that this means the integral (and therefore the area) over the entire domain of the random variable must be $1$—the probabilities of all possibilities add up to $1$, in other words.

$\endgroup$
  • $\begingroup$ Why does it has to do with limits? Like P(a<X<b). And what would mean f(1)? Is it only to know the image on y axis? $\endgroup$ – Vitor Aguiar Apr 3 '17 at 1:39
1
$\begingroup$

The probability that the random variable is between two numbers on the $x$-axis is the area under the graph of the density function above the part of the $x$-axis that is between those numbers. For example, look at the part of the graph between $x=2$ and $x=3.$ The area under the graph between those points is a triangle whose base has length $1$ and whose height is $1/3$, so whose area is $1/6.$ Thus the probability that the random variable is between $2$ and $3$ is $1/6.$

$\endgroup$
  • $\begingroup$ Yes, but what does it mean f(1) =? $\endgroup$ – Vitor Aguiar Apr 3 '17 at 1:36
  • $\begingroup$ The fact that $f(1)=2/3$ means for small intervals centered at $1$, the probability that the random variable is in the interval can be made as close as desired to $2/3$ the length of the interval, by making the interval short enough. $\endgroup$ – Michael Hardy Apr 3 '17 at 2:02
1
$\begingroup$

For a "continuous" (non-discrete) random variable the probability for any individual outcome is zero. The pdf does not represent probabilities, it is a non-negative function that integrates to $1$ in it domain, that is

$$\int_{-\infty}^\infty f_X(x)\mathrm dx=1$$

For a continuous random variable their probabilities are determined by areas under it pdf... by example

$$\Pr[0\le X\le 1]=\int_0^1f_X(x)\mathrm dx =\frac{x^2}3\bigg|_0^1=\frac13$$

or

$$\Pr[X\ge 1/2]=\int_{1/2}^\infty f(x)\mathrm dx=\int_{1/2}^1\frac{2x}3\mathrm dx+\int_1^3\left(1-\frac{x}3\right)\mathrm dx=\frac{x^2}3\bigg|_{1/2}^1+\left(x-\frac{x^2}6\right)\bigg|_1^3=\\=\frac13-\frac1{12}+\left(3-\frac32\right)-\left(1-\frac16\right)=\frac3{12}+\frac32-\frac56=\frac{11}{12}$$

obviously we have that $\Pr[X\le1/2]=1/12$. In the same way for any individual $x=a$ we have that

$$\Pr[X=a]=\int_a^a f_X(x)\mathrm dx=0$$

$\endgroup$
  • $\begingroup$ If I say I want the probability of X=1, it would be zero, as you said. I understand. It is right to say P(0<X<1)... so what does means f(1) = 2/3 ? $\endgroup$ – Vitor Aguiar Apr 3 '17 at 1:16
  • $\begingroup$ to be honest, I dont know it meaning Vitor. I read somewhere that it represent the "density" of the random variable at this point, by analogy with the physical concept of density, but I dont remember something more. If we have a physical analogy with pressure then the pressure in an infinitesimal point is zero but in some area it can have some positive value. Probably is the same idea here. $\endgroup$ – Masacroso Apr 3 '17 at 1:19
  • $\begingroup$ @VitorAguiar $f(1)=2/3$ basically means that $\lim_{\Delta x \to 0^+} \frac{P(X \in [1-\Delta x,1+\Delta x])}{2\Delta x}=2/3$. This division by $2\Delta x$ is why a pdf measures density (probability per length) rather than mass (probability, period). A caveat is that this may not be true at a point where $f$ is not continuous (for example, at $x=0$ for the exponential distribution). $\endgroup$ – Ian Apr 3 '17 at 1:33
  • $\begingroup$ @Ian could you answer it on an answer, to understand the code? $\endgroup$ – Vitor Aguiar Apr 3 '17 at 1:46

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.