Heya got a small doubt here

most of the discrete random variables utilize simple examples of situations like 3 coin toss or 2 dice rolls or girl/boys born. These examples are simple in the sense that we can chart out the entire sample space and discrete values of the random variable. We can then have a complete frequency or probability distribution.

but what happens in situations when it is not possible to prepare the complete sample space. say I note down childbirths in a hospital for 30 days. Can I use this sample and convert it to proportions and give out the probability of no of childbirths on a given date? (stupid example I know)


Hey hopefully I'm understanding your question correctly and this helps. It sounds like this is a general theory question about how to model a real-world event when we don't have access to the full population. This is the case most of the time, which is why we rely on probability distributions.

The long answer:

In your example, you want to predict either the number of childbirths on a given date or the probability of a certain number of childbirths on that date. The simplest way (that I can think of) to model this would be to assume that the number of childbirths a day follows a Binomial distribution:

i.e. Let X be the random variable "the number of childbirths a day", following a Binomial distribution with unknown n and p parameters: X ~ Binomial(n, p)

The question is how to estimate the values n (number of women who could give birth a day) and p (probability of giving birth that day). This is known as statistical inference and there are many available frequentist and Bayesian methods for performing this.

The short answer:

We cannot generalise a small sample to a population easily but we can make an assumption about which distribution the model may follow and then use statistical inference to estimate the shape of the distribution.

  • $\begingroup$ hi, yes got the problem right. Can you please explain on what basis did you select that the distribution will be binomial? @BigBowlOfXoop $\endgroup$ Jul 25 '19 at 10:20
  • $\begingroup$ Great! Well first of all, there is no right or wrong distribution to use. It's all about the question you want to answer. The Binomial distribution is used to model the "number of successful outcomes from a known number of trials", hence in this case we are modelling ."the number of pregnancies out of a known number of pregnant women in a given hospital". We also could have used the Beta distribution to directly predict the proportion of pregnancies, and then used statistical inference to estimate the shape parameters. The Beta distribution is very common in inference for this reason. $\endgroup$
    – RaphaelS
    Jul 25 '19 at 12:09
  • $\begingroup$ Poisson or negative binomial models might be more suitable If you don''t have a known N. $\endgroup$
    – Glen_b
    Jul 26 '19 at 4:00
  • $\begingroup$ Hm but are the Poisson assumptions not violated? e.g. Childbirths occurring single is violated depending on how common twins are. Doesn't neg bin have same problem if unknown N? $\endgroup$
    – RaphaelS
    Jul 26 '19 at 9:52

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.