This wiki page says

A continuous probability distribution is a probability distribution with a cumulative distribution function that is absolutely continuous. Equivalently, it is a probability distribution on the real numbers that is absolutely continuous with respect to Lebesgue measure. Such distributions can be represented by their probability density functions. If the distribution of X is continuous, then X is called a continuous random variable. There are many examples of continuous probability distributions: normal, uniform, chi-squared, and others.

It seems that the first sentence in the quotation implies that there exists some other types of probability distribution that could not be associated with a cumulative distribution function?

Is my understanding right, or I am over reading the part?

  • $\begingroup$ All random variables $X$ have a cumulative distribution function (CDF) defined by $$F_X(x) = P[X\leq x] \quad \forall x \in \mathbb{R}$$ The CDF function is not always continuous. I would say a random variable $X$ is "continuous type" if its CDF function is continuous. I would not complexify by requiring an "absolutely continuous" CDF. The advantage of adding the "absolutely" is that it allows you to also have a PDF. $\endgroup$ – Michael Oct 13 '19 at 4:36

Every probability distribution on $\mathbb R$ is associated to a cumulative distribution function (and every non-decreasing function $F$ with $\inf_x F(x)=0$ and $\sup_x F(x)=1$ is associated to a distribution!). The quoted text is probably using "with" to refer to the whole phrase "a cumulative distribution function that is absolutely continuous."

For the fact that every probability distribution does define a cumulative distribution function, just note that defining, for a probability measure $\mu$ on $\mathbb R$, the CDF as $$F(x)=\mu((-\infty,x))$$ gives a perfectly good CDF - although there's a little bit of ambiguity about what you should do at point masses where $(-\infty,x)$ and $(-\infty,x]$ have different measure - but this depends on what you want from a CDF and is a very manageable caveat.

Conversely, given a non-decreasing function on $\mathbb R$, you can define a measure with the property that $$\mu((a,b))=\left(\lim_{x\rightarrow b^-}F(x)\right) - \left(\lim_{x\rightarrow a^+}F(x)\right).$$ I won't give all the technical details of this construction, but this measure is called the Lebesgue-Stieltjes measure associated to $F$.

There's a slight issue in this association having to do with point-masses, but you can fix that by imposing conditions like right-continuity on $F$ which basically decide whether $F(x)$ includes a point mass at $x$ or not.

  • 1
    $\begingroup$ Can I convince you to use $F$ instead of $f$? If so, I will +1 vote! (Just to keep convention of usign $F_X(x)$ to denote CDF and $f_X(x)$ to denote PDF (when it exists)). $\endgroup$ – Michael Oct 13 '19 at 4:17
  • 1
    $\begingroup$ @Michael Sure - I'm not well aware of conventions with CDF and PDF, so I'll take your word for it. $\endgroup$ – Milo Brandt Oct 13 '19 at 4:18

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.