0
$\begingroup$

According to this website: https://datayze.com/labor-probability-calculator

One can model the probability distribution function of spontaneous labour given two pieces of information.

a) the due date of the pregnant women b) the current date

I have a set of information that provides the due date of labour, and actual date of labour for hundred's of women. It looks like this:

Est. Due Date Actual Labour Date
11th Sep 9th Sep
12th Oct 18th Oct
... ...
15th Aug 26th July

I would like to be able to generate a model similar to the above linked website that allows me to make predictions similar to the website. I.e. I can make a statement like

Your due date is 14th November 2022, given you haven't already given birth there is a x% chance you will give birth today.

or more generally a function that takes a due date as input and outputs the probability of the labour happening over the next number of days (until that probability hits 0 or close enough)

I.e. if the due date is 14th Nov and labour hasn't occurred yet it might produce:

Date Probability of labour
11th Nov 3.1%
12th Nov 3.6%
... ...
15th Dec 0.00003%

The website suggests a skewed normal distribution, but I am not sure how to generate these given the data I have.

Does anyone know how to model such a distribution given what I have and how to use that to make predictions?

$\endgroup$

1 Answer 1

1
$\begingroup$
  1. I disagree with the website author that, based on the information given, these data necessarily follow a skewed normal distribution. I agree the data most likely exhibit skewness (non-equal median and mean - for some reason the author discusses the mode on this point, perhaps a Freudian slip?), and a skewed normal may provide a reasonable approximation. What would be most useful is a continuous histogram of the data based on a kernal density estimator, and based day-by-day labor probabilities on that.
  2. Whereas you often see univariate statistics such as the mean and standard deviation being reported on key variables in a dataset, this does not imply that the data are normal, nor are these measures invalidated when the data do not follow normal rules. As a method of moment estimator, the mean and SD are a robust measure of location and scale.
  3. Estimation of a skewed normal distribution is discussed widely. The wikipedia article, even, gives a nice overview. Typically, you identify the power law that optimally transforms the data by a Box-Cox transformation, and after applying the inverse transform, estimating the mean and standard deviation to identify the 3 parameter normal curve.
$\endgroup$
1
  • $\begingroup$ Okay thank you for the reply. In light of what you have said do you think there is a better way to achieve what I am asking then? Do you have any suggestions? $\endgroup$
    – Tom
    Nov 11, 2022 at 17:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.