0
$\begingroup$

I have a table of values and their probabilities from a continuous distribution, for example say a normal distribution. What would be a good way to manually draw numbers from this distribution?

Imagine I just have this piece of paper with a table of values and I can use anything non-electronic to generate numbers like dice. Are there any methods I can use to generate numbers that follow this distribution?

  • 0
    To get you on the right track: Assume I have a table with the values $P(X=1) = \frac{1}{4}, P(X=2) = \frac{1}{4}, P(X=3) = \frac{1}{2}$ and I have to use a fair coin to draw samples from this distribution. How should one proceed? – 2017-02-13
  • 0
    Good start. I think for this, I would flip the coin twice and denote HH = 1, HT = 2, and TH and TT = 3. But if the distribution was continuous... – 2017-02-13
  • 0
    Looks good. Note that you can do with the assignment $T \to \{X = 3\}$ if you want to save some coin flips. As for drawing from a continuous distribution: For all practical purposes, you'll have to subdivide your sample space into a (finite) countable one and sample from the latter by the previously proposed idea. Of course, the more fine-grained your division is, the closer you'll be to the original distribution. In general the issue is due to the fact, that for continuous random variables we have $P(X = x) = \int_x^x f(z) \, dz = 0$. – 2017-02-13

1 Answers 1

1

If the distribution is continuous, you can flip a coin many times, since by the CLT this will approach a normal distribution. You can then relate this normal distribution to your own.